作者：Yue Qi (Mobile Programmer of Arena Breakout, Morefun Studios, Tencent Games)
With the advancement of hardware technology, an increasing number of mobile devices support high screen refresh rates such as 120Hz, 144Hz, and even higher. This creates the challenge of increasing frame rates without sacrificing image quality in mobile rendering.
This session introduces a frame-prediction-based rendering method (non-neural network) in Arena Breakout, specifically designed for achieving high frame rates, successfully implemented in the released game.. Frame prediction effectively reuses previously drawn pixels, significantly boosting the game's frame rate and reducing power consumption without adding latency to framebuffer presentation.
The presentation includes the implementation of frame prediction, the corresponding rendering pipelines, accuracy and performance analysis, and a comparison with other solutions.
随着硬件技术的进步，越来越多移动设备支持高屏幕刷新率，如120Hz、144Hz，甚至更高。这就产生了一个挑战，即在移动渲染中增加帧率而不牺牲图像质量。
本次专题介绍了一种基于帧预测的渲染方法（非神经网络）在《Arena Breakout》游戏中的应用，专门设计用于实现高帧率，在已发布的游戏中取得成功。帧预测有效地重复利用先前绘制的像素，显著提升游戏的帧率，并减少功耗，而不会给帧缓冲呈现增加延迟。
演示内容包括帧预测的实现，相应的渲染管线，准确性和性能分析，以及与其他解决方案的比较。
链接：https://schedule.gdconf.com/session/3d-toon-rendering-in-hi-fi-rush/899802
作者：Kosuke Tanaka (Lead Graphics Programmer, Tango Gameworks)、Takashi Komada (Graphics/Physics Programmer, Tango Gameworks)
'Hi-Fi RUSH' is a 60fps rhythm action game rendered in a stylish 3D toon art style using a customized Unreal Engine 4. In 'Hi-Fi RUSH', the developers at Tango Gameworks used toon shading for the entire world, both character and environment, while providing modern 3D graphics features, all at 60fps at native resolution.Learn about the development of a deferred toon renderer developed for toon shading a 3D world in a 2D way. Various customized UE4 render passes along with added render passes of their own are stylized and combined inside deferred rendered post process volumes.Comic shaders, toon lights, dynamic shadow map strategies, static shadow maps, light probe GI, and their face shadow implementation will be discussed.Learn about the core tech behind the game as well as select graphics implementations.
《Hi-Fi RUSH》是一款以时尚的3D卡通风格渲染的60帧节奏动作游戏，使用定制的虚幻引擎4。在《Hi-Fi RUSH》中，Tango Gameworks的开发人员使用卡通渲染技术对整个世界进行渲染，包括角色和环境，在提供现代3D图形功能的同时，以本机分辨率以60fps运行。了解为卡通渲染一个3D世界而开发的延迟卡通渲染器的开发过程，以二维方式进行卡通着色。各种定制的UE4渲染通道以及他们自己添加的渲染通道在延迟渲染的后期处理体积内进行风格化和组合。将讨论漫画着色器、卡通灯光、动态阴影图策略、静态阴影图、光探针全局光照（GI）以及它们的面部阴影实现。了解游戏背后的核心技术以及部分图形实现细节。
作者：Sig Narvaez (Executive Solutions Architect, MongoDB)
Games launch at the lowest cost, but need to be ready for that massive scale, or end up on the front page of reddit. And scale isn't your only worry, enriching your gaming data with AI and vector embeddings to improve outcomes such engagement and retention, shouldn't break the bank or create unnecessary complexity! This session will share some of the patterns and practices followed by some of the biggest online games, covering topics such as scaling from zero to a billion players and beyond, adding AI embeddings to your data, all while simplifying the live services backend stack. Learn how to make informed decisions during game development, and be ready to run hot without melting the database!
游戏在最低成本下启动，但需要为那种大规模准备好，否则最终会出现在Reddit的头条上。而规模并非你唯一需要担心的，利用人工智能和向量嵌入来丰富游戏数据，以提高用户参与度和留存率，不应该成为银行的负担或者带来不必要的复杂性！本课程将分享一些最大型在线游戏所遵循的模式和实践，涵盖从零到十亿玩家甚至更多的扩张过程，将AI嵌入到数据中，同时简化现场服务后端架构。了解如何在游戏开发过程中做出明智决策，并准备好高强度运行而不会使数据库崩溃！
链接：https://schedule.gdconf.com/session/ai-roundtable-day-1-deep-dive-with-experts/899046
作者：Neil Kirby (Senior Lecturer, The Ohio State University)
The AI Roundtables session provides discussion-based sessions on Wednesday and Friday and a Q&A format session on Thursday. The Monday session starts with a discussion of machine learning versus regular game AI and then goes on to topics from the audience. The Thursday session is Q&A bring questions and help answer questions and expect a larger number of topics to go by. The Friday session is completely driven by topics from the audience. The moderator helps to make sure that everyone gets a chance to speak and to keep the topics moving.
AI圆桌会议包括周三和周五的基于讨论的会话，周四则采用问答形式。周一的会议从讨论机器学习与普通游戏人工智能开始，然后转向观众提出的话题。周四的会议是问答环节，旨在提出问题并解答问题，预计将涉及更多话题。周五的会议完全由观众提出话题驱动。主持人帮助确保每个人都有机会发言，并推动话题的进行。
链接：https://schedule.gdconf.com/session/ai-roundtable-day-2-quick-qa-for-beginners/903094
作者：Neil Kirby (Senior Lecturer, The Ohio State University)
The AI Roundtables provides discussion-based sessions on Wednesday and Friday and a Q&A format session on Thursday. The Monday session starts with a discussion of machine learning versus regular game AI and then goes on to topics from the audience. The Thursday session is Q&A bring questions and help answer questions and expect a larger number of topics to go by. The Friday session is completely driven by topics from the audience. The moderator helps to make sure that everyone gets a chance to speak and to keep the topics moving.
AI圆桌会议在周三和周五提供基于讨论的会议，并在周四以问答形式举行会议。周一的会议从讨论机器学习与常规游戏人工智能开始，然后转向来自听众的主题。周四的会议是问答环节，听众提出问题并帮助回答问题，预计将涉及较多主题。周五的会议完全由听众提供主题驱动。主持人帮助确保每个人都有机会发言，并保持讨论话题的连贯性。
链接：https://schedule.gdconf.com/session/ai-roundtable-day-3-open-discussion/903095
作者：Neil Kirby (Senior Lecturer, The Ohio State University)
The AI Roundtables provides discussion-based sessions on Wednesday and Friday and a Q&A format session on Thursday. The Monday session starts with a discussion of machine learning versus regular game AI and then goes on to topics from the audience. The Thursday session is Q&A bring questions and help answer questions and expect a larger number of topics to go by. The Friday session is completely driven by topics from the audience. The moderator helps to make sure that everyone gets a chance to speak and to keep the topics moving.
AI圆桌会议在周三和周五提供基于讨论的会议，并在周四以问答形式举行会议。周一的会议从讨论机器学习与常规游戏人工智能开始，然后转向来自听众的主题。周四的会议是问答环节，听众提出问题并帮助回答问题，预计将涉及较多主题。周五的会议完全由听众提供主题驱动。主持人帮助确保每个人都有机会发言，并保持讨论话题的连贯性。
作者：Dante Camarena (Technical Director, Transitional Forms Inc)、Nick Counter (Art Director, Transitional Forms Inc)
Recent developments in Reinforcement Learning (RL) have resulted in near-human ability in competitive games such as Go, StarCraft and Dota 2, at the cost of significant computing power. An often under-looked element of training RL is the satisfaction that researchers and enthusiasts gain while watching their agents evolve and learn.
In this talk, Co-Directors Nick Counter and Dante Camarena present the work done in constructing Little Learning Machines, a game designed to bring the experience of training your own AI to players of any skill level. They show how careful game design and tricky engineering can help create a game that makes training accessible to people without technical knowledge or expensive computers. This presentation is intended to inspire developers to push this idea further.
最近深度强化学习（RL）领域的进展导致了在竞技游戏如围棋、星际争霸和《刀塔》中实现了接近人类水平的能力，但需要付出大量的计算力。强化学习训练中经常被忽视的一个元素是研究人员和爱好者在观看他们的智能体进化和学习过程中获得的满足感。
在这次讲话中，联合导演尼克·康特和丹特·卡马瑞纳介绍了构建《小学习机器》游戏的工作，该游戏旨在为各种技能水平的玩家带来训练自己人工智能的体验。他们展示了如何通过精心设计游戏和巧妙地工程方法，来创建一个让人们可以在没有技术知识或昂贵计算机的情况下进行训练的游戏。本次演讲旨在激励开发人员进一步推动这一想法。
作者：Gabriel Robert (Lead R&D Developer, Ubisoft)
In this presentation, Ubisoft La Forge's explores the approach to developing ML bots, designed to simulate human players in our games. These bots are not only integral to enhancing player experiences, particularly for onboarding in multiplayer games, but also play a crucial role in Quality Control (QC) testing across our expansive open-world titles. They delve into specific examples to demonstrate how these bots, underpinned by cutting-edge machine learning (ML) technologies, are collaboratively integrated into game development. This synergy between production teams and innovative ML applications is pivotal in upping game quality and pushing the boundaries of technological advancement in gaming.
在这个演示中，育碧拉福儿研究部探讨了开发机器学习（ML）机器人的方法，旨在模拟我们游戏中的人类玩家。这些机器人不仅对增强玩家体验至关重要，特别是在多人游戏中的新手引导方面，而且在我们众多庞大开放世界游戏的质量控制（QC）测试中发挥关键作用。他们深入具体示例，展示了这些机器人是如何借助尖端机器学习（ML）技术，并协同集成到游戏开发中的。制作团队与创新的 ML 应用之间的协同作用对于提升游戏质量，并推动游戏技术进步的边界至关重要。
作者：Hilary Mason (CEO & Founder, Hidden Door)
Generative AI, when combined with traditional game development techniques, can allow for a new generation of games, ones that possess truly infinite worlds that dynamically respond to player inputs. Drawing upon decades of work as a data scientist and her current role leading Hidden Door, Hilary walks attendees through the process of integrating AI into traditional game development. Within the talk, Hilary will address AI's shortcomings and ethical concerns, as well as outline an action plan for attendees to use as they explore incorporating A into their game development workflows.
生成式人工智能（Generative AI），当与传统游戏开发技术相结合时，可以实现一种新一代的游戏，具备真正无限的世界，动态响应玩家输入。作为一名数据科学家并担任 Hidden Door 领导的 Hilary，依托数十年的工作经验，将向与会者介绍将人工智能整合到传统游戏开发过程中的步骤。在演讲中，Hilary 将讨论人工智能的不足和伦理问题，同时概述一个行动计划，供与会者在探索将人工智能纳入游戏开发工作流程时使用。
作者：Andrea Tucci (Senior Gameplay Programmer, Rocksteady Studios)、Giovanni Campo (Lead Gameplay Systems Programmer, Rocksteady Studios)
In this session, Gio and Andrea provide a tour of the AI behavior systems used to bring Brainiac's army of corrupted to life in 'Suicide Squad: Kill the Justice League'. This talk covers all levels of the behavior systems, from game activity level through to groups of enemies and finally to individual agent decision-making and actuation of abilities. The speakers discuss the design and implementation of a spatial network and how this innovative approach enhances AI tactics in combat analysis and pathing. They dive into real content examples from the game and discuss the challenges of providing a fun player experience while scaling enemy behaviour up to large groups, and how technology created by the team at Rocksteady overcame these difficult problems.
在这个讲座中，Gio和Andrea向我们展示了用于赋予Brainiac的腐化军队生命的AI行为系统在《自杀小队：杀戮正义联盟》中的运用。本次讨论涵盖了所有级别的行为系统，从游戏活动级别到敌人群体，最终到个体代理的决策和能力激活。演讲者们讨论了空间网络的设计和实施，以及这种创新方法如何增强了战斗分析和路径规划中的AI战术。他们深入探讨了游戏中的真实内容示例，并讨论了如何在将敌人行为扩展到大群体的同时提供有趣的玩家体验的挑战，以及Rocksteady团队开发的技术是如何克服这些困难问题的。
作者：Theresa Duringer (CEO, Temple Gates Games)
Multiplayer boardgame adaptations require competitive AI opponents for solo play. But what's the best approach to creating these opponents? Temple Gates has scripted a heuristic AI for 'Ascension' and integrated a neural network for 'Race for the Galaxy'. Since then, each new game has used new techniques to improve AI competitiveness. Deckbuilder 'Shards of Infinity's' AI identifies symmetries, such as symmetries between players or cards, to supercharge the speed of training. 'Roll for the Galaxy's' AI is designed with holes in the inputs to speed training without creating strategic blindspots. 'Dominion' introduces an embedding layer, so the neural network can learn not only the cards themselves, but the value of their constituent features. This helps handle the highly variant game state where one of a quadrillion of permutations of cards may be in play. Understanding what techniques are available in ml and which best apply to different game scenarios is a shortcut to getting the most of a nn AI.
多人棋盘游戏改编需要在单人游戏中提供有竞争力的人工智能对手。但是，创建这些对手的最佳方法是什么？Temple Gates为《Ascension》设计了一种基于启发式算法的AI，并为《Race for the Galaxy》整合了神经网络。自那时起，每款新游戏都使用了新技术来提高AI的竞争力。卡牌构建游戏《Shards of Infinity》的AI识别对称性，如玩家或卡牌之间的对称性，以加速训练速度。《Roll for the Galaxy》的AI设计了输入中的空隙，以加快训练速度，同时避免出现战略盲点。《Dominion》引入了一个嵌入层，因此神经网络不仅可以学习卡牌本身，还可以学习它们构成特征的价值。这有助于处理游戏状态的高度变异性，其中可能有千亿级卡牌排列的其中一种在使用中。了解在机器学习中有哪些技术，并且哪些技术最适用于不同游戏场景，是利用神经网络AI的最佳捷径。
作者：Trevor Santarra (Senior Research Engineer, Unity Technologies)、Pierre-Arjun Dalaya (Senior Software Engineer, Unity Technologies)
Behavior trees have been a cornerstone of game AI for years, but what happens when you infuse them with the power of generative AI? Join Pierre Dalaya and Trevor Santarra as they explore the remarkable synergy between large language models (LLMs) and behavior tree authoring. Discover how these cutting-edge techniques can revolutionize your game development process by generating behavior trees and code, seamlessly integrating modern AI convenience into a familiar tool set. This talk showcases how generative AI techniques can supercharge your gameplay authoring tools, providing you with a whole new level of efficiency and creativity. Don't miss this opportunity to explore the future of game AI!
行为树（Behavior trees）已经是游戏人工智能的基石多年了，但是当你将它们与生成式人工智能的力量结合起来会发生什么呢？加入Pierre Dalaya和Trevor Santarra，他们探索了大型语言模型（LLMs）和行为树编写之间卓越的协同作用。发现这些前沿技术如何通过生成行为树和代码来彻底改变您的游戏开发流程，将现代人工智能便利无缝集成到熟悉的工具集中。这个讲座展示了生成式人工智能技术如何可以让您的游戏创作工具发挥更大作用，为您提供全新水平的效率和创造力。不要错过这个探索游戏人工智能未来的机会！
作者：David Partouche (Principal Software Engineer, The Multiplayer Group)
Existing solutions for navigation in standard game engines are mostly aimed at biped characters, and trying to apply them to vehicles is "doomed for failure" as it fails to take into consideration the kinematics constraints of the vehicle (i.e. most vehicle have a limited turning rate). Most games use navigation paths for vehicles, which are precomputed and already optimised for the vehicle's turning rate. But it means that the vehicle needs to stay on these paths. We provide a solution, adapted from Stanford's Hybrid A* algorithm to allow vehicles in game to navigate anywhere on an open map.
现有的标准游戏引擎中用于导航的解决方案大多针对双足角色，并试图将其应用于车辆是注定失败的，因为它未考虑到车辆的运动学约束（例如，大多数车辆具有有限的转向速率）。大多数游戏使用预先计算并针对车辆转向速率进行了优化的导航路径。但这意味着车辆需要留在这些路径上。我们提供了一种解决方案，改编自斯坦福大学的混合A*算法，可以让游戏中的车辆在开放地图上任意导航。
作者：Will Chambers (AI Lead, Oxide Games)
Traditional approaches to 4X game AI have relied on complex, hand-written code that struggles to deliver consistent, immersive gameplay experiences. In this talk, Oxide Games Software Engineer Will Chambers unveils 'Ara: History Untold's' groundbreaking AI approach that transforms the way we think about CPU player behavior.'Ara's' system leverages an AI framework to strategize and make decisions efficiently, planning many turns into the future. This system translates goals into a series of statements, breaking them down into subgoals.What sets 'Ara' apart is its ability to adapt and control AI behavior without the need for extensive programming. Whether managing nations or predicting player actions, Ara provides a tailored gaming experience without resorting to cheating tactics.In this session, Chambers explores the theory, methods, and implementation for goal-oriented AI while ensuring a seamless gaming experience. Learn how 'Ara's' approach redefines 4X game AI and player engagement!
传统的4X游戏人工智能方法倚赖复杂的手工编写代码，难以提供一致、沉浸式的游戏体验。在这次讲座中，Oxide Games软件工程师Will Chambers揭示了《Ara: History Untold》的开创性人工智能方法，改变了我们对CPU玩家行为的思考。《Ara》的系统利用人工智能框架高效地制定策略和做出决策，规划未来多个回合。这一系统将目标转化为一系列语句，将其分解为子目标。《Ara》的独特之处在于其能够在不需要大量编程的情况下自适应和控制人工智能行为。无论是管理国家还是预测玩家行为，Ara都能提供一种定制化的游戏体验，而无需诉诸作弊手段。在这个讲座中，Chambers探讨了面向目标的人工智能的理论、方法和实施，确保了游戏体验的流畅性。了解《Ara》的方法如何重新定义4X游戏人工智能和玩家参与度！
链接：https://schedule.gdconf.com/session/ai-summit-level-generation-with-large-language-models/902483
作者：Julian Togelius (Research Director, modl.ai)
Can LLMs generate video game levels, among the many other things they can do? This session provides an overview of how to generate game levels with large language models, including but not limited to GPT-3, GPT-4, and Llama. Various strategies based on prompting, finetuning, and bootstrapping are discussed. The talk is accessible and some of the ideas are immediately actionable.
LLM（Large Language Models）可以生成视频游戏关卡吗，除此之外它们还可以做很多其他事情吗？本次会话将概述如何使用大型语言模型生成游戏关卡，包括但不限于GPT-3、GPT-4和Llama。我们将讨论基于提示、微调和引导式学习的各种策略。这个讲座容易理解，其中一些想法也可以立即付诸实践。
作者：Marek Zeman (Principal AI Programmer for CRYENGINE, Crytek)
The Navigation System and NavMeshes in Cryengine have gone through significant changes in the past decade. In this session, we present the new major features added to the NavMeshes since their introduction during the development of 'Crysis 3'. We also share details about each added feature in chronological order, what problem each feature solves, what challenges we needed to overcome, how it was implemented and how it is used specifically in Crytek's latest game 'Hunt: Showdown'. We also show, whenever possible, what the performance implications of each feature are.
在过去的十年里，Cryengine中的导航系统和NavMeshes已经发生了重大变化。在这个讲座中，我们将介绍自从它们在开发《Crysis 3》期间引入以来，NavMeshes新增的主要功能。我们还将按照时间顺序分享每个新增功能的详细信息，包括每个功能解决的问题，我们需要克服的挑战，它是如何实现的，并且在Crytek最新游戏《猎狩：生存之战（Hunt: Showdown）》中具体如何使用。我们还会尽可能展示每个功能的性能影响。
作者：Sam Fay-Hunt (Game AI Developer, Kythera AI)
When authoring behavior for game AI we must reason about a broad range of systems and their associated data. Commonly, this reasoning is formulated by means of visual scripting languages and behavior trees (among others). These tools often introduce challenges for reusability, and managing data dependencies in complex worlds.In this talk, Sam Fay-Hunt demonstrates a framework to modularize game AI concepts into compartmentalized, reusable components. This system is highly suited to reasoning about, and describing properties of, environments or situations. He describes the core pillars of its design, and how it can be used to augment existing behavior authoring systems to empower designers to incorporate powerful concepts such as quality, danger, or utility into decision making.
在为游戏AI编写行为时，我们必须考虑一系列系统及其关联数据。通常，这种推理是通过可视化脚本语言和行为树（以及其他一些方式）来进行的。这些工具经常会引入可重用性方面的挑战，并在复杂的世界中管理数据依赖关系。在这次讲座中，Sam Fay-Hunt 展示了将游戏AI概念模块化为独立、可重用组件的框架。该系统非常适合用于推理环境或情境的属性描述。他描述了其设计的核心支柱，并介绍了如何将其用于增强现有的行为编写系统，从而让设计人员能够将强大的概念（如质量、危险或效用）融入决策过程中。
链接：https://schedule.gdconf.com/session/ai-summit-practical-utility-ai/902394
作者：Rez Graham (Creative / Technical Director, Independent)
Utility theory is a common method for handling decision making in games. Much has already been said about it, but most resources focus on theory rather than the nuts & bolts of actually building the system. This session solves that by focusing on implementation with only a light touch on theory.
The session starts by dissecting the architecture of a utility system, breaking down core concepts such as considerations, utility curves, and decision hierarchies. The session also examines the importance of data in tuning utility systems, offering insights into how designers manipulate data, curves, and more. Attendees discover essential tools for debugging utility systems and learn to navigate potential challenges and performance concerns.
The session concludes with a live demo, showing the full process of adding a new decision into the game, from design to implementation.
效用理论（Utility theory）是处理游戏决策的常见方法。关于这一理论已经有很多讨论，但大多数资源侧重于理论而不是实际构建系统的细节。本次会议通过着重于实施，只轻触理论，来解决这一问题。
本次会议首先解析效用系统的架构，分解核心概念，如考虑因素、效用曲线和决策层次结构。会议还探讨了数据在调整效用系统中的重要性，提供了洞见，展示设计师如何操纵数据、曲线等。与会者将发现调试效用系统的基本工具，学会应对潜在的挑战和性能问题。
会议以现场演示结束，展示了将新决策添加进游戏的全过程，从设计到实施。
作者：Jane Friedhoff (Senior UX Engineer, Google)、Suma Bailis (Staff Research Engineer, Google)、Feiyang Chen (Senior Software Engineer, Google)
Large language models (LLMs) present an exciting vision for the future of NPCs. Beyond just generating dialog, these language models can be equipped with agentic frameworks that allow them to reason about the present, reflect on the past, and plan for the future. This dynamic behavior has a perfect home in social deduction games, where long-term strategy, logical reasoning, and persuasion are key. In this talk, we present a game of 'Werewolf' in which your fellow villagers are all LLM-powered agents. We walk through our design process, sharing the architecture and key loops that LLM-curious developers can use for their own agentic games. We also share our learnings (from human-v-bots and bot-only games) about what tweaks facilitated the most clever, persuasive, and poker-faced agent behaviors.
大型语言模型（Large language models，LLMs）展示了未来NPC（Non-Player Character）的激动人心愿景。除了仅仅生成对话外，这些语言模型可以配备代理框架，使它们能够推理现在，回顾过去，并规划未来。这种动态行为在社交推理游戏中得以完美体现，那里长期策略、逻辑推理和说服力至关重要。在这个讲话中，我们提出了一个名为‘Werewolf’的游戏，在这个游戏中，你的村民都是由LLM提供动力的代理人。我们将详细介绍我们的设计流程，分享LLM-探究开发者可以用于自己的代理游戏的架构和关键循环。我们还会分享我们从人对战机器人和纯机器人对战中学到的知识，包括哪些调整促进了最聪明、最具说服力和面无表情的代理行为。
作者：Andrew Hoyt (Gameplay Engineer, Treyarch)
In this talk, Andrew Hoyt discusses the unique technical and design challenges involved in pioneering giant monster AI in 'Call of Duty'. The talk is presented in a postmortem format and focuses on the specific challenges that were faced when creating boss AI for the franchise in recent years (2021-2023). Due to the nature of 'Call of Duty's' annual release schedule, the talk features how they tackled complex issues with cost-effective solutions. Specific topics include their approach to pathfinding/movement, making attacks feel varied with limited animations, and what engine modifications were required when the assumption was that an AI was a human soldier.
在这个讲座中，安德鲁·霍伊特讨论了在《使命召唤》系列中开拓巨型怪物AI所涉及的独特技术和设计挑战。该讲座以事后总结的形式呈现，重点讨论了近年来为该系列创作Boss AI时所面临的具体挑战（2021-2023）。由于《使命召唤》系列每年都会推出新作，这个讲座介绍了他们如何以经济有效的解决方案应对复杂问题。具体主题包括他们对寻路/移动的方法、如何利用有限的动画制作多样的攻击感觉，以及当假设AI是人类士兵时需要进行的引擎修改。
链接：https://schedule.gdconf.com/session/ai-summit-the-evolution-of-ai-in-warframe/902416
作者：Daniel Brewer (Lead AI Programmer, Digital Extremes)
In 2012, we bet everything on a passion project to save the studio: a weird sci-fi MMO about space ninjas. Since then, 'Warframe' has developed a thriving community and grown from strength to strength. This talk covers the game AI tech we have built over the last 11 years for 'Warframe'. We also offer some insight into our methods and processes we use to continue to develop the technology while supporting, expanding and improving, a live service game.
2012年，我们把一切都押在了一个激情项目上，以挽救工作室：一个关于太空忍者的奇幻科幻MMO游戏。自那时起，《战争框架》发展出一个蓬勃发展的社区，并不断壮大。本次演讲涵盖了我们为《战争框架》这款游戏在过去11年中构建的游戏人工智能技术。我们还提供了一些关于我们的方法和流程的见解，这些方法和流程是为了继续发展技术，同时支持、扩展和改进一款在线服务游戏。
链接：https://schedule.gdconf.com/session/ai-summit-the-simplest-ai-trick-in-the-book/902333
作者：Jackson Barr (AI Programmer, Digital Extremes)、Rez Graham (Creative / Technical Director, Independent)、Daniel Brewer (Lead AI Programmer, Digital Extremes)、Jeff Orkin (CEO, Bitpart AI)
Most of the time when talking about game AI, you think of large, complex systems with many interconnected parts and delicate algorithms. But sometimes, you return to your true "smoke and mirrors" roots and get away with implementing something that is truly simple and yet has profound effects on the end product. This can be something that is apparent to the player, something that is only visible to the other devs around you, or might be something that only lives under the hood, known only to you. The common thread is that it is something that makes you ask, "why doesn't everyone do this?" This will be a rapid-fire series of presenters sharing something they have discovered and used over the years. They have to be simple, because each person will only have a few minutes to make their case!
在谈论游戏人工智能时，大多数时候你会想到拥有许多相互连接部分和精密算法的庞大、复杂系统。但有时，你会回归到真正的‘烟雾镜’本质，想方设法实现一些极其简单却在最终产品上产生深远影响的东西。这可能是玩家可见的特征，也可能只有你周围的其他开发人员能看见，或者可能是只存在于幕后、只有你能知道的东西。共同之处在于它会让你产生疑问：“为什么不每个人都这样做呢？”这将是一系列快速连续的演讲者分享多年来发现并应用的内容。它们必须简单，因为每个人只有几分钟时间陈述自己的观点！
作者：Ken Mitchell (Fellow, AMD)
Breakthrough CPU bottlenecks to reach higher frames-per-second and more builds-per-day equipped with AMD Game Engineering's expertise forged through its rich history of partnerships with AAA game developers! Learn about exciting AMD Ryzen™ products featuring advanced processor and graphics technologies powering today's laptop, desktop, and workstation PCs. Dive into data flow, simultaneous multi-threading, resource sharing, instruction set evolution, cache hierarchies, and coherency. Unlock powerful profiling tools and application analysis techniques using the Windows Performance Analyzer, Concurrency Visualizer, and AMD uProf. Discover best practices and lessons learned. Upgrade system software and build settings. Attack valuable code optimization opportunities. Examples include C/C++, assembly, and hardware performance-monitoring counters.
突破 CPU 瓶颈，实现更高的帧率和更多的每日构建，搭载着 AMD 游戏工程团队通过与 AAA 游戏开发者合作积累的丰富经验所打造的专业知识！了解激动人心的 AMD Ryzen™ 产品，拥有先进的处理器和图形技术，为当今的笔记本电脑、台式机和工作站 PC 提供动力。深入了解数据流、同时多线程、资源共享、指令集演进、缓存层次结构和一致性。解锁强大的性能分析工具和应用程序分析技术，包括 Windows 性能分析器、并发可视化器和 AMD uteProf。发现最佳实践和经验教训。升级系统软件和构建设置。探索有价值的代码优化机会。示例包括 C/C++、汇编和硬件性能监控计数器。
链接：https://schedule.gdconf.com/session/advanced-graphics-summit-all-you-can-ask/902717
作者：Julien Merceron (Worldwide CTO, Bandai Namco Studios)、Gauthier Viau (Graphics Programmer, Massive Entertainment - a Ubisoft Studio)、Dan Baker (Founder and Lead Graphics Architect, Oxide Games)、Matthaeus Chajdas (Fellow, AMD)、Quentin Kuenlin (Senior Rendering Programmer, Massive Entertainment – a Ubisoft Studio)、Evan Hart (Distinguished Engineer, NVIDIA)、Adam Marrs (Principal Engineer, NVIDIA)
The GDC Advanced Graphics Summit has designed this panel session to gather eminent panelists that will be on stage to answer questions from the audience, any question related to Advanced Graphics in the Video Game Industry. Unlike traditional panels where the moderator asks questions to the panelists, the 60 minutes are dedicated to the audience. We expect questions on topics such as Career Management, Working style and Processes, Cross-media and Convergence, Advanced Rendering Techniques, Ray Tracing, Participating Media, Streaming / Compression / Decompression, CPU side frame scheduling, GPU side frame scheduling, Post Effects, Machine Learning in Rendering, Hardware related aspects, and more ! This panel is all yours! All You Can Ask !
《GDC高级图形峰会》已经设计了这个小组讨论环节，旨在汇聚杰出的专家，他们将登台回答观众提出的任何与《视频游戏产业中的高级图形》相关的问题。与传统的小组讨论不同，主持人不会向小组成员提问，整整60分钟将完全留给观众。我们期待听到关于职业规划、工作风格和流程、跨媒体和融合、高级渲染技术、光线追踪、参与式媒体、流媒体/压缩/解压缩、CPU端帧调度、GPU端帧调度、后处理效果、渲染中的机器学习、硬件相关的方面等主题的提问！这个小组讨论完全属于你们！你问我答！
作者：Matthaeus Chajdas (Fellow, AMD)、Shawn Hargreaves (Dev Manager, Direct 3D, Microsoft)
Work graphs are a brand-new API, allowing the GPU to self-schedule work without going through the host. Unlike previous APIs like ExecuteIndirect, work graphs allow for fully parallel producer/consumer relationships on the GPU. It allows for high-performance GPU work creation, efficient recursive algorithms, expansion and compaction, and many other things.Join AMD’s Matthäus Chajdas and Microsoft’s Shawn Hargreaves for the first look at this brand new API and learn how it's going to transform the GPU programming landscape going forward. Together they will provide insights into what workloads can benefit from work graphs short-term and where we expect long-term benefits. The audience will get an in-depth overview of the API, the changes to HLSL, and walk through a sample to get everyone equipped for experiments and integration into their own games and applications.
工作图（Work graphs）是一种全新的API，允许GPU自行安排工作而无需通过主机。与之前的ExecuteIndirect等API不同，工作图允许在GPU上实现完全并行的生产者/消费者关系。它可以实现高性能的GPU工作创建，高效的递归算法，扩展和压缩，以及许多其他功能。加入AMD的Matthäus Chajdas和微软的Shawn Hargreaves，一起来第一时间了解这个全新API，并了解它将如何改变GPU编程领域。他们将深入探讨哪些工作负载可以从工作图中短期受益，以及我们期待长期受益的领域。观众将获得API的深入概述，HLSL的更改，以及通过一个示例讲解，使每个人都能为实验和在自己的游戏和应用中集成做好准备。
作者：Dan Baker (Founder and Lead Graphics Architect, Oxide Games)
Rendering technology in the gaming industry has evolved significantly, yet a promising approach called "Decoupled Shading" remains underexplored for widespread use. Dan Baker, founder of Oxide Games, shares his knowledge of this innovative rendering technique and how implementation can enhance rendering quality and efficiency. The lecture covers Decoupled Shading's principles, benefits, and practical use, drawing on insights from projects like 'Ara: History Untold'. Topics include challenges, anti-aliasing, virtualized shadel allocation, Oxide's shading language (ZNO), performance analysis, and future directions. Attendees will gain a comprehensive understanding of this technique's potential to revolutionize game rendering for both quality and performance.
在游戏行业中，渲染技术已经有了显著的发展，但一种被称为“解耦合着色（Decoupled Shading）”的有前途的方法仍未被广泛应用。Oxide Games 创始人 Dan Baker 分享了他对这种创新渲染技术的理解，以及如何通过实现来提升渲染质量和效率。这次讲座涵盖了解耦合着色的原理、优势和实际用途，借鉴了《Ara: History Untold》等项目的见解。讨论的主题包括挑战、抗锯齿、虚拟化着色分配、Oxide 的着色语言（ZNO）、性能分析和未来方向。与会者将全面了解这种技术在提升游戏渲染质量和性能方面的潜力，有可能彻底改变游戏渲染的方式。
作者：Gauthier Viau (Graphics Programmer, Massive Entertainment - a Ubisoft Studio)
To achieve good and reliable performance, games using DX12 on PC need to manage memory themselves. This presentation goes over Snowdrop's heaps allocation strategies, as well as how its texture streamer is used to manage memory.Attendees will learn how to use the Windows Performance Analyzer to diagnose VRAM overcommitment issues coming from OS automatic resource demotion, using real examples of performance degradations due to bad memory management in Snowdrop with DX12.Solutions to address these issues – a resource residency management system and the texture streamer usage of DirectX tiled resources to free large amount of memory while minimizing quality loss – are discussed, as well as what didn't work so well.
为了实现良好且可靠的性能，在PC上使用DX12的游戏需要自行管理内存。本次演示将介绍Snowdrop引擎的堆分配策略，以及其纹理流管理器如何用于内存管理。参与者将学习如何使用Windows性能分析器诊断由操作系统自动资源降级导致的VRAM超量使用问题，通过实际性能下降案例展示由于Snowdrop引擎在DX12中的不良内存管理而导致的问题。
讨论了解决这些问题的方法 - 资源驻留管理系统和DirectX平铺资源的纹理流管理器用于释放大量内存同时最小化质量损失，以及哪些方法效果不佳。
作者：Evan Hart (Distinguished Engineer, NVIDIA)、Adam Marrs (Principal Engineer, NVIDIA)
Path traced effects in games have been a dream of many since the early days of real-time rendering. The advent of hardware accelerated ray tracing and recent advances in algorithms derived from Reservoir-based Spatio-Temporal Importance Sampling (ReSTIR) have turned those dreams into reality. In this session, Evan and Adam discuss the theoretical and practical elements necessary to add real-time path–traced lighting to an advanced, modern game engine (Unreal Engine 5). Specifically, the team discusses changes to the light transport model and scalability that enhance existing engine systems to handle dramatically more complex lighting environments.
在游戏中的路径追踪效果自实时渲染早期便是许多人的梦想。随着硬件加速光线追踪的出现和基于‘ReSTIR（Reservoir-based Spatio-Temporal Importance Sampling）’的算法的最新进展，这些梦想变成了现实。在这个讨论中，Evan 和 Adam 讨论了将实时路径追踪光照添加到先进的现代游戏引擎（Unreal Engine 5）所需的理论和实践元素。具体来说，团队讨论了对光传输模型和可扩展性的改变，以增强现有引擎系统，使其能够处理更加复杂的光照环境。
作者：Quentin Kuenlin (Senior Rendering Programmer, Massive Entertainment – a Ubisoft Studio)
This presentation provides an overview of the pipeline used for the raytraced global illumination and raytraced reflections in the Snowdrop engine. It goes into a deep dive into the probe system used to light the forward rendered objects and also as a cache for the global illumination and reflections. The two probe placement strategies, a grid based one and a new octree based one, are described alongside their advantages and drawbacks. Next, it presents the raytracing done from the gbuffer for both the global illumination and the reflection. Various optimizations are presented, including the use of screen space tracing to reduce the number of world space ray needed, as well as improving raytraced lighting quality.The next section briefly covers the denoising strategy used in Snowdrop. Several denoiser options are provided, including AMD Fidelity FX Reflection Denoiser, Nvidia's raytracing Denoisers, as well as two custom in-house denoisers.Finally, the presentation ends on how the raytraced result are applied to the final image.
本演示提供了Snowdrop引擎中用于光线追踪全局光照和镜面反射的流程概述。它深入探讨了用于照亮前向渲染对象并作为全局光照和反射缓存的探针系统。描述了两种探针放置策略，一种基于网格的策略和一种新的基于八叉树的策略，以及它们的优势和缺点。接下来，介绍了从G缓冲区进行的光线追踪，既包括全局光照，也包括反射。提出了各种优化方法，包括使用屏幕空间追踪来减少需要的世界空间光线数量，同时改善光线追踪照明质量。接下来的部分简要介绍了Snowdrop中使用的去噪策略。提供了几种去噪选项，包括AMD Fidelity FX反射去噪器、Nvidia的光线追踪去噪器，以及两种自定义的内部去噪器。最后，演示介绍了光线追踪结果如何应用于最终图像。
链接：https://schedule.gdconf.com/session/advanced-uefn-features-presented-by-epic-games/903284
作者：Kurtis Schmidt (Framework Technical Director, Epic Games)、Tim Tillotson (Gameplay Engineer Director, Epic Games)
A look inside the new advanced UEFN features workflows and Verse APIs, which dramatically expand the power users have to construct content in UEFN.
深入了解新的高级 UEFN 功能工作流程和 Verse API，这显著拓展了用户在 UEFN 中构建内容的能力。
作者：Aurelio Reis (Director, Graphics Developer Tools, NVIDIA)、Ivan Fedorov (Developer Technology Engineer, NVIDIA)
As Ray Tracing continues to propel photorealistic rendering to new heights, the tools that drive this development have become more complex and powerful. In this session, we'll look at the improvements that have been made over the past year to tools like Nsight Graphics and Nsight Aftermath that help streamline your workflow and deliver breathtaking experiences to gamers worldwide. For the first time you'll see the brand new Ray Tracing Shader Debugger in action. There will also be a real world example of how the Shader Profiler has helped to profile and optimize complex ray tracing shaders by the NVIDIA DevTech pros.
随着光线追踪技术不断推动逼真渲染达到新的高度，推动这一发展的工具变得更加复杂和强大。在本次会议中，我们将看一看在过去一年中针对诸如 Nsight Graphics 和 Nsight Aftermath 这些工具所做的改进，这些工具有助于简化您的工作流程，并为全球玩家提供令人惊叹的体验。首次您将看到全新的光线追踪着色器调试器的实际效果。此外，还会有一个真实世界的例子展示着色器分析器如何帮助 NVIDIA DevTech 专家对复杂的光线追踪着色器进行分析和优化。
作者：Juha Sjöholm (Senior Developer Technology Engineer, NVIDIA)、Kiya Kandar (Graphics Programmer, Remedy Entertainment Plc, Remedy Entertainment Plc)
Alan Wake 2 was released last October with path tracing and DLSS Ray Reconstruction, receiving a lot of attention for the visual experience it offers. This session explains all aspects of the path tracer including acceleration structure management with large amounts of dynamic content, representation and accessing of material data, light sampling techniques and image upscaling and denoising methods. All key technologies are covered including using of the opacity micromaps (OMM) with alpha tested assets such as vegetation and hair, use of the shader execution reordering (SER) in the path tracing shader, how the DLSS Ray Reconstruction is used to produce the final image, and how smooth frame rates are achieved through the DLSS Frame Generation.
《艾伦·韦克 2》于去年十月发布，使用了光线追踪和DLSS射线重建技术，因其提供的视觉体验受到了很多关注。本次专题将解释光线追踪器的所有方面，包括管理大量动态内容的加速结构、材质数据的表示和访问、光采样技术，以及图像放大和去噪方法。涵盖了所有关键技术，包括使用不透明度微映射（OMM）处理Alpha测试的资源（如植被和头发）、在光线追踪着色器中使用着色器执行重排序（SER），DLSS射线重建技术如何用于生成最终图像，并通过DLSS帧生成实现平滑帧率的方法。
链接：https://schedule.gdconf.com/session/anatomy-of-a-frame-in-cyberpunk-2077-phantom-liberty/899324
作者：Charles Tremblay (VP of Technology, CD Projekt RED)
Cyberpunk 2077 features an incredibly dense and action-packed world that places huge demands on CPU processing—something that is pushed even further in the post-launch Episode Phantom Liberty. This talk dissects, in detail, a single CPU frame sampled from Phantom Liberty, in order to illustrate how the work for the various systems that drive the game are distributed across CPU cores in order to achieve 60 FPS gameplay in performance mode while updating one of the densest video game worlds to date. The process for how the team worked out their CPU budgets is described, as well as how they approached the challenge of balancing the desire to saturate CPU cores with useful work, and system update dependencies. Comparisons are made to CPU frames sampled in different gameplay situations to highlight the differences that can occur based on what is currently happening in the game.
《赛博朋克2077》展现了一个非常密集和动作场面丰富的世界，这给CPU处理带来了巨大的挑战，而在游戏发布后的“幻影自由”剧集中，这种挑战更是被进一步推到了极致。这次讨论将详细解析从“幻影自由”中采样的单个CPU帧，以便说明驱动游戏的各个系统的工作是如何分布在CPU核心上，从而实现在性能模式下保持60FPS的流畅游戏体验，同时更新迄今为止最为密集的游戏世界之一。团队如何确定CPU预算的过程被描述，以及他们如何应对平衡对CPU核心进行充分利用的渴望和系统更新依赖之间的挑战。通过对不同游戏情境中采样的CPU帧进行比较，突出显示了基于游戏当前发生的事情可能产生的差异。
作者：Scott Kircher (Principal Engine Programmer, Graphics, Insomniac Games)
The New York City of Marvel’s Spider-Man 2 is roughly twice as large as that of earlier titles in the series. At the same time, new quality and performance constraints required the creation of additional distant city LODs. The manual distant-LOD geometry pipeline of Marvel’s Spider-Man and Marvel’s Spider-Man: Miles Morales was no longer viable. This lecture describes how Insomniac Games transitioned to automated distant-LOD geometry generation. In particular, it covers in-depth the applications of fundamental mesh analysis concepts (dual meshes, the Euler Characteristic, curvature estimation, etc...) to several difficult problems (beyond standard reduction and remeshing) that arise from the unique constraints regarding distant city LODs present in Marvel’s Spider-Man 2.
《蜘蛛侠：复仇者联盟2》中的纽约市大约是该系列早期作品的两倍大。同时，由于新的质量和性能约束条件，需要创建额外的远程城市LODs。《蜘蛛侠》和《蜘蛛侠：迈尔斯·莫拉莱斯》中的手动远程LOD几何管道已不再可行。本讲座描述了Insomniac Games如何转向自动化远程LOD几何生成。具体而言，它深入介绍了基本网格分析概念（双重网格、欧拉特性、曲率估计等）在解决几个因《蜘蛛侠：复仇者联盟2》中关于远程城市LODs的独特约束条件所引发的若干复杂问题（超出标准减少和重塑）中的应用。
链接：https://schedule.gdconf.com/session/automated-testing-roundtable-day-1-process/898690
作者：Andrew Fray (Lead Programmer, Roll7)
ProcessAutomated testing provides a team with more stable software at an earlier point in development, allowing easier scheduling and fast iteration. Yet investing in automated testing has a bad reputation in the industry, where just-ship-it is the mantra. Maybe, as increased budgets and team sizes make stability even harder, its time has finally come. On Day 1 we prefer to discuss the people that write the tests and the workflows used to reinforce them. How does the process work in your studio? Are the engineers responsible for automated tests, or QA? Are QA silo'd or integrated? What could be better? What doesn't work?These roundtables bring together people of all experience levels and budgets to discuss how they currently, or would like to, use automated testing of any form. See autotestingroundtable.com for more details and to join our discord!
自动化测试为团队提供了在开发早期获得更加稳定软件的机会，有助于更容易地安排进度和快速迭代。然而，在行业中，投资于自动化测试却颇具负面声誉，因为“只管发布”一直是行业的信条。也许随着预算的增加和团队规模的扩大使稳定性变得更加困难，现在可能是时候采用自动化测试了。在第一天，我们更倾向于讨论编写测试的人员以及用于强化测试的工作流程。在贵工作室中，这一流程是如何运作的呢？工程师负责自动化测试，还是由质量保证部门（QA）负责？QA 是独立运作还是融为一体？有什么可以改善的地方？有哪些事情不起作用？这些圆桌会议聚集了各种经验水平和预算水平的人，讨论他们目前如何使用或希望如何使用任何形式的自动化测试。欢迎访问 autotestingroundtable.com 了解更多详情并加入我们的 Discord！
链接：https://schedule.gdconf.com/session/automated-testing-roundtable-day-2-legacy/903281
作者：Andrew Fray (Lead Programmer, Roll7)
LegacyAutomated testing provides a team with more stable software at an earlier point in development, allowing easier scheduling and fast iteration. Yet investing in automated testing has a bad reputation in the industry, where just-ship-it is the mantra. Maybe, as increased budgets and team sizes make stability even harder, its time has finally come.On Day 2 we prefer to talk about introducing automated testing to an organization. Do you have a culture or a codebase that doesn't currently embrace automated testing? Are you having trouble getting people on-board, or maybe it was very easy and you can share why? Grand failures are just as interesting as successes!These roundtables bring together people of all experience levels and budgets to discuss how they currently, or would like to, use automated testing of any form. See autotestingroundtable.com for more details and to join our discord!
自动化遗产测试为团队提供了更稳定的软件，使得在开发早期阶段就能更容易安排日程并快速迭代。然而，在行业内，投资自动化测试却声誉不佳，因为“只要发货”是口头禅。也许随着预算和团队规模的增加使稳定性变得更加困难，现在是时候投资自动化测试了。第二天我们更倾向于讨论如何将自动化测试引入组织中。您的企业文化或代码库目前是否不支持自动化测试？面临人员难以接受的问题，或者可能很容易，您可以分享一下原因吗？巨大的失败同样引人关注，正如成功案例一样！这些圆桌会议汇集不同经验水平和预算的人，讨论他们目前如何或希望如何使用任何形式的自动化测试。请访问autotestingroundtable.com了解更多详情，并加入我们的discord！
链接：https://schedule.gdconf.com/session/automated-testing-roundtable-day-3-implementation/903282
作者：Andrew Fray (Lead Programmer, Roll7)
ImplementationAutomated testing provides a team with more stable software at an earlier point in development, allowing easier scheduling and fast iteration. Yet investing in automated testing has a bad reputation in the industry, where just-ship-it is the mantra. Maybe, as increased budgets and team sizes make stability even harder, its time has finally come.On Day 3 we prefer to discuss the many faces of automation, from locally-run functional and unit tests, to server-driven continuous integration, and even external device farms. How do you do automated testing in your team? What works well for you? More interestingly, what didn't work and why?These roundtables bring together people of all experience levels and budgets to discuss how they currently, or would like to, use automated testing of any form. See autotestingroundtable.com for more details and to join our discord!
自动化测试的实施使团队在开发的早期阶段就获得了更稳定的软件，从而使调度更容易、迭代速度更快。然而，在行业中，投资自动化测试却声誉不佳，因为“只管上线”是行业的口头禅。也许，随着预算和团队规模的增加使稳定性变得更加困难，现在可能是时候了。在第三天，我们倾向于讨论自动化的许多方面，从本地运行的功能和单元测试，到由服务器驱动的持续集成，甚至是外部设备群。你们团队是如何进行自动化测试的呢？哪些方面对你们效果很好？更有趣的是，有哪些尝试失败了，为什么？这些圆桌会议汇集了各种经验水平和预算水平的人，讨论他们当前如何或希望如何使用任何形式的自动化测试。欢迎访问autotestingroundtable.com了解更多详情并加入我们的Discord！
链接：https://schedule.gdconf.com/session/automated-testing-of-shader-code/899160
作者：Keith Stockdale (Senior Software Engineer, Rare Ltd)
Rare has a rich culture of writing automated tests for their game and engine code. The 'Sea of Thieves' project has hundreds of thousands of automated tests that exist to validate behaviour and to ensure that bugs are kept to a minimum, as they continually deliver an ever-changing experience to players. Among this large test suite, are tests that validate and verify shader code. This session explores the technologies that Rare currently uses to write tests for their shader code, both on 'Sea of Thieves' and on their upcoming game, 'Everwild'. Throughout this session Rare discusses two forms of testing shader code:1. Screenshot Comparison2. Shader unit testsThey go through the pros and cons of both types of tests, as well as delving into the implementation of their shader unit testing framework to discuss the design goals of the system.
Rare 公司在他们的游戏和引擎代码中有着丰富的自动化测试编写文化。《Sea of Thieves》项目拥有数十万个自动化测试，用于验证行为并确保缺陷被最大程度地减少，因为他们不断为玩家提供不断变化的体验。在这个庞大的测试套件中，有一些测试用于验证和验证着色器代码。本次讨论将探讨 Rare 目前用于编写着色器代码测试的技术，包括对《Sea of Thieves》和即将推出的游戏《Everwild》的测试。在本次讨论中，Rare 探讨了两种测试着色器代码的形式：1. 屏幕截图比较 2. 着色器单元测试。他们将详细讨论这两种测试的优缺点，并深入探讨他们的着色器单元测试框架的实现，以讨论该系统的设计目标。
作者：Brent Schiestl (Director of Product Management, Perforce Software)、David Lewak (Senior Software Engineer, Perforce Software)
Building an award-winning game is complicated; your branching strategy shouldn't be. With projects spanning thousands-to-millions of files, limitations of your versioning system should not dictate how you set up your branches or make you think twice about implementing methodologies such as "branch per bug." Perforce Streams helps teams keep things simple, allowing you to design and automate development and release processes in Helix Core. In a recent survey, 73% of organizations agreed that Helix Core is best-in-class for visualizing your development workflow and preventing common issues like merge conflicts or broken builds. Streams easily organizes changes throughout your project's lifecycle, guiding teams through branching and merging with no external tools or scripts required. In this talk, we'll introduce a new way to achieve lightweight branching via Sparse Streams. This latest feature in Helix Core will enable your team to branch smarter to build better games, significantly reducing the amount of metadata produced. With benefits like increased developer velocity and lower data storage requirements, Sparse Streams solves the need for lightweight branching for both development and release scenarios. Join Director of Product Management, Brent Schiestl and Senior Software Engineer, David Lewak as they demonstrate how lightweight branching can: -Save your team time and resources. -Significantly reduce the amount of data stored per branch. -Quickly add new branches without paying the metadata population cost.
开发一款屡获殊荣的游戏是复杂的；而你的分支策略则不应该复杂。在涵盖数千到数百万个文件的项目中，你版本管理系统的限制不应该决定你如何设置分支，也不应该让你对实施“分支修复bug”等方法产生犹豫。Perforce Streams帮助团队保持简单，让你能够在Helix Core中设计和自动化开发及发布流程。在最近的一项调查中，73%的组织都认为Helix Core是最佳选择，可以使你的开发工作流程可视化，并预防合并冲突或构建失败等常见问题。Streams能够轻松组织项目生命周期中的变更，引导团队进行分支和合并，无需外部工具或脚本。在这里，我们将介绍通过Sparse Streams实现轻量级分支的新方法。Helix Core中的这一最新功能将使你的团队能够更智能地进行分支，构建更出色的游戏，大大减少生成的元数据量。通过增加开发速度和降低数据存储需求等好处，Sparse Streams解决了对于开发和发布场景都需要轻量级分支的需求。加入产品管理总监Brent Schiestl和高级软件工程师David Lewak，他们将演示轻量级分支可以：-节省团队的时间和资源。-显著减少每个分支存储的数据量。-快速添加新分支而无需支付元数据填充成本。
作者：Jason Bright (Senior Software Engineer, Electronic Arts)
Frostbite is known as a high-performance engine for graphically intense titles (Battlefield, FC, Need For Speed, Dead Space). Graphics technology is having a generational change from direct submission style APIs (e.g DX9/11, OpenGL) to a family of more modern and high performance technologies (DX12, Vulkan, Metal). This transition happened at the same time Electronic Arts wanted to bring Frostbite to a number of new platforms (iOS, Android, Stadia, Switch). In this talk, Senior Software Engineer Jason Bright shares how EA introduced new technologies and staged the introduction with minimal disruption to active development.
Frostbite 被认为是一个用于图形密集型游戏（《战地》、《刺客信条》、《极品飞车》、《死亡空间》等）的高性能引擎。图形技术正经历从直接提交式API（比如DX9/11、OpenGL）向更现代、高性能技术族（DX12、Vulkan、Metal）的一代变革。这种转变发生在当时 Electronic Arts 想要将 Frostbite 引擎引入一些新平台（iOS、Android、Stadia、Switch）的同时。在这个演讲中，资深软件工程师 Jason Bright 分享了 EA 如何引入新技术，并如何分阶段引入以尽可能减少对现有开发的干扰。
Build the Next Break Out Game with Helix Core Cloud' – Turnkey Version Control for Teams Under 50 (Presented by Perforce)
作者：Brad Hart (CTO and VP of Product Management , Perforce Software)
Whether you're an indie team of three or a mid-sized studio, building a game today means dealing with expanding dimensions of scale, such as large file sizes and number of assets.
Scalable version control is a necessity for efficient development and global collaboration, but it can be hard for small teams to get up and running. That's why Perforce launched Helix Core Cloud — a Perforce-managed and hosted offering of their industry-leading version control platform, Helix Core.
In this talk, Perforce CTO Brad Hart will introduce Helix Core Cloud and provide a primer on what scalable version control means within the context of digital creation efficiency and the changing landscape of game development.
无论你是由三人组成的独立开发团队，还是一个中等规模的工作室，如今开发游戏意味着要处理不断扩大的规模，比如庞大的文件大小和资产数量。
可扩展的版本控制对于高效的开发和全球协作至关重要，但对于小团队来说可能很难上手。这就是为什么 Perforce 推出了 Helix Core Cloud —— Helix Core 平台的 Perforce 管理和托管版本。该平台是业内领先的版本控制平台。
在这次演讲中，Perforce 首席技术官 Brad Hart 将介绍 Helix Core Cloud，介绍可扩展的版本控制在数字创作效率和游戏开发不断变化的环境中的意义。
链接：https://schedule.gdconf.com/session/building-game-worlds-with-openusd/899148
作者：Vesa Paakkanen (Lead Tools Programmer, Remedy Entertainment Plc)
The complexity and size of the game worlds is ever-increasing and tools need to meet that demand. At Remedy, we have been working on new tools and tech stack to allow us to build worlds faster and with more detail than ever before. At the core of our solution is Universal Scene Description (OpenUSD) by Pixar which provides a powerful composition engine as well as a format compatible with the DCCs. In this presentation, we share how we integrated OpenUSD into our tools and engine, what kind of technical and user experience challenges we have faced, and show examples how we have tackled those while having two major game projects using it throughout the development.
游戏世界的复杂性和规模不断增加，工具需要满足这一需求。在 Remedy，我们一直在研发新工具和技术堆栈，使我们能够比以往更快速、更详细地构建世界。我们解决方案的核心是由Pixar开发的Universal Scene Description（OpenUSD），它提供了强大的构图引擎，同时兼容于DCCs格式。在这个演示中，我们分享了如何将OpenUSD整合到我们的工具和引擎中，我们面临了哪些技术和用户体验挑战，以及展示了我们是如何在开发过程中处理这些挑战的，同时展示了两个主要游戏项目如何使用它。
链接：https://schedule.gdconf.com/session/building-the-fully-connected-multiverse-in-suicide-squad/902875
作者：Ingimar Guðmundsson (Advanced AI Programmer, Rocksteady Studios Ltd.)
This talk presents a method of how to support an environment design in a streaming open world that changes greatly over time within the game. The design allows for navigation data to be built, such that a predictable minimum set of offline data is readily available to complement the changing nature of the environment. Within this framework the talk will detail how to build connected navigation data that spans all surfaces, allowing agents to traverse walls, fly and reach the player at any location, quickly. Finally we address how to complement the system with dynamic elements that affect navigation.
这篇讲话介绍了一种在游戏中支持随着时间大幅度变化的流动开放世界环境设计的方法。这种设计允许构建导航数据，使得可预测的最小一组离线数据随时可用以补充环境变化的特性。在这个框架内，讲话将详细介绍如何构建跨越所有表面的连接导航数据，使得代理能够穿越墙壁、飞行并在任何位置迅速到达玩家。最后，我们将讨论如何通过动态元素来增强影响导航的系统。
链接：https://schedule.gdconf.com/session/choreographing-superpowered-ai-movement-in-suicide-squad/902876
作者：Lingxiao Fu (Advanced AI Programmer, Rocksteady Studios)
This talk presents an expressive and scalable AI movement stack that realizes diverse AI traversal behaviors in Suicide Squad: Kill The Justice League. A hierarchical multi-modal pathfinding approach operating on hyper-connected navigation data is used to produce effective paths in the complex open world of Metropolis. A novel path-planning approach is introduced between pathfinding and path following to create more varied and nuanced traversal decisions. AI path is annotated and broken down into featured patterns based on agent movement constraints and preference and then solved by an HTN-style planner into traversal plans. An adaptive traversal plan execution layer allows AI agents to carry out the planned traversal with the ability to repair and adjust their actions on the fly.
这个讲座介绍了一个富有表现力和可伸缩性的 AI 移动技术栈，在《自杀小队：杀戮正义联盟》中实现了多样化的 AI 遍历行为。采用在超连接式导航数据上运行的分层多模式寻路方法，以在大都市这一复杂开放世界中生成有效路径。在寻路和路径跟随之间引入了一种新颖的路径规划方法，以创建更丰富和细致的遍历决策。AI 路径根据代理移动约束和偏好被注释并分解为特征模式，然后由 HTN 风格的计划者解决为遍历计划。自适应遍历计划执行层允许 AI 代理执行计划的遍历，并可以在行动中进行修复和调整。
链接：https://schedule.gdconf.com/session/compile-score-slay-the-c-build-time-dragon/903039
作者：Ramon Viladomat (Senior Staff Gameplay Programmer, Santa Monica Studio)
Build times have been affecting productivity and iteration times since the dawn of programming. Developer improvement efforts have been based on guesswork so far considering the compiler has always been a closed box. 'Compile Score' allows us to peek inside and dissect how the compiler digests the code empowering the engineer to make targeted improvements and identify the real bottlenecks.This presentation covers how to extract, interpret, and act with the provided data, as well as go through C++ best practices to keep the dependencies healthy. It also goes through the setup in Santa Monica Studio to get up to date information automatically to the whole team.
自编程诞生以来，构建时间一直影响着生产效率和迭代速度。开发者的改进工作一直基于猜测，因为编译器一直是一个封闭的黑盒。『编译得分（Compile Score）』使我们能够窥探内部，并解剖编译器如何消化代码，赋予工程师有针对性地进行改进并识别真正的瓶颈的能力。本演示介绍如何提取、解释并处理提供的数据，以及介绍保持依赖关系健康的C++最佳实践。还将介绍在圣莫尼卡工作室设置环境如何自动获取团队整体的最新信息。
作者：Ryan Su (Senior SRE Manager, Tencent Games)、Jane Gao (Senior Director of Product Marketing, Tencent Cloud)
In this insightful session hosted by Tencent Cloud, experts delve into the ever-evolving world of game development in a multicloud environment. The focus is on how developers can leverage the power of multiple cloud services to create games that are not only resilient to various challenges but also scalable to meet the demands of a global audience. The speakers will discuss cutting-edge strategies and best practices for overcoming common hurdles in game hosting, such as handling high traffic spikes, ensuring consistent performance, and maintaining uptime during peak periods. They will also explore how multicloud solutions can foster innovation in game design and delivery, providing a seamless experience for players worldwide. This session is ideal for game developers, producers, and IT professionals looking to enhance their knowledge in multicloud gaming infrastructure and take their projects to the next level in terms of reliability, scalability, and player satisfaction.
在由腾讯云主办的这场富有洞察力的讨论会中，专家们深入探讨了多云环境中不断发展的游戏开发世界。重点是开发者如何利用多个云服务的力量来创建既能够应对各种挑战又能够满足全球受众需求的游戏。演讲者将讨论关于克服游戏托管中常见障碍的尖端策略和最佳实践，例如处理高流量峰值、确保稳定性能以及在高峰期间保持正常运行时间。他们还将探讨多云解决方案如何促进游戏设计和交付的创新，为全球玩家提供无缝体验。这场讨论会适合游戏开发者、制作人和IT专业人士，他们希望加强在多云游戏基础设施领域的知识，并在可靠性、可扩展性和玩家满意度方面将其项目推向新的高度。
作者：Jesse Werner (Senior Gameplay Programmer, Insomniac Games)
Marvel’s Spider-Man 2 is a fast-paced, high-spectacle, action game where Marvel Super Heroes team up to take down both street-level criminals and high-stakes Super Villains. Having a hero as agile and powerful as Spider-Man challenged us to find creative solutions to create powerful and responsive allies. Allies need to match pace with Spider-Man, adapt to each player’s playstyle, and be present throughout gameplay in a way that is believable. In this talk I will detail the methods we used to adapt to player input and playstyle, build sharable ally modules that can be customized for character expression, and create team up mechanics that dynamically support player actions to add spectacle without disrupting player intent.
《漫威蜘蛛侠2（Marvel’s Spider-Man 2）》是一款快节奏、高视觉效果的动作游戏，漫威超级英雄们团结一致，共同打倒街头犯罪分子和面临高风险的超级反派。拥有像蜘蛛侠这样敏捷且强大的英雄，挑战着我们寻找创造性解决方案，创造强大且反应灵敏的盟友。盟友需要与蜘蛛侠保持同步，适应每位玩家的游戏风格，并在整个游戏过程中以一种可信的方式存在。在本次讨论中，我将详细介绍我们用来适应玩家输入和游戏风格的方法，构建可共享的盟友模块，可以根据角色表现进行定制，并创建支持玩家行动的团队协作机制，以添加视觉效果，而不会影响玩家意图。
作者：Sylvie Sherman (Lead Engineer and Project Lead, Survios)
Eugene and Sylvie, the lead engineers of 'Creed: Rise to Glory', explore the long journey of supporting the code base over the last 6 years. The presentation explores mistakes and lessons learned from porting Creed across numerous VR platforms and making the transition to second generation VR. More broadly, the presentation provides an overview of the evolution of VR technology and design with respect to game development.The presentation offers insights of how 'Creed' was optimized for the tethered VR platforms like Quest. Furthermore, the lead engineers describe how the game was remastered for PS VR2 with the Championship Edition and delivered on cross-play with the Quest players for the first time.
尤金和西尔维，即《Creed: Rise to Glory》的首席工程师，探究了过去六年里支持代码库的漫长旅程。该演示探讨了在将Creed移植到众多VR平台并转向第二代VR过程中所学到的错误和教训。更广泛地，演示提供了关于VR技术和设计在游戏开发中的演变概述。演示提供了《Creed》如何针对像Quest这样的有线VR平台进行优化的见解。此外，首席工程师描述了如何使用冠军版对游戏进行了PS VR2的翻新，并首次实现了与Quest玩家的跨平台游戏。
作者：Bradley Pollard (Engineering Lead, Frontier Developments)
In this session, learn how and why Frontier Developments built a new deterministic simulation framework that underpins everything players see and interact with in Warhammer Age of Sigmar: Realms of Ruin. Engineering Lead Bradley Pollard covers the journey taken by the development team—from implementing a thread-safe ECS to creating tooling to analyze and prevent desynchronizations. Learn how determinism benefits both the developers, as well as the game, by enabling reproducibility, providing low-cost networking, and even providing a form of anti-cheat. The talk also addresses how Frontier was able to create a framework that is able to run on PC, Xbox, PlayStation, and Linux, and ensure determinism.
在这个讲座中，了解为什么Frontier Developments建立了一个新的确定性模拟框架，支撑着玩家在《战锤时代：毁灭领域》中看到并与之交互的一切。工程师领航 Bradley Pollard详细介绍了开发团队的工作历程——从实现线程安全的ECS到创建分析工具以预防数据不一致。了解确定性如何使开发人员和游戏受益，通过实现可再现性、提供低成本的网络连接，甚至提供一种防作弊方法。演讲还讨论了Frontier是如何创建能够在PC、Xbox、PlayStation和Linux上运行并确保确定性的框架。
'Crystal of Atlan': Cross-Server Architecture Design and Optimization (Presented by Crystal of Atlan)
作者：Shuangxing Gao (Lead Server Programmer of Crystal of Atlan, ByteDance)
The speech focuses on Crystal of Atlan, a MMOARPG with millions of DAU, and introduces the design of cross-server operations, optimization of server CPU, memory, and network, as well as its deployment and maintenance.
这篇演讲重点介绍了《亚特兰水晶》（Crystal of Atlan），这是一款拥有数百万日活用户的大型多人在线角色扮演游戏（MMOARPG），并且介绍了跨服务器操作、服务器CPU、内存和网络的优化，以及其部署与维护设计。
作者：Zay Miao (Sr. Software Engineer, very very spaceship)
Inclusion and representation have become ever more important in today's modern world, and giving players the tools to construct their virtual selves does wonders for enhancing immersion and engagement. It's almost a given that AAA console and PC games have robust character creators, but in the mobile space players are very seldom given the opportunity to build a character that represents themselves. In this session, Zay Miao, senior software engineer/technical artist at very very spaceship, will walk through the development process of building a robust avatar customization system. This talk will demonstrate the challenges of building this system and novel techniques for flexible avatar creation.
在当今现代世界中，包容性（Inclusion）和代表性（representation）变得越来越重要，在游戏中为玩家提供构建他们虚拟自我的工具有助于增强沉浸感和参与度。几乎可以肯定，AAA主机和PC游戏拥有强大的角色创建器，但在移动游戏领域，玩家很少有机会构建能代表自己的角色。在这个环节中，来自"very very spaceship"的高级软件工程师/技术美术师Zay Miao将介绍构建强大的角色自定义系统的开发过程。本次讲话将展示构建该系统的挑战和灵活角色创作的新技术。
Delta Force Techniques From Unified Production Pipeline to Cross-Platform Runtime Support (Presented by Tencent Games)
作者：Shangli Liang (Deputy Director of Game Client Engineering, TiMi Studio Group, Tencent Games)、Yang Shi (Game Client Engineering Team Lead of Delta Force TiMi Studio Group, Tencent Games)
Delta Force is preparing to deliver top-tier experiences to gamers for all platforms. This session will cover an efficient production process while pursuing exceptional development quality as TiMi Studio Group readies for the release of a new edition to a classic FPS franchise. This talk will summarize the valuable experiences learned through the development process of Delta Force regarding various aspects of cross-platform production including asset management, open world solutions, auto scalable 3C, adaptive UI framework and the performance tracking tools we build for better tracking during the unified development process.
'Delta Force'正在准备为各平台的玩家提供一流的游戏体验。本次讨论将涵盖一种高效的制作流程，同时追求卓越的开发质量，TiMi Studio Group正在为一款经典FPS系列的新作发布做准备。这次讨论将总结在开发Delta Force过程中学到的宝贵经验，涉及跨平台制作的各个方面，包括资产管理、开放世界解决方案、自适应UI框架和我们为了更好地追踪统一开发过程中的性能而构建的性能跟踪工具。
作者：Pete Kugler (Engineering Architect, Bungie)、Cameron Konner (Senior Technical Audio Designer, Bungie)
Wwise 2021 introduced the capability to transition across music containers, allowing us to build seamless transitions across music play lists. This inspired new innovations to our audio system, where we could now transition seamlessly back-and-forth between the free-roam open world experience, and in-world activities, like Lost Sectors and Public Events. This talk will be a deep dive into the implementation of our new music system that was shipped with the Lightfall expansion. It will cover the designer setup, scripting, and code architecture used to build up that system. Additionally, the talk will provide an overview of our unique audio data compilation pipeline and how we leverage WWAPI automation in our studio for content setup and testing.
Wwise 2021 引入了跨音乐容器过渡的能力，使我们能够在音乐播放列表之间构建无缝过渡。这激发了我们音频系统的新创新，我们现在可以在自由漫游的开放世界体验和地图内活动（如迷失领域和公共事件）之间实现无缝过渡。这次讨论将深入探讨我们携带 Lightfall 扩展中新音乐系统的实施。它将涵盖设计师设置、脚本编写和编码架构等构建该系统所使用的内容。此外，本次讨论还将概述我们独特的音频数据编译流程以及在内容设置和测试方面如何利用 WWAPI 自动化工具。
作者：Shawn Hargreaves (Dev Manager, Direct3D, Microsoft)、Austin Kinross (PIX Developer Lead, Microsoft)、Wessam Bahnassi (Principle Engineer, Developer Technology, Nvidia)、Rob Martin (Fellow Software Engineer, AMD)
The DirectX team will showcase the latest updates, demos, and best practices for game development with key partners from AMD and NVIDIA. Work graphs are the newest way to take full advantage of GPU hardware and parallelize workloads. Microsoft will provide a preview into DirectSR, making it easier than ever for game devs to scale super resolution support across Windows devices. Finally, dive into the latest tooling updates for PIX.
DirectX团队将展示与AMD和NVIDIA等重要合作伙伴一起进行游戏开发的最新更新、演示和最佳实践。工作图是充分利用GPU硬件并将工作负载并行化的最新方法。微软将提供DirectSR的预览，使游戏开发人员能够更轻松地在Windows设备上扩展超级分辨率支持。最后，深入了解PIX的最新工具更新。
作者：Zac Litton (VP of Engineering, The Believer Company)
This presentation is going to discuss the vectors of engineering leadership, in theory and example, focusing on how they operate at different studio levels and organizations. It is going to discuss key ways teams can find, nurture, and evaluate success in leadership, as well as ways individuals can grow themselves. Additionally, it shares engineering organizational processes that have proven to aid in visibility, collaboration, and trust in highly creative studios.
这次演示将讨论工程领导力的向量，理论和实例，重点关注它们在不同工作室层面和组织中的运作方式。它将讨论团队发现、培养和评估领导力成功的关键方法，以及个人成长的途径。此外，它还分享了在高度创意的工作室中证明有助于可见性、协作和信任的工程组织流程。
链接：https://schedule.gdconf.com/session/ecs-in-practice-the-case-board-of-alan-wake-2/899798
作者：Alexander Balakshin (Senior Gameplay Programmer, Remedy Entertainment)
For Alan Wake 2, Remedy Entertainment is using a new data-oriented ECS gameplay framework instead of the previous object-oriented one. As a result, we have performance gains such as cache coherency and multithreading out of the box. Also, being implemented using some of the latest C++ features, the new framework changes the way gameplay programmers write their code by providing them some extra convenience and productivity.We will present how this framework is used for implementing one of the game's features – the Case Board. We will cover how we describe the Case Board objects with components, how we register their systems, and how we interact with other modules (camera, cursor, etc.). Finally, we'll show how the interaction between C++ code and Lua scripts is organized.
对于《阿兰·韦克 2》，Remedy Entertainment使用一种新的面向数据的ECS（Entity-Component-System）游戏框架，而不是之前的面向对象的框架。因此，我们获得了诸如缓存一致性和多线程等性能提升。此外，新框架使用了一些最新的C++特性，改变了游戏程序员编写代码的方式，为他们提供了额外的便利性和生产力。我们将展示这一框架如何用于实现游戏特性之一——“案例面板”。我们将介绍如何用组件描述案例面板对象、如何注册它们的系统，以及如何与其他模块（相机、光标等）进行交互。最后，我们将展示C++代码与Lua脚本之间的交互是如何组织的。
作者：Mans Isaksson (Game Programmer, Embark Studios)
THE FINALS is a fast-paced, competitive first person shooter with a focus on emergent gameplay and dynamic systems. Front and center of these systems is the destruction system which enables most of the environment to be destroyed, buildings to collapse, and debris from buildings to re-shape the play-space, all while being physically simulated.In this session, Måns Isaksson (Game Engineer at Embark Studios) walks through the technical challenges the team at Embark had to overcome to enable this level of destruction in THE FINALS. He walks you through the workflow – from constructing and fracturing structures to integrating them into the playable environment, and explore technical aspects such as replication, simulation performance, rendering, and structural analysis calculations.
《THE FINALS》是一款以新生玩法和动态系统为重点的快节奏、竞技性第一人称射击游戏。系统的核心是破坏系统，使得大部分环境可以被摧毁、建筑物会倒塌，建筑残骸会重新塑造游戏空间，同时这一切都是通过物理仿真实现的。在本次讲座中，Embark Studios的游戏工程师Måns Isaksson将演示团队在使《THE FINALS》中的破坏效果达到这一水平时所面临的技术挑战。他将带领你了解从构建和破坏结构，到将其整合到可玩环境中的工作流程，并探讨复制、仿真性能、渲染和结构分析计算等技术方面的内容。
作者：Kylan Gibbs (Co-Founder, Inworld AI)、Haiyan Zhang (General Manager of Gaming AI, Xbox)
AI has long been a component of game development from Nim, the first AI-based game developed in 1951, to procedural generation in Rogue-likes. With the rise of generative AI, there are now new opportunities to leverage AI to create innovative games.
In this session, we'll provide an early look at AI tools designed to help game devs prototype their game narratives – and more easily facilitate real-time interactivity with AI NPCs. We'll share our roadmap for what's to come and how we're working with developers to create tools that matter to them.
人工智能 (AI) 一直是游戏开发的一部分，从1951年开发的第一个基于AI的游戏《Nim》到Rogue-like游戏中的程序化生成。随着生成式AI的兴起，现在有新的机会利用AI来创造创新的游戏。
在这个场景中，我们将提供早期介绍设计用于帮助游戏开发者原型他们游戏故事情节的AI工具 – 并更轻松地促进与AI NPC进行实时互动。我们将分享未来的发展路线图，以及我们如何与开发者合作，创造对他们有意义的工具。
链接：https://schedule.gdconf.com/session/fp16-shaders-in-frostbite/899054
作者：Alexis Griffin-Lira (Rendering Engineer, Electronic Arts)
This talk describes how Frostbite implemented multi platform 16 bit type (FP16) support for shaders in order to increase shader performance, especially on mobile. 16 bit shader types like half and short have advantages over 32 bit types, like occupancy and ALU performance. However, converting shaders to take advantage of half precision types is often not a straightforward process, in particular when accounting for platform and driver quirks.
This talk will cover the process of creating Frostbite's FP16 API, which enables shader programmers to write FP16 shaders while the tools bridge the platform gaps under the hood. The talk will also explain the rationale behind embracing FP16 in our shaders, the various challenges and quirks of FP16 in different platforms, and how we worked around them, and, finally, it concludes by presenting some of our performance and visual results on various console and mobile platforms.
这篇演讲描述了Frostbite 如何实现了多平台 16 位类型（FP16）支持用于着色器，以提高着色器性能，特别是在移动设备上。16 位着色器类型如 half 和 short 相比于 32 位类型具有优势，比如占用率和 ALU 性能。然而，将着色器转换以利用半精度类型通常不是一个简单的过程，特别是要考虑平台和驱动程序的怪癖。
这次演讲将涵盖创建 Frostbite 的 FP16 API 的过程，该 API 使着色器程序员能够编写 FP16 着色器，而工具将在底层处理平台差异。演讲还将解释在我们的着色器中采用 FP16 的原因，不同平台上使用 FP16 遇到的各种挑战和怪癖以及我们如何解决这些问题，最后，通过在各种游戏主机和移动设备平台上的性能和视觉效果结果来总结。
作者：Stefan Maurus (Gameplay Programmer & Lead Integrator, GIANTS Software GmbH)
The Farming Simulator game series showcases perhaps the most complex vehicles in the industry. In this talk it is all about the integration of those vehicles into the game, progressing from static 3D models to fully animated vehicles controllable by the user. Stefan Maurus, the lead integrator of GIANTS Software will explain the challenges his team faced by increasingly complex farming machinery and share insights into how they addressed specific cases. Additionally, he'll show clever techniques implemented to optimize performance. Given the game's highly modifiable nature, he'll also show you what has been done ensuring that all this technology is accessible to modders, providing a deeper understanding of the game's mod support.
《农业模拟器》游戏系列展示了游戏行业中可能是最复杂的车辆。本次讲座主要讨论如何将这些车辆整合到游戏中，从静态的3D模型发展到完全由用户控制的完全动画化的车辆。GIANTS Software首席整合师Stefan Maurus将解释他的团队面临的不断增加复杂度的农业机械带来的挑战，并分享他们如何解决特定情况的见解。此外，他还将展示已实施的优化性能的巧妙技术。鉴于游戏的高度可修改性，他还将展示确保所有这些技术对修改者都是可访问的措施，以便更深入地了解游戏的修改支持。
链接：https://schedule.gdconf.com/session/game-audio-programming-roundtable-day-1-game-integration/899288
作者：Guy Somberg (Lead Gameplay Engineer, Echtra Games)、Helen Yang (Audio Programmer, Epic Games)
Wednesday: "Audio Programming Roundtable: Game Integration"
This roundtable session will focus on tools and techniques for integrating with game engines. Topics will center around technology choices, game audio engine design, and tool design. Newbies and experts are welcome to come and share their wisdom.
Thursday: "Audio Programming Roundtable: Low-Level Topics"
This roundtable session will focus on low-level topics such as DSPs and mixing. Topics will center around algorithms, interesting effects to implement, and integrating those effects into workflows. Newbies and experts are welcome to come and share their wisdom.
Friday: "Audio Programming Roundtable: Free-for-all"
In this roundtable session we will cover any topics that we didn't get to in the previous two days. We will also have a Q&A session where your questions can be answered by expert audio programmers. Newbies and experts are welcome to come and share their wisdom.
周三：「音频编程圆桌讨论：游戏集成」
本次圆桌讨论将聚焦于与游戏引擎集成的工具和技术。讨论主题将围绕技术选择、游戏音频引擎设计和工具设计展开。欢迎新手和专家前来分享经验。
周四：「音频编程圆桌讨论：低层级主题」
本次圆桌讨论将聚焦于诸如 DSP 和混音等低层级主题。讨论主题将围绕算法、有趣的效果实现以及将这些效果集成到工作流程中展开。欢迎新手和专家前来分享经验。
周五：「音频编程圆桌讨论：自由讨论」
在本轮圆桌讨论中，我们将讨论前两天没能涉及到的任何主题。我们还将设置一个问答环节，让专业音频程序员回答你的问题。欢迎新手和专家前来分享经验。
链接：https://schedule.gdconf.com/session/game-audio-programming-roundtable-day-2-low-level-topics/903298
作者：Guy Somberg (Lead Gameplay Engineer, Echtra Games)、Helen Yang (Audio Programmer, Epic Games)
This roundtable session focuses on low-level topics such as DSPs and mixing. Topics will center around algorithms, interesting effects to implement, and integrating those effects into workflows. Newbies and experts are welcome to come and share their wisdom.
这个圆桌讨论会专注于低级话题，比如 DSPs 和混音。讨论的主题将围绕算法、有趣的实现效果以及将这些效果整合到工作流程中展开。新手和专家均欢迎前来分享他们的经验。
链接：https://schedule.gdconf.com/session/game-audio-programming-roundtable-day-3-free-for-all/903299
作者：Guy Somberg (Lead Gameplay Engineer, Echtra Games)、Helen Yang (Audio Programmer, Epic Games)
In this roundtable session, we cover any topics that we didn't get to in the previous two days. We also have a Q&A session where your questions can be answered by expert audio programmers. Newbies and experts are welcome to come and share their wisdom.
在这个圆桌讨论会上，我们会涵盖前两天未能讨论到的任何话题。我们还会设置一个问答环节，专家音频程序员会回答你的问题。新手和专家都可以前来分享他们的智慧。
Game Optimization With The Radeon™ Developer Tool Suite, and Shader Instrumentation With GPU Reshape (Presented by AMD)
作者：Miguel Petersen (Senior Rendering Engineer, Striking Distance Studios (Spain))、Amit Mulay (Senior Software Development Engineer, AMD)、Chris Hesik (Principal Member of Technical Staff, AMD)
In this two-part session, we will first present the latest enhancements in the Radeon Developer Tool Suite, a rich set of GPU profilers and analyzers and a recently-added GPU crash analysis tool. We will show how developers can use the tools to optimize game performance and get insights into how games are running on AMD Radeon GPUs. We will also present a completely redesigned Radeon Developer Panel and show how it can be used to quickly obtain the data to feed into the other tools in the suite.Next, Miguel (Striking Distance Studios) will present GPU Reshape, a just-in-time instrumentation framework with instruction level validation of shader operations. We discuss the challenges with instrumenting across multiple intermediate representations, and how a unified model was achieved through a custom intermediate language featuring seamless bi-directional translation. We'll take a deep dive into current validation methodologies and what the future of instrumentation may look like.
在这两部分的会议中，我们首先会介绍最新的 Radeon 开发者工具套件的增强功能，这是一个丰富的 GPU 分析器和分析工具集，以及最近新增的 GPU 崩溃分析工具。我们将展示开发者如何使用这些工具来优化游戏性能，并了解游戏在 AMD Radeon GPU 上的运行情况。我们还会介绍一个全新设计的 Radeon 开发者面板，并展示如何快速获取数据以供套件中其他工具使用。接下来，Miguel（Striking Distance Studios）将介绍 GPU Reshape，这是一个即时的仪器化框架，用于对着色器操作进行指令级验证。我们将讨论在多个中间表示之间进行仪器化的挑战，以及如何通过一个自定义中间语言实现了统一模型，具有无缝双向转换功能。我们将深入探讨当前的验证方法论以及仪器化的未来发展方向。
Global Illumination With AMD FidelityFX™ Brixelizer, Plus AMD FidelityFX SDK Updates (Presented by AMD)
作者：Dihara Wijetunga (Senior Software Engineer, AMD)
This talk will briefly discuss how the AMD FidelityFX™ Brixelizer works, then explore how diffuse and specular global illumination is implemented with sparse distance fields in Brix GI. This will provide a step-by-step breakdown of the algorithm and provide some of the solutions to problems such as populating the radiance cache and handling static and dynamic geometry. The talk will also feature image quality comparisons to ground truth ray-traced global illumination and some initial performance numbers to provide the audience with an idea of what to expect from Brix GI. It will also showcase the Brix GI API and how it integrates into an existing Brixelizer integration.This talk will also include a preview of new samples releasing with the new version of the FidelityFX SDK, as well as an update on the changes made to existing samples.
这次讲话将简要讨论 AMD FidelityFX™ Brixelizer 的工作原理，然后探讨在 Brix GI 中如何利用稀疏距离场来实现漫反射和高光全局照明。将逐步解析算法，并提供一些解决方案，如如何填充辐射缓存以及处理静态和动态几何体等问题。讲话还将展示图像质量与基准光线追踪全局照明的比较，以及一些初始性能数据，让观众了解对于 Brix GI 可能会有什么期待。同时，还会展示 Brix GI API 以及其如何集成到现有的 Brixelizer 集成中。这次讲话还将包括一个新版本的 FidelityFX SDK 中发布的新样本的预览，以及对现有样本所做更改的更新。
作者：Shenshen Cui (Staff Developer Relations Engineer, Google)、Patrick Martin (Developer Relations Engineer, Google)
This session will discuss a high level overview of best practices for handling input in mobile vs PC games, including technical details on how to implement these best practices in Android for phones, tablets, PCs, and Chromebooks to help developers give their players a first class gaming experience.
本次讨论将提供移动和PC游戏中处理输入最佳实践的高层概述，包括如何在Android手机、平板电脑、PC和Chromebook上实现这些最佳实践的技术细节，帮助开发人员为玩家提供一流的游戏体验。
Google Developer Summit: Beyond "Set and Forget": Advanced Debugging with Firebase Crashlytics (Presented by Google)
作者：Joe Spiro (Developer Relations Engineer, Google)
It's time to move beyond simply catching bugs; Crashlytics has added a number of features that make detecting, tracking and understanding bugs all the way from high level to native code even easier. Take your fixes to another level with native stack traces, memory debugging and the ability to log uncaught exceptions as fatal. Now developers can better understand and fix issues that occur at some of the deepest and most obscure parts of their apps.
是时候超越简单地捕捉错误了；Crashlytics 已经添加了许多功能，使得从高级别到本地代码层面检测、跟踪和理解错误变得更加容易。通过原生堆栈跟踪、内存调试以及记录未捕获异常为致命错误，将修复问题提升到另一个水平。现在开发者可以更好地理解和修复发生在其应用程序最深和最隐蔽部分的问题。
作者：Zoey Fan (Senior Product Manager, Flutter, Google)、Brett Morgan (Developer Relations Engineer, Google)
Today, Flutter is used in more than 1M apps from small studios to large companies like Lotem and PUBG who are building beautiful, natively compiled, multi-platform applications from a single codebase. Casual games have become a big part of the Flutter ecosystem with the Casual Games Toolkit, a collection of free and open source tools, templates, and resources to make game dev more productive with Flutter.
今天，Flutter 被用于超过 100 万款应用程序，从小工作室到大公司，如 Lotem 和 PUBG，他们正在利用单一代码库构建漂亮、本地编译的跨平台应用程序。休闲游戏已成为 Flutter 生态系统的重要组成部分，通过休闲游戏工具包提供了一套免费开源的工具、模板和资源，使使用 Flutter 开发游戏更加高效。
作者：Mark Mandel (Developer Advocate, Google)
Google Cloud is an ecosystem for live games. In this session, developers will survey the breadth of Google Cloud offerings through a working project framework capable of supporting a global-scale game.
Google Cloud 是一个针对即时游戏的生态系统。在这个课程中，开发人员将通过一个能够支持全球规模游戏的工作项目框架来全面了解 Google Cloud 提供的各种服务。
Harnessing Generative AI to Create Unlimited Content Within Game Worlds (Presented by Amazon Web Services)
作者：Russ Harding (Chief Gaming Officer, Saltwater Games)
Join Saltwater Games and walk through the generative AI architecture of the world of Resurgence, an upcoming free-to-play, post-apocalyptic, open-world crafting and survival game. Learn about the generative AI technology behind the in-game ThorAI (powered by Amazon Bedrock), and dive deep into fully voiced, interactive in-game NPCs in the companion mobile game Missing. Explore real-world use cases for a variety of generative AI services and solutions.
加入Saltwater Games，并探索即将推出的自由游戏《Resurgence》，这是一个后启示录、开放世界手工制作与生存游戏，其中包含生成式人工智能架构。了解游戏内部ThorAI（由亚马逊Bedrock技术提供支持）背后的生成式人工智能技术，并深入探索同伴手机游戏《Missing》中完全配音的互动NPC。探索各类生成式人工智能服务与解决方案的真实世界应用案例。
High Performance Rendering in Snowdrop using AMD FidelityFX™ Super Resolution 3 (FSR 3) (Presented by AMD)
作者：Hampus Siversson (Associate Lead Engine Programmer, Massive Entertainment, Massive Entertainment)、Colin Riley (Senior Manager Software Development, AMD)
AMD FidelityFX™ Super Resolution 3 (FSR 3) is AMD's open-source upscaling and frame interpolation solution. Massive Entertainment integrated AMD FSR 3 technology in the Snowdrop engine, which supports a variety of titles and platforms.
Learn about how the integration has been done to ensure FSR 3 is well integrated into their pipeline, which issues they were facing and how the integration helped to improve FSR 3 itself during its development process.
AMD FidelityFX™ Super Resolution 3（FSR 3）是AMD的开源提升分辨率和帧插值解决方案。Massive Entertainment将AMD FSR 3技术集成到了Snowdrop引擎中，该引擎支持各种游戏和平台。
了解他们如何进行集成以确保FSR 3良好地融入他们的开发流程，他们所面临的问题以及这一集成如何帮助改进FSR 3在其开发过程中。
作者：Doug Sheahan (Sr. Programming Director, Insomniac Games)
In building up on traversal from Marvel's Spider-Man, we wanted to provide players a fresh experience built upon the mechanics they loved. This session talks about how we expanded and evolved our traversal system to include additional mechanics like web wings. You also learn about how the team had to evolve several of our existing mechanics to handle the new speeds the player could traverse at.
在Marvel's Spider-Man游戏中，我们基于已有的穿梭系统，希望为玩家提供一个建立在他们喜爱的机制之上的全新体验。这个讨论会谈到我们如何拓展和演化我们的穿梭系统，增加了像蜘蛛翼这样的额外机制。您还将了解到团队是如何演化我们现有的几项机制，以处理玩家可能以全新速度穿梭的挑战的。
作者：Stef Corazza (Senior Engineering Director, Roblox)、Chen Zheng (Senior Director, Product, Roblox)
The latest AI and LLM (Large Language Models) innovations have paved the way for completely re-imagining the game development process, driving the discovery of new games, and making safer gaming experiences. What was a weeks-long process of creating the first playable prototype can now be achieved within a day. We will explore how developers can leverage AI for everything from texturing 3D models to generating and debugging code and arranging 3D scenes. We will also explore how AI is being used at Roblox to help drive the discovery of new games and match players to the right content. And how AI is helping to create safer and more civil gaming experiences. Lastly, we will showcase how AI enables a new generation of User-Generated Content (UGC) with a conversational interface and how these capabilities will empower developers to make more engaging experiences where their users participate in the creation process.
最新的人工智能和LLM（大型语言模型）创新为完全重新构想游戏开发过程铺平了道路，推动了新游戏的发现，并使游戏体验更加安全。过去需要数周时间才能创建出第一个可玩原型的过程，现在可以在一天内完成。我们将探讨开发者如何利用人工智能从为3D模型贴图、生成和调试代码，到布置3D场景等方面。我们还将探讨Roblox如何利用人工智能来推动新游戏的发现，并将玩家与合适的内容匹配。以及人工智能如何帮助创造更安全、更文明的游戏体验。最后，我们将展示人工智能如何通过对话界面实现新一代用户生成内容（UGC），以及这些能力如何赋予开发者创造更具吸引力的体验，在这些体验中，用户参与到创作过程中。
作者：CJ Williams (Director of Product Management Microsoft Game Creator Services, Microsoft)、Rob Fox (Designer and Producer Gun Interactive, Gun Interactive)
Gun Interactive launched Texas Chainsaw Massacre in August 2023 and player engagement immediately exceeded the team’s expectations. With Playfab as their backend service, they were able to scale quickly and smoothly to meet the demands of rapidly escalating DAU, ensuring the team didn’t miss a beat landing this massive success. Learn how early integration of solutions ranging from Game Servers to CloudScript, Player Profile and more, allowed Gun to depend on Playfab for a smooth and scaled launch and continued ongoing operations.
Gun Interactive 在 2023 年 8 月推出了《德州电锯杀人狂》，玩家参与度立即超出了团队的预期。凭借 Playfab 作为他们的后端服务，他们能够快速、顺利地扩展规模，以满足迅速增长的 DAU 的需求，确保团队在取得这一巨大成功时不落后。了解早期整合解决方案（从游戏服务器到云脚本、玩家资料等），让 Gun 能够依赖于 Playfab 进行平稳和规模化的发布，并持续进行运营。
How We Got 10,000 Players Into a Level (The Tech Behind the Aurora Concert in 'Sky: Children of the Light')
作者：Sam McGrath (CTO, Thatgamecompany Inc)
In December of 2022, thatgamecompany hosted a concert for the artist AURORA inside its game Sky: Children of the Light. For this event, Sam McGrath (CTO) and his engineering team developed technology running in TGC's custom engine that was capable of rendering and networking an unprecedented 4000 simultaneous players together in a single arena. In August of 2023, a revamped version of the concert was hosted, this time supporting over 10,000 simultaneous players. In this talk, McGrath tells the story of the development of Festival Tech, and describes some of the techniques and algorithms employed for rendering and networking thousands of players at once. He tells the story of how the tech was developed, highlighting some engineering insights that transformed a seemingly impossible task into an awe-inspiring experience for Sky's players.
在2022年12月，thatgamecompany 在其游戏《Sky: Children of the Light》内举办了为艺术家 AURORA 所举办的音乐会。为了这次活动，Sam McGrath（首席技术官）及其工程团队开发了能够在 TGC 自定义引擎中运行的技术，能够在一个单一竞技场中渲染和网络连接超过4000名同时玩家。到了2023年8月，音乐会的改版版本再次举办，这一次支持超过10,000名同时玩家。在这次演讲中，McGrath 讲述了节日科技（Festival Tech）的开发故事，并描述了用于同时渲染和联网成千上万名玩家的部分技术和算法。他讲述了这项技术是如何开发的，突出了一些工程领域的见解，这些见解将一项看似不可能的任务转化为《Sky》玩家引以为傲的令人惊叹的体验。
作者：Christopher Shankland (CTO, Odyssey Interactive)
Launch day for an online game is a very exciting time, but often not in a good way for the online or service engineering team. Odyssey Interactive's debut title, Omega Strikers, a competitive online sports-combat game launched on 7 platforms with a services team of 3 reaching hundreds of thousands of players on day 1 with no outages. This session explores the core role that Observability played in making this 'boring launch' the expected outcome. An introduction to what Observability is, some basics of the tech that makes it work, and in-depth examples of how it provides crucial focus and context in high pressure situations will be presented. Don't be scared of launch day again!
线上游戏的上线日是非常令人兴奋的时刻，但通常对于线上或服务工程团队来说，并非总是令人愉快的。Odyssey Interactive的首款作品《Omega Strikers》，一款竞技线上体育格斗游戏在7个平台上发布，服务团队由3人组成，在首日即吸引了数十万玩家，且没有出现任何服务中断。这个讨论将探讨可观测性在实现这种“无聊发布”中扮演的核心作用。介绍了可观测性的概念，一些使其运作的技术基础以及详细示例，展示了它在高压情况下提供关键的焦点和背景的方式。不要再因发布日而感到害怕！
作者：Robin-Yann Storm (Tool UX & Workflow Designer, Independent)
When we make game development tools it is very easy to pick what we think is a 'sensible default', and have all users deal with it. We think those defaults makes sense, and work for most users. However, even a cursory look at DCCs such as Unity, Unreal, 3dsMax, Maya, and Blender shows us that this is wrong. Unity, Max, and Maya have the hierarchy on the left side of the screen, while Unreal and Blender have the hierarchy on the right side of the screen. The same differences can be found between all of them when it comes to toolbars, the 3D view, the property panel, and the asset browser. There is no 'sensible default' for any of those panels in the UI, yet we often hear they 'should' be on the left or 'should' be on the right.
What do we have to unlearn, to improve the UX of our tools?
在开发游戏工具时，很容易选择我们认为是‘合理默认值’，让所有用户去适应。我们认为这些默认设置是合理的，并适用于大多数用户。然而，对于诸如Unity、Unreal、3dsMax、Maya和Blender等DCC软件的简单查看就告诉我们这是错误的。Unity、Max和Maya在屏幕左侧拥有层次结构，而Unreal和Blender则在屏幕右侧拥有层次结构。当涉及到工具栏、3D视图、属性面板和资源浏览器时，它们之间存在相同的差异。在界面中，这些面板没有任何‘合理默认值’，但我们经常听到它们‘应该’在左侧或者‘应该’在右侧。
为了改进我们工具的用户体验，我们需要忘掉什么？
Innovation Unleashed: High-performance UE5 Mobile Rendering and Next-gen Character Creation Pipeline Powered by Machine Learning (Presented by Lightspeed Studios)
作者：Bo Li (Principal Software Engineer, LIGHTSPEED STUDIOS)、Jing Gong (Principal Software Engineer Lead, LIGHTSPEED STUDIOS)、Xin Qiao (Senior Software Engineer, LIGHTSPEED STUDIOS)、Quan Wen (Senior Game Engineer, LIGHTSPEED STUDIOS)
The bar of quality and performance on next-gen platforms has been raised once again with Unreal Engine 5. However, a significant gap has emerged between these advanced features and their mobile counterparts. In this presentation, we will explore how the presenters addressed the challenge of developing high-quality UE5 mobile games involving the implementation of multiple optimizations and creation of new mobile rendering pipelines.
Similarly, MetaHuman by Epic revolutionized character creation technology, yet challenges in performance and customization persist. This session will introduce a groundbreaking character creation pipeline designed for crafting high-definition, animation-ready characters that enhance the immersion in the next-level metaverse. The creators will showcase a novel system, built upon multimodal machine learning techniques, that empowers both artists and gamers to generate hyper-realistic avatars seamlessly from photos, sketches, or even text within seconds.
品质和性能在次世代平台上再次得到提升，而 Unreal Engine 5 的出现将这一水平再次提高。然而，这些先进功能与移动平台的相应功能之间出现了明显差距。在这个介绍中，我们将探讨主持人是如何解决开发高质量 UE5 移动游戏的挑战，涉及多项优化措施的实施以及新移动渲染流程的创建。
同样，Epic 推出的 MetaHuman 形象创作技术已经发生了革命性变革，但在性能和定制化方面仍然存在挑战。这个环节将介绍一个为打造高清晰度、动画就绪角色而设计的开创性角色制作流程，以增强在下一级元宇宙（Metaverse）中的沉浸感。创作者将展示一个新颖的系统，基于多模态机器学习技术构建，赋予艺术家和玩家们生成超逼真化身的能力，可以在几秒钟内从照片、草图甚至文本中无缝生成。
链接：https://schedule.gdconf.com/session/keeping-online-communities-healthy-presented-by-microsoft/903871
作者：Jennifer Mills-Perov (Principal Product Manager Gaming Trust and Safety, Microsoft)、Ivory Harvey (Senior Technical Program Manager Gaming Trust and Safety, Microsoft)
Online gaming communities face new challenges every day that can harm the well-being and security of players. And Content Moderators have to deal with policy breaches, threats, and creating a safe and secure space for everyone. Meanwhile, government regulators are pushing for policies that can help the industry tackle online risks. Community Managers have a large responsibility and Microsoft can support them with innovative tools and solutions that make it easier to build trust with players, drive community engagement, and reduce churn by keeping players playing. In this talk, we will show how AI-powered safety can improve your gaming community creating a safe place that players expect and deserve, making everyone feel welcomed, safe and included when they play.
在线游戏社区每天都面临着可能危害玩家福祉和安全的新挑战。内容审核员需要处理政策违规、威胁等问题，并努力为每个人创造一个安全可靠的空间。同时，政府监管机构正在推动政策，帮助行业应对在线风险。社区经理担负着重大责任，微软可以用创新的工具和解决方案支持他们，帮助他们更轻松地建立与玩家之间的信任，推动社区参与，并通过让玩家持续游戏来减少流失。在这次演讲中，我们将展示如何利用人工智能技术提升你的游戏社区安全性，创造玩家期待和值得的安全场所，使每个人在游玩时都感到受欢迎、安全和被包容。
作者：Kiyavash Kandar (Graphics Programmer, Remedy Entertainment)
Alan Wake 2's forests and vegetation of the Pacific Northwest carry the atmosphere of the horror survival genre almost like a character on their own.This talk showcases Remedy's new GPU-based skinning pipeline, built to handle detailed and dense environments that are mostly comprised of dynamically animated vegetation geometry. To match the team's ambitions in geometric fidelity and complexity of motion, they also developed a new GPU-based animation system that could animate skeleton rigs at runtime via compute shaders.The talk provides a look under the hood at the techniques used to hit performance targets, create feature rich systems for artists, and enable full representation in the Northlight engine's raytracing implementation.
《艾伦·韦克2》中的太平洋西北地区的森林和植被承载着恐怖生存体裁的氛围，几乎就像一个角色一样。这次讨论展示了Remedy新开发的基于GPU的蒙皮管线，用于处理主要由动态动画植被几何体组成的详细而密集的环境。为了匹配团队在几何保真度和动作复杂性方面的野心，他们还开发了一套新的基于GPU的动画系统，通过计算着色器在运行时对骨骼装配进行动画化。这次讨论将深入探讨用于达到性能目标、为艺术家建立功能丰富系统，以及在Northlight引擎的光线追踪实现中实现全面表现的技术。
作者：
Havok technology has been integrated with a broad set of proprietary engines. In this talk we want to share our learnings from integrating the Havok SDKs into Unreal Engine. We show how we approached the integrations of the three Havok SDKs - Havok Navigation, Havok Cloth and Havok Physics - unlocking Havok’s industry renowned performance and rich feature set within Unreal. Join this session to learn how to integrate and maintain your own existing tech within Unreal or see how Havok can work for you.
Havok技术已经被整合到了一系列专有引擎中。在本次讲座中，我们想要分享从将Havok SDK整合到虚幻引擎中所学到的经验。我们展示了我们是如何进行三个Havok SDK（Havok Navigation、Havok Cloth和Havok Physics）的整合，解锁Havok在虚幻引擎中享有盛誉的性能和丰富功能。加入本次会议，了解如何在虚幻引擎中整合和维护您自己现有的技术，或者看看Havok如何为您助力。
Machine Learning Summit: 'Naruto Mobile': Optimization for Large-scale Reinforcement Learning in Fighting Games
作者：Elvis Liu (Expert Engineer, Tencent Games)
Naruto Mobile is a popular Fighting Game with over 100 million registered players. AI agents are deployed to the game for various applications such as level challenges. Although deep reinforcement learning is an excellent approach to creating agents with diverse strategies, it is difficult to apply to Naruto Mobile--a game that is built on a pool of 300 characters with unique skills. A traditional approach of self-play training at such a scale may require substantial training costs and time. This talk presents a new training approach to improve agents' generalization ability and optimize massive-scale self-play. The proposed algorithm has already been employed by Naruto Mobile to create agents, which have been used in more than 300 million human-AI fighting matches. To the best of our knowledge to date. This is the first time that deep reinforcement learning has been employed by a commercial fighting game.
《火影忍者手游》是一款受欢迎的格斗游戏，拥有超过1亿注册玩家。AI代理被部署到游戏中，用于各种应用，如关卡挑战。尽管深度强化学习是创建拥有多样化策略的代理的绝佳方法，但很难应用到《火影忍者手游》这样一个以拥有独特技能的300个角色为基础构建的游戏中。在如此规模上进行传统的自我训练方法可能需要大量的训练成本和时间。本讲座提出了一种新的训练方法，以提高代理的泛化能力并优化大规模自我训练。所提出的算法已经被《火影忍者手游》采用，用于创建代理，在超过3亿场人机对战中使用。据我们目前所知，这是深度强化学习首次被商业格斗游戏采用。
作者：Miyake Youichiro (Lead AI Researcher, SQUARE ENIX)
Battle system is one of the most vital parts of SaGa and costly to balance due to the huge volume of the game. We present a machine learning (ML) framework to help us balance the battle system of the latest title of the series under development, SaGa Emerald Beyond, with Reinforcement Learning (RL). We show how our game designers use the AI to help their daily work, the challenges, the detail of our in-house algorithm, and the engineering backend to deliver the system. Also, our system and algorithm are generally applicable to other games and address major key challenges on implementing reliable ML system within general production line. With this system, we can evaluate more than 700 stages in one day and find critical bugs, speeding up the QA process significantly.
战斗系统是《SaGa》系列中最关键的部分之一，由于游戏内容庞大，因此平衡起来是一项耗费巨大的任务。我们提出了一个机器学习（ML）框架，以帮助我们平衡正在开发中的系列最新作《SaGa Emerald Beyond》的战斗系统，采用强化学习（RL）。我们展示了我们的游戏设计师如何利用人工智能来辅助他们的日常工作，面临的挑战，我们内部算法的细节，以及工程后端如何交付该系统。此外，我们的系统和算法通常适用于其他游戏，并解决了在生产线内实施可靠的机器学习系统所面临的主要挑战。有了这个系统，我们可以在一天内评估超过700个阶段，并发现关键错误，极大加快QA过程。
作者：Yijie Tu (Senior Game Engineer, Thunderfire Universe X Studio)
As an important rendering optimization technique, frustum culling is widely used in game development. In recent years, research on the application of neural networks in real-time rendering has been very popular, and we have also applied this idea to frustum culling. Through this speech, game developers learn about our solution of using NN for frustum culling, and the experimental data can also provide developers with a reference.
作为一项重要的渲染优化技术，视锥剔除在游戏开发中被广泛使用。近年来，关于神经网络在实时渲染中的应用研究变得非常流行，我们也将这一思想运用到了视锥剔除中。通过这次演讲，游戏开发者可以了解我们利用神经网络进行视锥剔除的解决方案，并且实验数据也可以为开发者提供参考。
链接：https://schedule.gdconf.com/session/machine-learning-summit-all-you-can-ask-day-1/903336
作者：Olivier Pomarez (Director of development, Ubisoft la Forge)、Antoine Houdard (Research & Development Scientist, Ubisoft)、Georges Nader (Research & Development Scientist, Ubisoft)、Jianrong Tao (CTO of Game Dpt. at GrowEase, NetEase)、Hau Nghiep Phan (Technical Art Director, Electronic Arts)、Yusuke Mori (AI Researcher, SQUARE ENIX CO., LTD.)、Guozhi Xu (AI Team Leader, ByteDance)、Qixuan Zhang (CTO, Deemos Tech)、Yijie Tu (Senior Game Engineer, Thunderfire Universe X Studio)
The GDC Machine Learning Summit has designed this panel session to gather eminent panelists that will be on stage to answer questions from the audience, any question related to Machine Learning in the Video Game Industry. Unlike traditional panels where the moderator asks questions to the panelists, the 60 minutes will be dedicated to the audience. Questions are expected on topics such as Career Management, Working style and Processes, Simulation-to-AI, AI for behavior, AI for Online simulation, AI for Audio or Rendering, AI for Game-as-a-Service features, Generative AI, Build Stability approaches, QA, Game Balancing, Advancements from Academia, AI Developer development environment, and more!This Panel is all yours! All You Can Ask !
GDC 机器学习峰会（GDC Machine Learning Summit）特别设计了这个专题讨论，邀请了一些杰出的专家组成面板，他们将登台回答观众提出的任何与《游戏产业中的机器学习》相关的问题。与传统专题讨论不同的是，这 60 分钟将完全留给观众。预期问题的主题包括职业管理、工作风格和流程、从模拟到人工智能、行为人工智能、在线模拟人工智能、音频或渲染人工智能、作为服务特性的游戏人工智能、生成式人工智能、构建稳定性方法、质量保证、游戏平衡、学术界的进展、人工智能开发者开发环境等。这个专题讨论完全属于你们！你想问什么就问什么！
作者：Chenwei Ding (Senior Technical Engineer, NetEase)
Existing ways to find bugs including client logs, server monitors, and feedbacks from user community. Finding a bug is far from satisfactory. What we need to know is dive into the mechanism and dig out the factors that cause the bug. That is why we need root cause analysis. Existing bug root analysis requires a data science engineer diving into the game logs, analyzing the data of many variables and factors, even playing the game in person to reproduce the bugs. Those efforts could be time-consuming and labor-consuming. We will introduce casual analysis and explain how causal models can help with root cause analysis and improve user experience.
现有的发现错误的方法包括客户端日志、服务器监控以及用户社区的反馈。仅仅发现一个错误是远远不够的。我们需要了解的是深入机制，挖掘导致错误的因素。这就是为什么我们需要根本原因分析。现有的错误根本分析需要一名数据科学工程师深入游戏日志，分析许多变量和因素的数据，甚至亲自玩游戏以重现错误。这些工作可能会耗费大量时间和精力。我们将介绍因果分析，解释因果模型如何帮助根本原因分析并提高用户体验。
Machine Learning Summit: ChatAvatar: Multi-agents Enhanced 3D Character Creation in 'Earth: Revival'
作者：Guozhi Xu (AI Team Leader, ByteDance)、Qixuan Zhang (CTO, Deemos Tech)
LLM has substantially accelerated the pace of game development. Recently, they successfully apply LLMs in the game 'Earth: Revival', for offline production of game NPCs and online creation of player characters.This talk will delve into how multi-agent systems assist in the creation of avatars within 'Earth: Revival'. The most recent LLM of ByteDance will also be introduced. The details of driving agents across multiple tasks will be described. The cooperating agents increased the art team's speed in character creation by 50%.
This multi-agent includes the Dialog Agent, Brain Strom Agents, and Prompt Agent, etc. In addition, they also apply three generative AI modules, including SD for character original painting generation, DreamFace for text/image-to-avatar conversions, and Rodin Search. Furthermore, a text-to-face entrance can be directly experienced by players in 'Earth: Revival.
LLM（Large Language Models）已经大大加快了游戏开发的速度。最近，他们成功地将LLM应用于游戏《地球：复兴》中，用于离线制作游戏NPC和在线创建玩家角色。本次讲话将深入探讨多智能体系统如何协助在《地球：复兴》中创建角色扮演者。还将介绍最新的字节跳动LLM。将描述跨多个任务的驱动智能体的详细内容。协作智能体提高了美术团队在角色创建方面的速度，提升了50％。
这个多智能体包括对话智能体、头脑风暴智能体、提示智能体等。此外，他们还应用了三个生成式人工智能模块，包括用于人物原画生成的SD，用于文本/图像转换成角色头像的DreamFace，以及Rodin Search。此外，玩家还可以在《地球：复兴》中直接体验文本转换成面部的效果。
作者：Yusuke Mori (AI Researcher, SQUARE ENIX CO., LTD.)
In creating interactive digital content including characters, text-to-speech technology is a promising option to flexibly respond to arbitrary user input. We propose a method of applying text-to-speech technology to generate speech in a fictional language that the user has never heard before. We will share with attendee how this method can generate a pronunciation language for a fictional language that fits the worldview and creates an immersive experience, as if the user is listening to the language spoken in that world. We will introduce the simple but effective approach to synthesis fictional language speech, and explain the possible application in games.
在创建包括角色在内的互动数字内容时，文本转语音技术是一种有前景的选项，可以灵活地响应任意用户输入。我们提出了一种将文本转语音技术应用于生成用户从未听过的虚构语言言语的方法。我们将向与会者分享这种方法如何可以为虚构语言生成一个符合世界观、创造沉浸式体验的语音语言，就像用户在那个世界听到该语言被说出来一样。我们将介绍一种简单但有效的合成虚构语言语音的方法，并解释在游戏中的可能应用。
作者：Hau Nghiep Phan (Technical Art Director, Electronic Arts)
In the presentation the speaker discusses how machine learning can improve facial animation in the gaming industry. A showcase on their solution using in-house facial rig, FaceRig, and will demonstrate how blendshapes can be automatically created to suit various character identities. The presentation covers: FaceMixer, which can generate blendshapes from a neutral input, and FaceOptim, which automates the creation of blendshapes and animations from scanned expressions or 4D animation. The presenter also introduces FaceBot, a solution that generates photorealistic face textures and meshes with expressions that can then be fed back into the previous system. By combining FaceBot with FaceOptim or FaceMixer, users can generate high-quality rigged faces for game characters with minimal effort. Future developments include incorporating FaceAnim animation, allowing for more organic and realistic facial animation.
在演讲中，演讲者讨论了机器学习如何提高游戏行业中的面部动画。展示了他们的解决方案使用内部面部绑定系统 FaceRig，并演示如何自动生成适合不同角色身份的融合形状。该演示涵盖了以下内容：FaceMixer，可以从中立输入生成融合形状；FaceOptim，可以自动从扫描表情或4D动画创建融合形状和动画。演讲者还介绍了 FaceBot，这是一个可以生成逼真面部纹理和网格的解决方案，这些表情可以反馈到之前的系统中。通过将FaceBot与FaceOptim或FaceMixer相结合，用户可以轻松生成高质量的游戏角色绑定面部，未来的发展还包括整合FaceAnim动画，从而实现更加有机和逼真的面部动画。
作者：Antoine Houdard (Research & Development Scientist, Ubisoft)、Georges Nader (Research & Development Scientist, Ubisoft)
We present a novel neural material compression technique that is designed to take advantage of hardware accelerated texture compression formats. This approach leans on a multiresolution set of block compressed (BC6H) textures that is subsequently decoded by a small neural network. In practice, our method is integrated into the current pipelines in two stages. At production stage, we use deep learning to jointly learn the texture set and the decoder network. At runtime, we sample the learned textures and then reconstruct the material information by inferring neural network in a shader.
我们提出了一种新颖的神经材质压缩技术，旨在利用硬件加速的纹理压缩格式。这种方法依赖于一个多分辨率的块压缩（BC6H）纹理集，随后通过一个小型神经网络进行解码。在实践中，我们的方法分两个阶段整合到当前的流程中。在生产阶段，我们使用深度学习来联合学习纹理集和解码器网络。在运行时，我们对学习到的纹理进行采样，然后通过在着色器中推断神经网络来重建材质信息。
Machine Learning Summit: Introduction & Training Human-like and High-performance Basketball AI Bot for 'Streetball Allstar'
作者：Jianrong Tao (CTO, NetEase)
"Streetball Allstar" is a 3v3 street basketball game that attracts players with its real-time competition and fairness. AI bots play an important role in the game. Firstly, they assist beginners in learning and improving, providing examples of exceptional skills and teamwork. Secondly, AI bots serve as teammates or opponents, increasing the diversity and competitiveness of the game. Most importantly, AI bots ensure fair competition, fill the gap when there are not enough players and match players of different tiers with highly customized parameters. In terms of training, we combine reinforcement learning and imitation learning, allowing AI bots to learn a wide range of strategies and simulate human thinking and behavior patterns, possessing human-like qualities and human-AI interaction while also having rich strategy diversity. Through dynamic difficulty control, we can control the operation level of the AI accurately and reasonably, adapting to different skill levels. With the client-side deployment solution, AI bots are decoupled from the core logic of game, improving response speed and ensuring data security. In our demonstration, AI bots have shown outstanding performance in various aspects such as difficulty switching, accurate response, strong adaptability, human-AI interaction, and high stability, receiving enthusiastic praise from both players and our customer.
《街头篮球全明星》是一款3v3的街头篮球游戏，以其实时竞技和公平性吸引玩家。在游戏中，AI机器人起着重要作用。首先，它们协助初学者学习和提高，展示出色的技能和团队合作的示例。其次，AI机器人充当队友或对手，增加了游戏的多样性和竞争性。最重要的是，AI机器人确保公平竞争，在玩家数量不足时填补空缺，并通过高度定制的参数将不同等级的玩家进行匹配。在训练方面，我们结合了强化学习和模仿学习，使AI机器人能够学习多种策略，模拟人类的思考和行为模式，具备类人的素质和人机交互，同时拥有丰富的策略多样性。通过动态难度控制，我们可以准确合理地控制AI的操作级别，适应不同的技能水平。通过客户端部署解决方案，AI机器人与游戏的核心逻辑解耦，提高了响应速度并确保数据安全。在我们的演示中，AI机器人在难度切换、准确响应、强大的适应能力、人机交互和高稳定性等各个方面表现出色，受到玩家和我们客户的热情好评。
作者：Jianrong Tao (CTO of Game Dpt. at GrowEase, NetEase)
As a blockbuster FPS game, CSGO enjoys a high reputation in the world, but it suffers from the attacks of cheating players who use aim-bots and wallhack cheating. The talk will share how to use Replay data to detect aim-bots and wallhack cheating, so as to restore a clean gaming environment in CSGO, bringing players a competitive environment that's fair.
作为一款大型的第一人称射击游戏，《反恐精英：全球攻势》在全球享有很高的声誉，但遭受作弊玩家使用瞄准助手和透视作弊的攻击。本次讲座将分享如何利用回放数据来检测瞄准助手和透视作弊，以恢复《反恐精英：全球攻势》中清洁的游戏环境，为玩家带来一个公平的竞技环境。
链接：https://schedule.gdconf.com/session/machine-learning-summit-panel-all-you-can-ask-day-2/903115
作者：Julien Merceron (Worldwide CTO, Bandai Namco Studios)、Sarra Habchi (R&D Scientist, Ubisoft)、Elvis Liu (Expert Engineer, Tencent Games)、Chenwei Ding (Senior Technical Engineer, NetEase)、Mateusz Pilski (Co-Founder and COO, Ice Code Games S.A.)、Jakub Marszałkowski (Co-Founder and Board Member, Indie Games Poland)、Miyake Youichiro (Lead AI Researcher, SQUARE ENIX)
The GDC Machine Learning Summit has designed this panel session to gather eminent panelists that will be on stage to answer questions from the audience, any question related to Machine Learning in the Video Game Industry. Unlike traditional panels where the moderator asks questions to the panelists, the 60 minutes will be dedicated to the audience. Questions are expected on topics such as Career Management, Working style and Processes, Simulation-to-AI, AI for behavior, AI for Online simulation, AI for Audio or Rendering, AI for Game-as-a-Service features, Generative AI, Build Stability approaches, QA, Game Balancing, Advancements from Academia, AI Developer development environment, and more!This Panel is all yours! All You Can Ask!
《The GDC Machine Learning Summit》为了收集知名专家并回答观众关于视频游戏行业中机器学习的任何问题，特别设计了这个专题讨论环节。与传统专题讨论不同，本次60分钟将完全由观众提问。预计观众可能会问及职业管理、工作风格与流程、模拟到人工智能、行为人工智能、在线模拟人工智能、音频或渲染人工智能、以游戏为服务的功能人工智能、生成式人工智能、构建稳定性方法、质量保证、游戏平衡、学术界的进展、人工智能开发者开发环境等话题。这个专题讨论完全属于你们！你们想问什么就问什么！
Machine Learning Summit: Taming the Beast: Leveraging Machine Learning to Curb the Cost of Continuous Integration
作者：Sarra Habchi (R&D Scientist, Ubisoft)
Continuous Integration (CI) is a common practice in modern software organizations. It plays an especially important role for AAA game development, where thousands of build jobs are submitted daily. However, the size and complexity of video games tend to inflate build size and duration, leading to a slower CI and longer feedback time. To provide faster CI feedback, studies have explored optimization techniques that can skip steps, and sometimes the entirety, of build jobs. This talk presents an overview of these techniques and highlights their costs and benefits. Then, it introduces RavenBuild, a novel ML-based technique tailored for video games. Contrarily to existing approaches, RavenBuild predicts build outcomes for both code and data submissions, providing full coverage for a game development CI. RavenBuild skips passing builds and therefore reduces the CI cost, up to 43%, while ensuring that failing builds are properly anticipated.
持续集成（Continuous Integration，CI）是现代软件组织中的常见实践。对于 AAA 游戏开发来说尤为重要，每天都会提交数千个构建任务。然而，游戏的规模和复杂性往往会增加构建的大小和持续时间，导致持续集成速度变慢，反馈时间变长。为了提供更快的持续集成反馈，研究已经探索了一些优化技术，可以跳过构建步骤，有时甚至整个构建任务。本讲话概述了这些技术，并强调它们的成本和收益。然后，介绍了 RavenBuild，这是一种针对视频游戏量身定制的基于机器学习的新技术。与现有方法相反，RavenBuild 针对代码和数据提交都可以预测构建结果，为游戏开发持续集成提供全面覆盖。RavenBuild 跳过成功构建，从而降低了持续集成成本，最高可达 43%，同时确保了失败构建得到适当的预测。
作者：Mateusz Pilski (Co-Founder and COO, Ice Code Games S.A.)、Jakub Marszałkowski (Co-Founder and Board Member, Indie Games Poland)
While we're flooded with generative AI approaches, we intend to show a different possibility to facilitate ML for better and faster gamedev: AI-assisted game design. AI-assisted development is barely emerging, usually as recommendations in programming IDEs. We tried a similar approach to build tools for game designers in three areas: level design, narrative, and game object design (game mechanics related). In this talk, we will show which areas benefited the most, with very promising ML recommendations to ease the designers' work, and what algorithms are showing the most promise despite being really low effort. We will share where, from our experience, ML will not deliver good results anytime soon and how we went from a hardly working system to what seems to be a promising tool.
在我们被各种生成式人工智能方法淹没的同时，我们打算展示一种不同的可能性，以促进更好、更快的游戏开发：AI辅助游戏设计。AI辅助开发刚刚开始出现，通常在编程IDE中作为建议。我们尝试了类似的方法，为游戏设计师在三个领域构建工具：关卡设计、叙事和游戏物体设计（游戏机制相关）。在这次讨论中，我们将展示哪些领域受益最多，具有非常有前景的机器学习建议来简化设计师的工作，以及哪些算法表现最有前景，尽管投入资源很少。我们将分享从我们的经验来看，机器学习不会很快带来好的结果的地方，以及我们是如何从一个几乎无法正常运行的系统发展成一个看起来很有前景的工具的。
作者：Ben Walker (Developer Relations Engineer, Meta)
Whats different about making games for a mobile, immersive headset? Well, a lot. Ben Walker shares his guidance from years of helping developers publish experiences on Meta Quest. From architectural differences like Compositor Layer rendering and CPU/GPU levels, to immersive-only algorithms like Application SpaceWarp, to iteration timesavers like Meta XR Simulator and optimization tools like RenderDoc Meta Fork—learn as much about developing for Meta Quest as 30 minutes will allow.
制作手机、沉浸式头显游戏有什么不同？嗯，有很多。Ben Walker分享了他多年来帮助开发者在Meta Quest上发布体验的指导意见。从诸如Compositor Layer渲染、CPU/GPU级别等架构上的差异，到仅限沉浸式的算法，比如Application SpaceWarp，再到节省迭代时间的工具，比如Meta XR Simulator，以及优化工具，比如RenderDoc Meta Fork——在30分钟内了解尽可能多关于为Meta Quest开发的知识。
作者：Alexis Vaisse (Senior Technical Director, VIRTUOS)、Marios Michaelides (Engineering BU Director, VIRTUOS)
A technical walkthrough on the development of a virtualized geometry technology that can be integrated with any game engine, and an overview of the challenges faced.
The talk explores how Virtuos achieved optimal performance with ultra high-resolution meshes, thanks to a multitude of different techniques - automatic split of mesh LODs, data streaming, occlusion culling at the chunk level, data compression, and more.
Other experimental features, such as geomorphing to enable smooth transition between LODs, will also be discussed.
一项关于虚拟几何技术开发的技术性指南，该技术可以与任何游戏引擎集成，并概述所面临的挑战。
该讲座探讨了Virtuos如何通过多种不同的技术手段实现了超高分辨率网格的最佳性能 - 包括网格LOD的自动分割、数据流式传输、区块级别的遮挡剔除、数据压缩等等。
另外，还将讨论其他的实验性功能，比如几何变形以实现LOD之间的平滑过渡。
链接：https://schedule.gdconf.com/session/mergebot-activisions-merge-and-resolve-infrastructure/901277
作者：Damianos Gkouzkouris (Senior Tools Engineer, Activision)
Source control is one of the foundational elements in a software's lifecycle. Unfortunately, not all source control systems handle code merges and conflict resolution gracefully, particularly in a game development environment with numerous large files scattered across branches that must be kept in sync. This talk serves as a postmortem on a tool that emerged from our ongoing struggle to smoothly integrate Perforce changelists between streams, all while ensuring it doesn't disrupt our continuous development pipeline. The presentation will delve into the architecture of the tool, the challenges encountered while making it as hands-off as possible, the complexities involved in the file resolution process, which needed to be universally accessible, and a discussion of the compromises we had to make while developing and customizing the tool to suit our needs.
源代码控制是软件生命周期中的基础元素之一。不幸的是，并非所有的源代码控制系统都能优雅地处理代码合并和冲突解决，特别是在一个游戏开发环境中，存在大量分散在各个分支上的大型文件，这些文件必须保持同步。这次讨论将作为一项事后检讨，讨论我们如何在确保不影响持续开发流程的同时，如何顺利集成Perforce变更列表的工具。演示将深入探讨该工具的架构，制作过程中遇到的挑战，包括尽可能使其自动化的复杂性，以及用于文件解决过程的普遍可访问性，以及我们在开发和定制工具以满足需求时不得不做的讨论。
链接：https://schedule.gdconf.com/session/mesh-shaders-in-amd-rdna-3-architecture-presented-by-amd/903035
作者：Lou Kramer (Member of Technical Staff, AMD)、Max Oberberger (Senior Software Engineer, AMD)
This talk describes the mesh shader pipeline and how it maps to the AMD RDNA™ 3 architecture. It compares it with the traditional vertex and pixel shader pipeline and the compute pipeline highlighting similarities and differences from both an implementation perspective and a performance perspective. The performance aspects of mesh shaders will be discussed and how one can utilize the hardware efficiently when using mesh shaders.
The talk will also focus on the topic of meshlet generation and meshlet compression and how to optimize assets for mesh shaders. The audience should be able to determine whether using mesh shaders will benefit their use case or if the traditional vertex pipeline is preferable. This discussion will also explain how to use tools to analyze and debug mesh shaders and will explore a practical example of where mesh shaders might be useful.
这次讲座描述了网格着色器流水线以及它如何映射到 AMD RDNA™ 3 架构。它将其与传统的顶点和像素着色器流水线以及计算流水线进行比较，从实现和性能两个角度突出了它们的相似之处和不同之处。将讨论网格着色器的性能方面，以及如何在使用网格着色器时有效利用硬件。
此次讲座还将聚焦于网格块生成和网格块压缩的主题，以及如何为网格着色器优化资产。听众应该可以确定使用网格着色器是否有利于他们的用例，还是传统的顶点流水线更可取。本讨论还将解释如何使用工具来分析和调试网格着色器，并会探讨一个实际示例，演示网格着色器可能会有用的场景。
Meta Developer Summit: Build Next Generation XR Experiences with Meta Quest and Presence Platform (Presented by Meta)
作者：Rangaprabhu Parthasarathy (Director, Product Management, Meta)、Giray Ozil (Senior Engineering Manager, Meta)、Mike Lamprinos (Product Manager, Meta)
Are you interested in developing your next title in VR or MR for Meta Quest? Join Rangaprabhu Parthasarathy, Mike Lamprinos, and Giray Ozil as they take you from project setup to an XR app in less than one hour. You will learn everything you need to get started building XR on Meta Quest during this crash-course. We will cover Meta's Presence Platform capabilities and SDKs that will let you create truly immersive experiences you cannot create on a flatscreen. We will discuss a wide range of Presence Platform features including hand and body tracking, intuitive interactions, and audio and haptics for a more immersive experience. We will also discuss how you can take everything you've learned to bring your experiences to the real world with MR on Meta Quest 3. This fast-paced session will move from concept understanding to practical implementation. Additional sessions throughout the day will provide even deeper dives into designing for MR, adding body tracking to your game, and creating greater immersion with haptics and interaction.
你对在 Meta Quest 上开发下一个标题的虚拟现实（VR）或混合现实（MR）应用感兴趣吗？加入 Rangaprabhu Parthasarathy、Mike Lamprinos 和 Giray Ozil，他们将在不到一个小时的时间内将你从项目设置带入到一个 XR 应用程序中。在这个速成课程中，你将学会一切你需要开始在 Meta Quest 上构建 XR 应用的知识。我们将涵盖 Meta 的 Presence Platform 能力和 SDK，这将让你创造出在平面屏幕上无法实现的真正沉浸式体验。我们将讨论一系列 Presence Platform 功能，包括手部和身体跟踪、直观互动以及更加沉浸式的音频和触觉体验。我们还将探讨如何将你所学到的内容应用到在 Meta Quest 3 上的混合现实中，将你的体验带入现实世界。这个快节奏的会话将从概念理解转向实际实施。全天还将有更深入的专题讨论，涉及在混合现实中设计、将身体追踪应用于游戏以及通过触觉和互动创造更大沉浸感方面。
Meta Developer Summit: Building for Immersion: Haptics and Interactions with Presence Platform (Presented by Meta)
作者：Andrew Lazdins (Sound & Haptic Designer, Meta RL)、David Nelson (Software Engineer, Meta)
Immersion is key when building for XR. Right out of the box, Meta's Presence Platform provides you with the tools and SDKs you need to immerse your players. Discover how to quickly integrate natural-feeling interactions (with hands and controllers) and add HD haptics to your app with David Nelson and Andrew Lazdins. You can thrill your players with this one-two immersive punch. During this session we will introduce you to the Interaction SDK, its capabilities, how we’ve made it even easier to use by integrating it directly into the Unity editor, and show you how to add interactions to your app. Our haptics discussion will begin with a concise 2-minute overview of the design and integration workflow using our tools, showing how you can quickly create most of your haptics in seconds. We'll also share haptic design guidelines, principles, and processes derived from popular apps like Asgard's Wrath 2, Beat Saber, and Max Mustard, so you can enhance those 'special moments' in your app. Haptics We'll kick off with a concise 2-minute introduction to the haptic design and integration workflow using meta tools, demoing how you can quickly create the majority of your haptics in seconds. We will then deep dive into examples and lessons drawn from popular apps like Asgards Wrath 2, Beat Saber, First Hand and First Encounters as we teach you how to enhance those 'special moments' in your app by sharing a new set of haptic design guidelines, principles, and processes.
投入感在构建 XR 时至关重要。Meta 的 Presence 平台提供了您需要让玩家沉浸其中的工具和 SDK。发现如何快速整合自然的手部和控制器交互，以及如何在您的应用中添加高清触觉，David Nelson 和 Andrew Lazdins 将会带领您。您可以通过这种沉浸感十足的方法激动您的玩家。在这个会议中，我们将为您介绍交互 SDK，它的功能，通过将其直接集成到 Unity 编辑器中，我们如何使其使用起来更加轻松，并向您展示如何向您的应用添加交互。我们的触觉讨论将从我们的工具的设计和集成工作流程的简洁2分钟概述开始，演示如何在几秒钟内快速创建您的大部分触觉。我们还将分享触觉设计准则，原则和流程，这些是从像《阿斯加德之怒2》、《Beat Saber》和《Max Mustard》等热门应用中推导出的，这样您就可以增强您的应用中那些‘特别时刻’。触觉 我们将以简明扼要的方式开始，介绍 Meta 工具的触觉设计和集成工作流程，演示如何在几秒钟内快速创建大部分触觉。接着，我们将深入探讨从热门应用中如《阿斯加德之怒2》、《Beat Saber》、《First Hand》和《First Encounters》中汲取的例子和经验教训，教您如何通过分享一系列新的触觉设计准则、原则和流程来增强您的应用中那些‘特别时刻’。
作者：Kirk Barker (Product Manager, Meta)
Using body tracking to animate a character in game play presents some unique challenges. What happens when your character is small compared to the person playing the game? What happens if the proportions are different like having a large bobble head? How do you trade off positional accuracy for hands or head while maintaining the character's look and feel?
In this session, Kirk Barker will teach you how to incorporate body tracking using Meta's Presence Platform Movement SDK with Meta Quest HMDs to animate characters with natural movement for action, fitness, or social. Learn how to enable body tracking and overcome common problems like character scaling, allowing for missing bones, and maintaining positional accuracy of hands, head, or hips desired for the gameplay.
在游戏中使用身体追踪来为角色添加动画呈现了一些独特的挑战。当你的角色与玩游戏的人相比较小时会发生什么？如果比例不同，比如有一个巨大的大脑袋又会怎样？在保持角色外观和感觉的同时，如何在手部或头部的位置精确度上进行权衡？
在这个讲座中，Kirk Barker将教你如何使用Meta的Presence Platform Movement SDK和Meta Quest HMDs来实现身体追踪，为动作、健身或社交类游戏中的角色添加自然的动作。学习如何启用身体追踪，克服角色缩放、缺失骨骼以及保持对游戏玩法所需的手部、头部或臀部位置精确度等常见问题。
作者：David Borel (Engineering Manager, Developer Technology, Meta)、Alexandre Thivierge (Software Engineer, Developer Technology, Meta)
Meta Quest has now reached tens of millions of users and made over a billion dollars, but the technical challenges remain immense, with a wearable form factor and a power-constrained mobile SoC. David gives an overview of the Quest build and optimize flow, geared towards console and PC developers, with an eye toward graphically-demanding, immersive games. Mixed Reality enables new creative possibilities, but it puts unique demands on programmers, artists, and designers. We share the lessons of a ten-year journey from developer kits to Quest 3.
Meta Quest现已吸引了数千万用户，创造了超过十亿美元的收入，但技术挑战仍然巨大，因为其采用了可穿戴外形设计和功耗受限的移动SoC。David概述了Quest的构建和优化流程，针对主机和PC开发者，专注于图形要求高、沉浸感强的游戏。混合现实为创造新的可能性带来了机会，但对程序员、艺术家和设计师提出了独特的要求。我们分享了从开发者套件到Quest 3的十年历程中所学到的经验。
作者：Francisco Rius (Head of Data at Mojang Studios)
This talk explores the vast and diverse world of Minecraft, focusing on the role of data-driven insights in improving player experiences. We’ll delve into lessons from player pathing, segmentation, social listening, and discuss how these insights are applied to different areas of Minecraft. We’ll demonstrate how effective prediction and recommendations can seamlessly integrate into the player experience to deepen player relationships.
本次讲座探讨了广阔而多样化的《我的世界》（Minecraft）世界，重点关注数据驱动洞察在改善玩家体验中的作用。我们将深入研究从玩家路径、细分、社交监听中得出的经验教训，并讨论这些见解如何应用于《我的世界》的不同领域。我们将演示有效的预测和推荐如何无缝地融入玩家体验，以加强玩家关系。
作者：Yuchen Liao (Senior Engine Programmer, TiMi Studio Group, Tencent Games)、Songnan Li (Deputy Director, Machine Learning Team , TiMi Studio Group, Tencent Games)
This presentation will introduce to you a character animation system using machine learning, dubbed MotorNerve, which implements two core character animation functions: locomotion control and transition animation generation. First, by combining Motion Matching and Learned Motion Matching technologies, MotorNerve achieves high-quality, low-consumption locomotion control while speeding up the tuning process for Learned Motion Matching. Secondly, MotorNerve also implements an inhouse Motion In-Betweening (MIB) algorithm, which uses a variational autoencoder to encode leg movements, significantly improving the foot skating problem of existing methods. This method, a joint work by Zhejiang University and TiMi Studio Group, a Tencent Games studio, has been published in SIGGRAPH 2022. MotorNerve applies MIB to interactive animation scenarios, generating high-quality transition animations while saving the project costs.
这份演示将向您介绍一种使用机器学习的角色动画系统，名为MotorNerve，它实现了两个核心角色动画功能：移动控制和过渡动画生成。首先，通过结合运动匹配和学习运动匹配技术，MotorNerve实现了高质量、低消耗的移动控制，同时加速了学习运动匹配的调整过程。其次，MotorNerve还实现了一种内部运动插入（MIB）算法，该算法使用变分自动编码器来编码腿部运动，显著改善了现有方法中存在的脚滑问题。这种方法是由浙江大学和腾讯游戏工作室TiMi Studio Group联合进行的成果，在SIGGRAPH 2022上发表。MotorNerve将MIB应用于交互式动画场景，生成高质量的过渡动画的同时节约项目成本。
链接：https://schedule.gdconf.com/session/nvidia-ace-anatomy-of-a-smart-npc-presented-by-nvidia/903306
作者：Jaakko Haapasalo (Production Director, NVIDIA)
NVIDIA ACE is a suite of technologies for creating Smart NPCs in games. ACE offers a flexible set of building blocks that tools and middleware vendors are using to assemble end-to-end NPC pipelines. Join us for a look behind the scenes at our latest demo, and see how these technologies can be combined via middleware to create a state of the art character pipeline.
NVIDIA ACE 是一套用于在游戏中创建智能非玩家角色（Smart NPCs）的技术套件。ACE 提供了一组灵活的构建模块，工具和中间件供应商正在使用这些模块来组装端到端的 NPC 管线。加入我们，一起深入了解我们最新演示背后的技术，并看看这些技术如何通过中间件结合在一起，创建一个先进的角色管线。
链接：https://schedule.gdconf.com/session/nanite-gpu-driven-materials/899486
作者：Graham Wihlidal (Engineering Fellow (Graphics), Epic Games)
This advanced technical talk takes a deep dive into Nanite's novel GPU driven material pipeline.
A high-level overview is presented, detailing all of the core components and explaining how they fit into the overall design.
This talk also covers low-level technical details, including how artist-authored materials drive the rasterizer, how final shading is evaluated, an innovative variable rate shading technique, and some challenges of GPU driven rendering.
Additionally, this talk discusses several learnings and optimizations that were discovered during the development of these components.
这次高级技术讲座将深入探讨Nanite独创的GPU驱动材质渲染管线。
演讲将以高层次概述的形式呈现，详细解释所有核心组件，并说明它们如何融入整体设计。
此次讲座还涉及低级技术细节，包括艺术家创作的材质如何驱动光栅化器、最终着色是如何评估的、一种创新的可变速率着色技术以及GPU驱动渲染的一些挑战。
此外，该讲座还讨论了在开发这些组件过程中发现的几种学习和优化方法。
作者：Junhong Wang (Lead Mobile Programmer of Arena Breakout, Morefun Studios, Tencent Games)
"Arena Breakout" is a next-gen immersive tactical FPS powered by Unreal Engine 4.26 on mobile devices.The game integrates Vulkan Ray Query-based reflection, ambient occlusion and soft shadow, all optimized to run seamlessly on the latest Android platform. In this session, Junhong Wang, Lead Mobile Programmer of Arena Breakout, will introduce the implementation and optimization of the Vulkan Ray Query in 'Arena Breakout', including its ability to maintain a stunning 60 FPS within the expansive open world map.
《竞技场突围（Arena Breakout）》是一款由虚幻引擎4.26打造的下一代沉浸式战术FPS游戏，适用于移动设备。该游戏整合了基于Vulkan Ray Query的反射、环境光遮蔽和软阴影，经过优化使其能够在最新的Android平台上顺畅运行。在本次分享中，《竞技场突围》的首席移动程序员王军红（Junhong Wang）将介绍Vulkan Ray Query在游戏中的实现和优化，包括其在广阔的开放世界地图中保持惊人的60FPS的能力。
作者：Gus Martins (Developer Advocate, Google)、Dan Zaratsian (AI/ML Solutions Architect, Google)、Lei Zhang (Director, Play Partnerships, Global GenAI & Greater China BD, Google)、Jack Buser (Director, Game Industry Solutions)、Simon Tokumine (Director of Product Management, Google AI)
Join us for a keynote from Google's leading AI minds as they delve into the transformative applications of Generative AI in the gaming landscape. This session brings together senior executives from Google Cloud, Google Play, and Labs, offering unique perspectives from across the gaming ecosystem, development and business management lifecycle.This 1 hour session includes:- A Fireside Chat with Google's Cloud, Play, and Labs AI experts- Open Source AI development tool overview for games including Keras, TensorFlow, JAX, and more- Introduction of Google Cloud Gen AI services supporting scalable architecture and quick-to-deploy design patterns for games
加入我们，参加由谷歌领先的人工智能专家主持的主题演讲，他们将深入探讨生成式人工智能在游戏行业中的变革应用。本次会议汇聚了谷歌云、谷歌Play和实验室的高级管理人员，为游戏生态系统、开发和业务管理生命周期提供独特的视角。本1小时的会议包括：- 与谷歌云、Play和实验室AI专家进行的炉边聊天- 为游戏提供的开源人工智能开发工具概述，包括Keras、TensorFlow、JAX等- 介绍支持游戏的可扩展架构和快速部署设计模式的谷歌云Gen AI服务
链接：https://schedule.gdconf.com/session/obstacle-traversal-in-the-organic-world-of-pandora/899509
作者：Joel Nilsson (Senior Gameplay Programmer, Massive Entertainment - A Ubisoft Studio)
Stepups and mantles are an established part of the game character's repertoire, yet with increasingly complex game worlds traditional markup solutions have become increasingly cumbersome. In Avatar: Frontiers of Pandora we wanted to portray the full dense, organic beauty of Pandora while making the motion feel as flowing and unobstructed as you would expect from a Na'vi. In order to deal with these challenges we created a highly generic and (almost) markup-free system for stepups and mantling relying on a minimum of assumptions while trying to maximize freedom of movement. This talk presents the basic philosophy, concepts, and implementation of this solution. The focus is on providing a solid starting point for those just starting out with this problem and some useful tools for anyone working with it.
台阶（Stepups）和爬台（mantles）已经成为游戏角色技能库中的一个重要组成部分，然而随着游戏世界变得越来越复杂，传统的标记解决方案变得越来越繁琐。在'阿凡达：潘多拉前线'中，我们希望展现出潘多拉充满密集有机美感的同时，使玩家的动作感觉像纳美族一样流畅且不受阻碍。为了应对这些挑战，我们创建了一个高度通用且（几乎）无需标记的台阶和爬台系统，减少了假设的数量，同时努力最大化移动自由度。本讲演介绍了这一解决方案的基本理念、概念和实施方法。重点是为那些刚刚开始处理这个问题的人提供一个扎实的起点，以及为任何正在处理此问题的人提供一些有用的工具。
链接：https://schedule.gdconf.com/session/open-world-rendering-techniques-in-hogwarts-legacy/902654
作者：Rob Hall (Advanced Software Engineer, Avalanche Software)
Rendering a game as complex as Hogwarts Legacy required a focus on adapting the Unreal engine for an open world game. This included handling the massive scale of the Hogwarts castle and scaling rendering systems for a seamless indoor and outdoor experience. This talk will expound on the technical aspects of this work. Topics will include how to render with hundreds of shadow casting light sources, rendering multi-layered cloth at a reduced cost, and how to batch light probes, reflections, and foliage. Hogwarts Legacy also contains several other techniques that make it an immersive experience. This will involve topics like Dynamic time of day, camera exposure, seasons, and weather. Lastly, we will discuss how to render animated paintings without the typical way of rendering to a render target.
将像《霍格沃茨遗产》这样复杂的游戏呈现出来，需要专注于将虚幻引擎(Unreal Engine)改编成一款开放世界游戏。这包括处理霍格沃茨城堡的庞大规模，以及调整渲染系统，以实现室内和室外体验的无缝衔接。本次讨论将深入探讨这项工作的技术方面。主题将包括如何使用数百个投射阴影的光源进行渲染，以较低成本渲染多层布料，以及如何批处理光探针、反射和植被。《霍格沃茨遗产》还采用了其他几种技术，使其成为一种沉浸式体验。这将涉及到动态日夜更替、相机曝光、季节和天气等主题。最后，我们将讨论如何在不使用传统的渲染目标的情况下渲染动画画作。
链接：https://schedule.gdconf.com/session/pondering-orbs-the-rendering-and-art-tools-of-cocoon/902878
作者：Mikkel Svendsen (Render Programmer, Geometric Interactive)
This talk details the graphical techniques developed for COCOON, the debut title of Geometric Interactive, founded by Jeppe Carlsen, lead gameplay designer on LIMBO and INSIDE. Rendering the game's beautiful worlds within worlds takes a few tricks—from thick billowing froxel fog and spherical harmonics volumetric lightmaps, to a host of visual effects like crystalline Voronoi bridges, SDF shoreline ripples on water, and warping through ponderable orbs.Mikkel Svendsen presents how these features, tools and effects were implemented and shipped on the game's target platforms, how they contributed to the game's look, and how it all works together with an MSAA-friendly rendering pipeline by carefully considering the gotchas of fitting these features and effects with that.
这个讲座详细介绍了为 Geometric Interactive 的首款游戏《COCOON》开发的图形技术，该公司由《LIMBO》和《INSIDE》的首席玩法设计师 Jeppe Carlsen 创立。在游戏的世界内呈现美丽的世界需要一些技巧——从厚重的滚滚的 froxel 雾和球谐体积光照图，到一系列视觉效果，如晶格 Voronoi 桥、SDF 水面波纹和通过感知光球进行扭曲等。Mikkel Svendsen 展示了这些功能、工具和效果是如何实现并在游戏目标平台上发布的，它们如何为游戏的外观做出贡献，以及如何通过谨慎考虑将这些功能和效果与 MSAA 友好的渲染管线配合，使一切协同起来。
作者：Adam Sawicki (Principal Member of Technical Staff, AMD)、Amit Mulay (Senior Software Development Engineer, AMD)、Marco Bouterse (Principal Graphics Programmer, Nixxes Software)
The AMD Radeon GPU Detective (RGD) is AMD's powerful new developer tool for post-mortem analysis of GPU crashes on Windows (TDRs). In this session you will learn about these types of failures - when they occur, why they are difficult to debug, and how RGD can help to quickly narrow down the search for a crash's root cause.
You will learn how to capture and analyze AMD GPU crash dumps from crashing games, and how to interpret the information in the Crash Analysis reports that the tool generates. You will also hear a story about a specific crash in a game that the tool helped to investigate and fix.
AMD Radeon GPU Detective（RGD）是AMD强大的新开发者工具，用于Windows上GPU崩溃的事后分析（TDRs）。在本次课程中，您将了解这些故障类型 - 它们发生的时间，为什么它们难以调试，以及RGD如何帮助快速缩小崩溃根本原因搜索范围。
您将学习如何捕获和分析从崩溃游戏中获得的AMD GPU崩溃转储，并如何解释工具生成的崩溃分析报告中的信息。您还将听到有关工具帮助调查和修复特定游戏中崩溃的故事。
RT: Overdrive in Cyberpunk 2077 Ultimate Edition - Pushing Path Tracing One Step Further (Presented by NVIDIA)
作者：Pawel Kozlowski (Principal Developer Technology Engineer , NVIDIA)、Jakub Knapik (VP of Art, Global Art Director, CD Projekt RED)、Michael Murphy (Senior Developer Technology Engineer, NVIDIA)
Ray Tracing: Overdrive mode of Cyberpunk 2077, released as a Technology Preview back in April 2023, revamped most of the game’s lighting pipeline with path tracing. Ever since then, the team has been working on addressing the most prominent issues of this new technology. Join our session to find out how ReSTIR GI, radiance caching, and DLSS Ray Reconstruction helped us improve image quality at near-zero performance cost and how they became the key components behind removing the Technology Preview label in the Ultimate Edition of the game that includes the Phantom Liberty expansion.
射线追踪：在2023年四月作为技术预览发布的《赛博朋克2077》超频模式，通过路径追踪优化了大部分游戏的光照流程。自那时起，团队一直在努力解决这项新技术中最突出的问题。加入我们的讨论，了解ReSTIR GI、辐射缓存和DLSS射线重建如何帮助我们在几乎没有性能消耗的情况下改善了图像质量，以及它们如何成为删除“技术预览”标签的关键组件，在包括“幻影自由”扩展的游戏终极版中。
作者：Oleksandr Koshlo (Render Architect, Massive Entertainment - A Ubisoft Studio)
How do you achieve high fidelity lighting with ray tracing, while maintaining high performance target in an expansive world such as the world of Avatar: Frontiers of Pandora? It requires a good implementation of ray tracing pipeline at multiple levels. This session describes BVH and material management part. The talk shows in detail two different acceleration structure implementations: a custom one used on consoles, and a DXR-based one for PC platforms. It highlights advantages and tradeoffs available with a custom approach, and tackles both common and unique challenges of each implementation, and a range of possible optimizations. The second part talks about "average" material representation, and how Snowdrop handles calculating the average for artist-authored materials, often dependent on runtime inputs.
如何在像《阿凡达：潘多拉前线》这样的开放世界中实现高保真光线追踪效果，并同时保持高性能目标？这需要在多个层面上良好地实现光线追踪渲染管线。本讨论主要涉及 BVH（Bounding Volume Hierarchy，包围体层次树）和材质管理部分。具体介绍了两种不同的加速结构实现方案：一种是用于游戏主机的自定义方案，另一种是针对PC平台的基于DXR（DirectX Raytracing）的方案。文章突出了自定义方案的优势和权衡，并探讨了每种实现方案所面临的常见和独特挑战，以及一系列可能的优化方式。第二部分讨论了“average（平均）”材质表示法，以及Snowdrop引擎如何处理为艺术家制作的材质计算平均值的过程，这些计算通常依赖于实时输入。
作者：Varun Khare (Chief Executive Officer, NimbleEdge)
Personalization is well-known to improve gamer engagement significantly. However, most developers rely on outdated, historical user data for personalization, missing out on rich insights regarding session context from real-time user-game interactions. This is largely due to a) prohibitively high cloud costs of handling and processing real-time session data on cloud and b) difficulty of deploying and maintaining real-time ML models on-device
We propose a novel approach to overcome these challenges, using our managed on-device ML platform. The platform executes data processing, inference as well as training on-device to deliver cost-efficient, real-time personalization at scale. This results in significant improvement in gamer experience and conversion, with minimal end-to-end latency and CPU and battery usage spike
个性化已被广泛认为能显著提升玩家参与度。然而，大多数开发者仍然依赖过时的、历史性的用户数据进行个性化，错过了实时用户与游戏交互中关于游戏场景的丰富见解。主要原因在于a)在云端处理和处理实时会话数据的云成本过高，b)难以在设备上部署和维护实时机器学习模型。
我们提出了一种新颖的方法来克服这些挑战，利用我们的托管设备上的机器学习平台。该平台在设备上执行数据处理、推理以及训练，以实现规模化的成本高效、实时个性化。这将显著提升玩家体验与转化，同时端到端延迟、CPU以及电池使用的增加将被最小化。
链接：https://schedule.gdconf.com/session/real-time-world-editing-technology-in-northlight/899582
作者：Sebastian Eriksson (Lead Core Engine Programmer, Remedy Entertainment)
Requirements for game worlds are ever-increasing, being more realistic, more interactive, and larger than ever before. This means that there is a ton of content to be authored, but also large amounts of data to pass through pipelines and workflows to the engine runtime. Remedy has been working on a new data pipeline and technology stack for the Northlight engine that allows us to optimize entity data for authoring at edit time, and performance at runtime, while keeping near-instant iteration times for world editing workflows. The fundamental parts of this technology are OpenUSD for authoring, and Northlight's transformative live-edit pipeline that enables entity modifications to be applied to a running game process on the fly. In this presentation, Sebastian shares how we keep the editor and runtime data representations separated, how the data is transformed, and how the real-time editing between the editor and runtime works in practice.
游戏世界的需求在不断增加，变得比以往更加真实、互动性更强，规模也比以往大得多。这意味着需要创作大量内容，同时也需要通过管道和工作流将大量数据传递到引擎运行时。Remedy一直在为《北光引擎》开发新的数据管道和技术堆栈，这使得我们能够在编辑时优化实体数据，同时在运行时提高性能，同时保持编辑世界工作流程的几乎即时迭代时间。这项技术的基本部分是使用 OpenUSD 进行创作，并使用北光引擎的转换性实时编辑管道，使得实体修改能够实时应用到运行中的游戏进程中。在这个演讲中，Sebastian 分享了我们如何保持编辑器和运行时数据表示分离，数据是如何转换的，以及编辑器和运行时之间的实时编辑是如何实际运作的。
作者：Adam Noonchester (Associate Game Director, Insomniac Games)
Boss fights with Spider-Man's rogues gallery are a key part of the fantasy, and delivering on them is a key component to the Marvel’s Spider-Man series. For the team, improving our boss-fights for Marvel’s Spider-Man 2 meant leaning into core combat. This required making a lot of fundamental changes to how we designed and implemented bosses. This talk reveals design, programming and production lessons we learned as we went through this process. The attendees learn how we made unique, epic and memorable fights that achieved spectacle through core combat instead of via one-off, lock-and-key mechanics and disabling player abilities.
蜘蛛侠的反派大Boss对战是游戏中的重要组成部分，成功呈现这些对决是《漫威蜘蛛侠》系列的关键要素之一。为了给《漫威蜘蛛侠2》改进我们的Boss战，团队不得不深入探讨核心战斗。这需要我们对设计和实现Boss进行大量基础性改变。本次演讲将揭示我们在这一过程中所学到的设计、编程和制作经验。与会者将了解我们如何设计独特、史诗和难忘的对决，通过核心战斗来实现壮观，而非通过一次性机制和禁用玩家能力来达成。
作者：Bobby Wilkinson (Technical Director, Respawn Entertainment)
The Jedi series of games presents a number of challenges in managing persistent global state due to its mix of a non-linear Metroidvania structure and a strong central story. Tracking information such as if a shortcut is unlocked, a line of dialogue has played, or a boss has been beaten are all vital. All of it is interconnected with different dependencies, but at the same time there is no single correct order for all of these things to occur. This talk will start with a brief overview of the non-structured approach of how STAR WARS Jedi: Fallen Order™ managed this information, discuss the lessons learned from that approach's failures, then spend the majority of its time discussing the new centralized system we developed for STAR WARS Jedi: Survivor™ to address those failings.
《绝地武士》系列游戏以非线性的Metroidvania结构和紧凑的主线故事为特色，管理持久全局状态面临许多挑战。跟踪诸如快捷方式是否解锁、是否播放了某行对话或是否打败了某个boss等信息都至关重要。所有这些都彼此相关，具有不同的依赖关系，但同时不同事情发生的顺序也没有单一正确的方法。本演讲将从简要概述《星球大战 绝地：陨落秩序™》是如何管理这些信息的非结构化方法开始，讨论从该方法失败中学到的教训，然后重点讨论我们为《星球大战 绝地：幸存者™》开发的新集中式系统，旨在解决这些问题。
链接：https://schedule.gdconf.com/session/scaling-git-workflows-for-game-development/900038
作者：Vikram Saran (Senior Lead Engineer, Independent)
Game development studios often face a dilemma when choosing version control systems—specific requirements like exclusive file locking can limit options, while common solutions remain financially out of reach for up-and-coming studios. Surprisingly, even with widely used solutions, many projects struggle to adhere to the 'not rocket science' rule of software engineering: maintaining a code repository that consistently passes all tests. Join Vikram in this session as he delves into the challenges of version control in game development and explores how modern git tooling, coupled with strategic workflow adjustments, can pave the way for a cost-effective solution. Discover practical insights to start utilising modern DevOps practices with your projects.
游戏开发工作室在选择版本控制系统时常常面临两难选择——特定要求比如独占文件锁定可以限制选项，而常见解决方案对新兴工作室来说在经济上并不可及。令人惊讶的是，即使使用了广泛的解决方案，许多项目仍然难以遵循软件工程中的“并非火箭科学”规则：维护一个始终通过所有测试的代码存储库。加入Vikram在本次会议，他将深入探讨游戏开发中版本控制的挑战，并探讨现代的git工具以及与之配合的策略性工作流调整如何为成本效益的解决方案铺平道路。发现开始利用现代DevOps实践来开展项目的实用见解。
作者：Theresa Foley (Senior Research Scientist, NVIDIA)、Ivan Fedorov (Developer Technology Engineer, NVIDIA)
The Slang shading language is compatible with existing HLSL codebases, adding extensions needed for building maintainable and high-performance real-time rendering codebases, along with deeply integrated support for automatic differentiation to unblock innovation in AI-accelerated rendering, inverse rendering, and more.
This session will cover our experiences integrating Slang into multiple renderers, give advice for developers who want to try Slang out, and share exciting new applications made possible by automatic differentiation.
‘The Slang着色语言兼容现有的HLSL代码库，添加了构建可维护和高性能实时渲染代码库所需的扩展，同时深度集成了对自动微分的支持，以推动AI加速渲染、反向渲染等创新。
本次讲座将介绍我们整合Slang到多个渲染器的经验，为那些想尝试Slang的开发者提供建议，并分享由自动微分带来的令人兴奋的新应用。’
作者：Yusuf Duman (Software Engineer, Arm)
Discover the cutting-edge advancements in Unity Sentis as we discuss the support for INT8 quantized networks. Dive into a dynamic showcase revealing real-time depth map generation from live camera feeds, demonstrating the transformative power of quantized nets.
Uncover the game-changing advantages, including increased performance, optimized memory, and reduced bandwidth usage – vital elements for resource-constrained environments such as mobile. Empower game developers to unleash potent neural networks on mobile devices, crafting immersive and compelling experiences that redefine the boundaries of gaming.
在Unity Sentis中探索最前沿的进展，我们将讨论对INT8量化网络的支持。深入展示实时从摄像头拍摄中生成深度图的动态展示，展示量化网络变革力量的示范。
揭示这一变革性优势，包括提高性能、优化内存和减少带宽使用等方面 - 这些对于诸如移动设备等资源受限环境至关重要。赋予游戏开发者在移动设备上释放强大的神经网络的力量，打造沉浸式和引人入胜的体验，重新定义游戏的边界。
作者：Edwin Jones (Senior Software Engineer, Epic Games)
At launch, the player population of “Fall Guys” peaked at 650,000 concurrent users, which was 8-times the projected load. In this session, walk through how the game services team at Mediatonic prepared for launch, the challenges encountered with this sudden influx of users, and describe improvements made over the last few years to scale to millions of players.
在游戏《Fall Guys》发布时，玩家人数达到了65万同时在线用户的峰值，是预期负荷的8倍。在这个讨论中，我们将一起了解Mediatonic的游戏服务团队是如何为发布做准备的，他们在面对突如其来的用户涌入时遇到的挑战，并描述近几年为承载数百万玩家而做出的改进。
作者：Sun Jeong (Machine Learning Engineer, LoadComplete)
Playtesting is a crucial aspect of modern game development. However, small-to-mid-sized teams often face difficulties in meeting their testing demands due to limited QA resources.
In this talk, Sun Jeong discusses how he and his team built and harnessed a machine-learning-assisted automated playtesting solution to fulfill the testing requirements for the launch and live service of an action-packed roguelite game, Frame Arms Girl: Dream Stadium.
He talks about the technical details of the automated playtesting stack, introduces various detailed use cases and the workflow used to debug and discover valuable design insights, and shares the tips and lessons learned during the development of the stack under dynamic and unstable production settings.
游戏测试（Playtesting）是现代游戏开发的一个关键方面。然而，小至中型团队常常面临测试需求的难题，因为质量保证（Quality Assurance，QA）资源有限。
在这场讲座中，Sun Jeong讨论了他和他的团队如何建立和利用基于机器学习的自动游戏测试解决方案，以满足动作充沛的roguelite游戏《Frame Arms Girl: Dream Stadium》的上线和运营服务所需的测试要求。
他讲述了自动游戏测试堆栈的技术细节，介绍了各种详细的使用案例以及用于调试和发现有价值的设计见解的工作流程，并分享了在动态和不稳定的生产环境下开发这一堆栈时学到的建议和教训。
链接：https://schedule.gdconf.com/session/the-evolution-of-matchmaking-in-apex-legends/899644
作者：Christopher Pierse (Principal Data Scientist, Electronic Arts)、Samy Duc (Technical Director, Respawn Entertainment)
Apex Legends launched four years ago with no skill-based matchmaking (SBMM). While players often claim that the best SBMM is no SBMM, significant evidence proving otherwise quickly emerged and the team has been iterating on in-house approaches to matchmaking since. This session is a full chronological review of how and why the matchmaking in Apex Legends has evolved since launch. Technical Director Samy Duc and Principal Data Scientist Christopher Pierse will discuss the details of Respawn's tech stack, skill measurement algorithms, and matchmaking algorithms at every step of this journey.
《Apex Legends》游戏在四年前推出时没有技能匹配系统（Skill-Based Matchmaking，SBMM）。虽然玩家们经常声称最好的SBMM就是没有SBMM，但很快就出现了大量证据证明相反，因此团队一直在不断尝试内部匹配方法。本场讨论将全面回顾自推出以来《Apex Legends》中的匹配系统是如何演变的，以及为何会有这样的变化。技术总监Samy Duc和首席数据科学家Christopher Pierse将讨论Respawn的技术架构、技能评估算法以及匹配算法在这一过程的每一个阶段的细节。
作者：David Block (Sr. Engine Programmer, CD Projekt RED)
This talk covers the evolution of REDengine 4's thread-based job system: its API, internal implementation details, and our overall efforts to achieve greater parallelism. The final design was the result of many iterations and stages of feedback while the game was under development. The final approach chosen resulted in nearly all engine and gameplay systems taking advantage of multiple cores, which helped toward reaching our performance goals. Particular topics include solutions for a thread-based approach using counters for interjob dependencies, debugging and profiling, practical philosophy towards CPU resource sharing, integration with other systems, and overall approaches to the composability of jobs.
这次演讲涵盖了 REDengine 4 的基于线程的作业系统的演进：它的API、内部实现细节，以及我们为实现更大并行性所付出的努力。最终设计是在游戏开发过程中经过多次迭代和反馈阶段的结果。最终选择的方法导致几乎所有引擎和游戏系统都能充分利用多核，有助于实现我们的性能目标。特定主题包括使用计数器处理作业间依赖关系的基于线程的方法，调试和性能分析，CPU 资源共享的实用哲学，与其他系统的集成，以及作业组合性的整体方法。
链接：https://schedule.gdconf.com/session/tools-roundtable-day-1-engineering/899982
作者：Geoff Evans (Principal Engineer, Nvidia)
These roundtable sessions focus on discussion of best practices, novel techniques, and technical trade-offs for building game development tools and services. Game development tools are constantly evolving, so come to share what works (and what doesn't) at your studio, and listen to others' take on how to keep their creative teams producing awesome games.
Wednesday: "Engineering", focuses on tools programming, testing, deployment, frameworks, scripting, automation, extensibility, and telemetry.
Thursday: "Assets" focuses on export/import, interchange, persistence, collaboration, search and organization, development models (branching and/or trunking), revision control, and outsourcing.
Friday: "Build" focuses on processing assets through to the game in all its forms: build systems, caching, distribution, scalability, optimization, disc mastering and patching.
这些圆桌讨论会侧重讨论构建游戏开发工具和服务的最佳实践、新技术和技术权衡。游戏开发工具不断发展，所以来分享在你工作室取得成功的经验（和不成功的经验），听听其他人如何让他们的创意团队持续制作出令人惊叹的游戏。
星期三：“工程”，侧重于工具编程、测试、部署、框架、脚本编写、自动化、可扩展性和遥测。
星期四：“资产”侧重于导入/导出、交换、持久性、协作、搜索和组织、开发模型（分支和/或主干）、修订控制以及外包。
星期五：“构建”侧重于将资产处理到游戏中的各种形式：构建系统、缓存、分发、可伸缩性、优化、光盘制作和打补丁。
链接：https://schedule.gdconf.com/session/tools-roundtable-day-2-assets/903302
作者：Geoff Evans (Principal Engineer, Nvidia)
These roundtable sessions focus on discussion of best practices, novel techniques, and technical trade-offs for building game development tools and services. Game development tools are constantly evolving, so come to share what works (and what doesn't) at your studio, and listen to others' take on how to keep their creative teams producing awesome games."Assets" focuses on export/import, interchange, persistence, collaboration, search and organization, development models (branching and/or trunking), revision control, and outsourcing.
这些圆桌讨论会侧重于讨论构建游戏开发工具和服务的最佳实践、新颖技术以及技术取舍。游戏开发工具不断发展，因此来分享在你的工作室起作用（以及不起作用）的内容，并听听别人对如何让他们的创意团队创作出令人惊叹的游戏的看法。"资源"关注于导出/导入、交换、持久性、合作、搜索和组织、开发模型（分支和/或主干）、版本控制和外包。
链接：https://schedule.gdconf.com/session/tools-roundtable-day-3-build/903303
作者：Geoff Evans (Principal Engineer, Nvidia)
These roundtable sessions focus on discussion of best practices, novel techniques, and technical trade-offs for building game development tools and services. Game development tools are constantly evolving, so come to share what works (and what doesn't) at your studio, and listen to others' take on how to keep their creative teams producing awesome games."Build" focuses on processing assets through to the game in all its forms: build systems, caching, distribution, scalability, optimization, disc mastering and patching.
这些圆桌讨论会专注于讨论构建游戏开发工具和服务的最佳实践、新颖技术以及技术折衷方案。游戏开发工具不断发展变化，因此请分享您工作室的成功经验（以及失败经验），并倾听他人对如何让他们的创意团队打造出优秀游戏的见解。“Build”关注的是将各种形式的资源处理至游戏中的过程：构建系统、缓存、分发、可扩展性、优化、光盘制作和补丁。
作者：Rajiv Gandi (Master Principal Cloud Architect, Oracle)、Yasmina Benkhoui (Generative AI Strategic Partnerships, NVIDIA)、Nathan Yu (Director of Product, Inworld AI)
Explore the possibilities offered by AI NPCs as they become essential catalysts in engaging gameplay. This session will take a look at how developers have successfully used AI NPCs to drive core game loops and mechanics that keep players coming back for more.
探索AI NPCs为游戏提供的可能性，它们正成为引人入胜的游戏关键。本次讲座将探讨开发者如何成功地利用AI NPCs来推动核心游戏循环和机制，从而吸引玩家不断回流。
作者：Takuhiro Dohta (Senior Director, Entertainment Planning & Development Division, Nintendo Co., Ltd.)、Junya Osada (Sound Programmer, Entertainment Planning & Development Division, Nintendo Co., Ltd.)、Takahiro Takayama (Physics Programmer, Entertainment Planning & Development Division, Nintendo Co., Ltd.)
Developers of The Legend of Zelda: Tears of the Kingdom discuss structuring an expanded Hyrule around physics-based gameplay and evolved sound design! Join the game’s Technical Director Takuhiro Dohta, Lead Physics Programmer Takahiro Takayama, and Lead Sound Engineer Junya Osada as they explore challenges their teams faced when approaching this sequel. They will share insights on Link’s new abilities, composing Hyrule’s expanded sound design and music, and how these go hand-in-(Ultra)hand to create a new experience for players. By the end of Mr. Dohta, Mr. Takayama, and Mr. Osada’s talk, they aim to give attendees and players everywhere new perspective on the world of Tears of the Kingdom and the innovations within.
《塞尔达传说：王国之泪》的开发团队讨论围绕基于物理的游戏玩法和进化的音频设计构建扩展版的海拉尔！加入游戏的技术总监土田拓广，主力物理程序员高山隆博，和主力音效工程师长田纯也，他们将探讨团队在处理这部续作时所面临的挑战。他们将分享关于林克（Link）的新技能、创作海拉尔扩展版音效设计和音乐的见解，以及这两者如何携手合作为玩家打造全新体验。在土田先生、高山先生和长田先生的讨论结束时，他们的目标是为与会者和各地玩家带来对《王国之泪》世界和其中的创新的新视角。
链接：https://schedule.gdconf.com/session/uefn-roadmap-presented-by-epic-games/903285
作者：Bjorn Book-Larsson (VP, Creator Platform, Epic Games)
Join us as we share the current state of Unreal Editor for Fortnite (UEFN) and discuss future development plans. This presentation will guide developers and studios through the latest capabilities they can leverage to engage the Fortnite audience. It will include tips on how to get started within the Fortnite Ecosystem, as well as discussing current limitations.
加入我们，分享《堡垒之夜》（Fortnite）的虚幻编辑器（Unreal Editor for Fortnite，UEFN）的当前状态，并讨论未来的开发计划。本次演示将指导开发人员和工作室了解他们可以利用的最新功能，以吸引《堡垒之夜》的受众群体。演示将包括如何在《堡垒之夜》生态系统内开始的提示，以及讨论当前的限制。
链接：https://schedule.gdconf.com/session/uipainter-tile-based-ui-rendering-in-one-draw-call/901128
作者：Wessel Mast (Tools Programmer, Guerrilla B.V.)
Every game editor needs a user interface. It is how we allow developers to create content for the game in a visual way. A good UI renderer has clear-cut constraints, like drawing nicely anti-aliased primitives, yet programmers have vastly different approaches to rendering game editor UI.In this session, Wessel Mast, Tools Programmer at Guerrilla, will tell you about how Guerrilla’s editor ‘Decima’ renders UI: how it uses a tile-based renderer, commonly used in light culling, SSAO, depth of field and more, for drawing complex UI in an efficient manner and how it manages to draw thousands of primitives like rectangles, lines and curves in a single draw call using signed distance fields.
每个游戏编辑器都需要一个用户界面。这是我们让开发人员以可视化方式为游戏创建内容的方法。一个优秀的UI渲染器具有明确的约束条件，比如绘制出漂亮的反赇抗锯齿图形，尽管程序员在渲染游戏编辑器UI方面有着完全不同的方法。在这个讲座中，Guerrilla的工具程序员Wessel Mast将告诉你Guerrilla的编辑器“Decima”如何渲染UI：它如何使用基于瓦片(tile-based) 的渲染器，通常在光线遮蔽(light culling)、SSAO、景深(depth of field)等方面常用，以高效的方式绘制复杂的UI，以及如何通过有符号距离场(signed distance fields)一次绘制调用来绘制成千上万的矩形、线条和曲线。
作者：TBA . (TBA, Unity)
Its expensive to create and challenging to transform 3D content across multiple pipelines and applications. Just finding and using the right content can be a huge source of lost time and broken workflows. In this session, we discuss the unique challenges of 3D digital asset management (DAM) and how the new Unity Asset Manager can help organize your content across the studio and get great-looking games to market faster.
在创建和跨多种流程和应用程序转换3D内容方面是非常昂贵且具有挑战性的。仅仅是找到和使用正确的内容就可能是巨大的时间浪费和工作流程中断的主要原因。在本次论坛中，我们将讨论3D数字资产管理（DAM）的独特挑战，以及新的Unity资产管理器如何帮助您在整个工作室间组织内容，并更快地将高质量的游戏推向市场。
Unity Developer Summit: The Great Data Bake Off: Improving Performance with DOTS Baking (Presented by Unity)
作者：TBA . (TBA, Unity)
Attend this session to see how you can maximize your game's performance using specialized features of Unity's Data-Oriented Tech Stack (DOTS). This includes baking – a crucial process that converts scene data into entities and components. You also learn how to benefit from configurable components, custom allocators, and other features that help with performance optimization, more efficient memory utilization, and effective debugging.
参加这个会话，了解如何利用Unity数据导向技术堆栈（DOTS）的专业特性，最大化你游戏的性能。这包括烘焙（baking）- 将场景数据转换为实体和组件的关键过程。你还会学习如何充分利用可配置的组件，自定义分配器和其他功能，这些功能有助于性能优化，更高效的内存利用以及有效的调试。
Unity Developer Summit: Unity 6 and Beyond: A Roadmap of Unity Engine and Services (Presented by Unity)
作者：TBD . (TBD, Unity)
This is the can't-miss session of GDC 2024. Join Unity to discover the latest developments in the Unity Engine and how they're supported through additional tools and services to maximize your creative freedom and efficiency, while enabling you to produce gorgeous, immersive games that will wow players over and over.
In just an hour, we show you how to achieve fantastic graphics fidelity and flexibility through rendering pipelines, shader/VFX Graph, and AI tools like Muse. In addition, with new tools like Multiplayer Center and Multiplayer Playmode, you get a head start on streamlining your multiplayer development. And of course we haven't forgotten that a lot of you work in teams, so learn about some great collaboration tools such as Unity Cloud (with Azure) and Muse project-aware chat. Finally, we know it's all about platform reach – so we have you covered there, too.
这是2024年GDC绝对不能错过的演讲。加入Unity，探索Unity引擎的最新发展，以及通过额外的工具和服务支持，以最大限度地提升您的创造自由和效率，同时让您制作出令玩家们反复惊叹不已的华丽沉浸式游戏。
在仅仅一个小时内，我们将向您展示如何通过渲染管线、着色器/VFX Graph和像 Muse 这样的人工智能工具，实现出色的图形保真度和灵活性。此外，借助诸如 Multiplayer Center 和 Multiplayer Playmode 等新工具，您可以提前了解如何简化多人游戏开发。当然，我们也不会忘记许多人是团队合作，所以了解一些出色的协作工具，例如 Unity Cloud （集成Azure） 和 Muse 项目感知聊天。最后，我们知道平台覆盖面至关重要 - 所以在这方面也已为您考虑周全。
作者：Joshua Simmons (Senior Programmer, Massive Entertainment - A Ubisoft Studio)
Avatar: Frontiers of Pandora is Massive Entertainment's largest and most complex world to date. This session describes the changes made to the Snowdrop engine to support a huge world densely filled with procedurally generated objects, as well as the optimizations necessary to support online co-op server hosting and local servers for offline play.
《阿凡达：潘多拉前线》是至今为止育碧最大最复杂的世界。本节描述了对Snowdrop引擎的改动，以支持一个充满程序生成对象的庞大世界，以及为支持在线合作服务器托管和本地服务器提供离线游玩所需的优化。
作者：Louis Marcoux (Technical Specialist | Omniverse for Games Developers, NVIDIA)
NVIDIA has created AI models which are ideal for the implementation of Smart NPCs in games. In this session we’ll present NVIDIA’s Audio-To-Face technology, which uses AI to generate facial animation based on speech input as audio – especially useful in real-time for games where dialog and speech are synthesized, not pre-canned. We’ll explain how anyone developing smart NPC tools or experiences can integrate A2F in their own projects, at production or runtime stages. Attendees will learn how to integrate speech-driven AI facial animation with standard and custom games character rigs.
NVIDIA已经开发出适用于游戏中智能角色NPC实现的人工智能模型。在本次会议中，我们将介绍NVIDIA的“Audio-To-Face”技术，该技术利用人工智能基于语音输入生成面部动画，尤其在需要实时合成对话和语音的游戏中非常有用。我们将解释任何开发智能NPC工具或体验的人如何在其项目中集成A2F，在生产或运行时阶段。与会者将学习如何将基于语音驱动的人工智能面部动画与标准和自定义游戏角色绑定。
作者：Seth Schneider (Product Manager, AI for Games, NVIDIA)
This GDC, NVIDIA and our partners will show a technology demo based on Generative AI technologies that lets the player explore a new way of playing with NPCs in a game. In this session we will discuss how these technologies are implemented and how Smart NPCs will change approaches to game design and narrative.
SmartNPCs are quite different from classic NPCs. In the Classic model, designers can deploy a backstory via a dialog manager. Player choices during interaction mostly are limited to pre-defined paths in the dialog manager.
With SmartNPCs, the paradigm is completely changed: the limited dialog tree is no longer needed. Instead, designer will give the SmartNPC a back story, a knowledge base, a personality; basically fill up the "brain of the NPC" and give it some latitude to answer the questions of the player while still sticking to the lore of the game.
We will discuss some different gamplay scenarios, and detail the process we went through to design a game demo using this new technology that will greatly impact the future of gaming.
在今年的GDC上，NVIDIA和我们的合作伙伴将展示基于生成式人工智能技术的技术演示，让玩家在游戏中以全新方式与NPC互动。在这个环节中，我们将讨论这些技术是如何实现的，以及智能NPC将如何改变游戏设计和叙事的方式。
'SmartNPCs'和经典NPC完全不同。在经典模型中，设计师可以通过对话管理器部署背景故事。玩家在互动过程中的选择大多局限于对话管理器中预定义的路径。
而在'SmartNPCs'中，范式完全改变：不再需要有限的对话树。相反，设计师会为'SmartNPC'提供一个背景故事、知识库、个性；基本上填满“NPC的大脑”，并为其提供回答玩家问题的一定自由度，同时仍然坚守游戏的故事背景。
我们将讨论一些不同的游戏玩法场景，并详细介绍我们为设计一款使用这项新技术的游戏演示所经历的过程，这将极大地影响游戏的未来。
链接：https://schedule.gdconf.com/session/verse-update-presented-by-epic-games/903280
作者：Phil Pizlo (Sr. Director, Language Engineering, Epic Games)、Tim Tillotson (Gameplay Engineer Director, Epic Games)
A look at the latest Verse features and what's coming soon to the language and framework.
查看最新的Verse特性以及即将到来的语言和框架。
链接：https://schedule.gdconf.com/session/warp-high-fidelity-cloth-simulation-in-ea-sports-fc-24/898847
作者：Robin Taillandier (Senior Software Engineer, Electronic Arts)
This talk will present Warp, a new GPU compute shader-based cloth simulator developed for Electronic Arts Frostbite engine and shipped with EA Sports FC24. Warp is used for the simulation of footballers jerseys and shorts and thanks to a drastically increased cloth mesh density and realistic collisions with character bodies, it provides improved authenticity while retaining excellent performance. Using videos of FC24, in-game debug rendering, screenshots and diagrams, this technical presentation will illustrate the solutions adopted to produce a step-change in runtime cloth behavior. It will explain how to exploit the potential and overcome the limitations of running physics simulations on the GPU, and showcase its impact on character authenticity in FC24.
这次演讲将介绍“Warp”，这是一个基于GPU计算着色器开发的布料模拟器，专为《Frostbite》引擎开发，随后与EA Sports的《FC24》一同发布。Warp用于模拟足球运动员的球衣和短裤，通过大幅提高布料网格密度并实现与角色身体的真实碰撞，提升了真实性同时保持出色的性能。通过展示FC24的视频、游戏内调试渲染、截图和图表，本技术演示将展示为实现运行时布料行为的重大飞跃采取的解决方案。它将解释如何充分利用GPU运行物理模拟的潜力并克服限制，并展示其对FC24中角色真实性的影响。