MetaDrive-0.4.2.3

Summary

• Fix pull_asset functionality.
• Improve the human in the loop env.

What's Changed

• Fix pull_asset by @pengzhenghao in #607
• Render() should be called after engine.after_step() to make sure consistency by @QuanyiLi in #608
• Update default argument for TopDownRenderer by @QuanyiLi in #609
• Fix agentManager by @QuanyiLi in #610
• Fix issue on human-in-the-loop env by @pengzhenghao in #613
• Fix static obj by @QuanyiLi in #616
• Fix Xbox controller; introduce Takeover policy w/o brake by @pengzhenghao in #615
• Update description about MetaDrive by @pengzhenghao in #618

Full Changelog: MetaDrive-0.4.2.2...MetaDrive-0.4.2.3

MetaDrive-0.4.2.2

This release includes the following updates:

• Hotfix the pull_asset functionality. (Not completely fixed yet, try new release!)

MetaDrive-0.4.2.1

This release includes the following updates:

• Update the assets.zip for a tiny bug fix related to the object_summary in dataset_summary.pkl of Waymo example data.

MetaDrive-0.4.2.0

This release includes the following updates:

1. SUMO integration
2. Extract a BaseAgentManager class from the old AgentManager, so we can control more things, i.e. traffic lights in SUMO
3. Make the new doc system
4. Add more content to the doc
5. Fix memory leak when repeatedly calling reset() and close()
6. Use multi-pass rendering for sensors, so we can render the terrain in the camera
7. Include SimplePBR in the source code
8. Add more tests for sensors and install them from different sources
9. Remake the PSSM shadow
10. MeshTerrainSystem
11. ROSbridge
12. InstanceCamera
13. Rename TopDownRenderer. Now it supports semantic top-down renderer

MetaDrive-0.4.1.2

Include the folders under render_pipeline to fix the rendering bug.

MetaDrive-Scenario-0.4.1.1

We release the new MetaDrive with the following changes:

0. For supporting ScenarioNet, various updates are made like unifying all real-world environments and introducing scenario description.
1. Formally support the render_pipeline for advanced 3D rendering
2. Upgrade the 'gym' support to 'gymnasium'
3. Move the asset out of the repo
4. Refactor camera/sensor API allowing the creation of sensors with config
5. Automatically infer the render mode from none', 'onscreen', and 'offscreen
6. Unify coordinates for MetaDrive, Panda3D, and real-world data

MetaDrive-0.3.0.1

MetaDrive-0.3.0.1

In this new MetaDrive release, we have several important updates including:

• Rendering quality improvement via enabling PBR
• Nuplan dataset support, now more than 400,000+ scenarios can be loaded in MetaDrive
• Pedestrian, Cyclist and Traffic Light support
• Fix memory leak problem in Waymo/PG environment
• Image-on-cuda support which allows keeping the rendered image on GPU as pytorch.tensor and increase the image-observation sample efficiency by a large margin (20x faster)

MetaDrive-0.2.6.0

MetaDrive-0.2.6.0

This new release is for scenario-based autonomous driving research. We provide several new features like:

• Top-down view for Waymo Map
• New navigation system/reward function for real scenarios
• Efficiency improvement
• Release the updated Waymo dataset at https://github.com/metadriverse/metadrive-scenario/releases

MetaDrive-0.2.5

We now release MetaDrive-0.2.5, enabling the following features:

• Bug fixing in MARL environments, except ParkingLotEnv
• Replaying Waymo data and loading IDM Traffic in real environments
• Providing 1000+ Waymo scenarios split into training and test set to benchmark the performance of generalizability in real scenarios

For using real data, please note:

• Waymo motion dataset provides almost 30,000 cases collected in the real world. Due to the space limitation, we only release a part of them.
• Each training dataset contains a different number of scenarios randomly sampled from all cases.
• Validation set has 100 cases
• DO NOT violate the license terms of Waymo dataset

MetaDrive-0.2.4

MetaDrive-0.2.4

Recently, we add full support to Waymo and Argoverse datasets. Also, this draft contains several patches about bug-fixing and performance optimization.