2023 Challenge

Our 2023 AI2-THOR Rearrangement Challenge has several upgrades distinguishing it from the 2022 version:

1. New AI2-THOR version. We've upgraded the version of AI2-THOR we're using from 5.0.0, this brings performance improvements and bug fixes.
2. New dataset. We've released a new rearrangement dataset to match the new AI2-THOR version. This new dataset has a more uniform balance of easy/hard episodes and requires interaction with more objects.
3. Improved object-opening logic. In previous versions of the rearrangement challenge there was no downside to attempting to open all objects that the agent came across as the open action would only execute when opening an object that was in a state different from that during walkthrough phase. In this version of the challenge, all openable objects have open and closed states that are toggled when agent issues the open action upon them.
4. Misc. improvements. We've fixed a number of minor bugs and performance issues from the 2022 challenge improving consistency.

2022 Leaderboards and Embodied CLIP Model

This release adds:

1. Links to the 2022 leaderboards.
2. Pretrained model and experiment configuration file for the embodied CLIP model that is SOTA for the 1-phase 2021 leaderboard. Relevant paper: https://arxiv.org/abs/2111.09888 .

2022 AI2THOR-Rearrangement Challenge

🔥🆕🔥 2022 AI2THOR-Rearrangement Challenge

Our 2022 AI2-THOR Rearrangement Challenge has several upgrades distinguishing it from the 2021 version:

1. New AI2-THOR version. We've upgraded the version of AI2-THOR we're using from 2.1.0 to 4.1.0, this brings:
   • Performance improvements
   • The ability to use (the recently announced) headless rendering feature, see
     here this makes it much easier to run
     AI2-THOR on shared servers where you may not have the admin privileges to start an X-server.
2. New dataset. We've released a new rearrangement dataset to match the new AI2-THOR version. This new dataset
   has a more uniform balance of easy/hard episodes.
3. Misc. improvements. We've fixed a number of minor bugs and performance issues from the 2021 challenge improving
   consistency.

v0.4.1 Patch

This patch updates the AI2-THOR commit version in order to improves the quality of the data returned by the semantic mapping sensors. This fixes a bug where semantic map sensors might return incorrect convex hulls for "Drawer" objects.

Semantic Mapping Experiments Release

This release includes three new experiments corresponding to the semantic mapping experiments from the paper:

• baseline_configs/walkthrough/walkthrough_rgb_mapping_ppo.py - This experiment is used to pretrain the active neural SLAM (ANM) model. We have released pretrained ANM model weights which are automatically downloaded when running the below 1- / 2- phase mapping experiments (e.g. see here for an example of this being done).
• baseline_configs/one_phase/one_phase_rgb_resnet_frozen_map_dagger.py - The 1-phase mapping experiment where the pretrained (semantic) map is frozen.
• two_phase/two_phase_rgb_resnet_frozen_map_ppowalkthrough_ilunshuffle.py - The 2-phase mapping experiment where the pretrained (semantic) map is frozen.

Leaderboard Details and Improved Physics Determinism

This release includes two central updates:

1. New instructions and links to our online challenge leaderboard along with instructions describing how to create submissions.
2. Use of a new AI2-THOR build that improves physics determinism (necessary for making leaderboard evaluation repeatable).

2021 Rearrangement Challenge

This release includes brings this repository up to date for our 2021 challenge. This includes several substantial upgrades including:

1. Definitions of new challenge tracks.
2. Integration of our rearrangement environment into AllenAct abstractions.
3. New baseline models with associated pretrained models.
4. Large amounts of documentation.
5. New training, validation, and test sets.

🎉 0.1.0

fix held object rotation scales