v0.2.1

EmbodiChain v0.2.1 Release Notes

Highlights

This release upgrades EmbodiChain to DexSim v0.4.1, introduces the sim-ready data pipeline with full-mesh grasp annotation, adds emissive light randomization mode, and fixes CI/CD issues with multiversion documentation deployment.

Breaking Changes

• Upgraded dexsim_engine from 0.4.0 to 0.4.1 by @yuecideng in #272

New Features

• Added sim-ready data generation pipeline by @XuanchaoPENG in #271
• Added emissive light mode to randomize_indirect_lighting by @yuecideng in #274
• Added full-mesh antipodal grasp pose annotation by @matafela in #260

Improvements

• Adapted simulation API call sites to DexSim v0.4.1 by @yuecideng in #272
• Improved documentation navigation, cross-references, and guides by @yuecideng in #262

Bug Fixes

• Fixed atomic action behavior by @XuanchaoPENG in #273
• Fixed OPW solver TCP inverse pose calculation by @matafela in #261
• Fixed multiversion docs overwrite on main branch push by @yuecideng in #263 and #266
• Fixed multiversion docs deployment to GitHub Pages by @yuecideng in #267

Documentation

• Added academic publications page by @yuecideng in #265

All Commits Since v0.2.0

• Sim-ready pipeline by @XuanchaoPENG in #271
• Fix atomic action by @XuanchaoPENG in #273
• Add emissive light mode to randomize_indirect_lighting by @yuecideng in #274
• Adapt DexSim v0.4.1 by @yuecideng in #272
• Annotate full mesh by @matafela in #260
• Fix OPW TCP by @matafela in #261
• Fix multiversion docs deployment by @yuecideng in #267
• Fix multiversion docs overwrite on main push by @yuecideng in #266
• Add academic publications page by @yuecideng in #265
• Fix multiversion docs overwrite on main branch push by @yuecideng in #263
• Improve navigation, cross-references, and guides by @yuecideng in #262

---

Full Changelog: v0.2.0...v0.2.1

v0.2.0

EmbodiChain v0.2.0 Release Notes

Highlights

This release upgrades EmbodiChain to DexSim v0.4.0, adds the atomic action abstraction layer for embodied
motion generation, expands RL and robotics benchmarking, improves solver correctness, and introduces multi-
version documentation and PyPI release workflow fixes.

Breaking Changes

• Upgraded dexsim_engine from 0.3.11 to 0.4.0 by @yuecideng in #226
• Upgraded pytorch_kinematics from 0.7.6 to 0.10.0 by @matafela in #244
• Made lerobot>=0.4.4 a required dependency by @yuecideng in #227
• Upgraded formatter requirement to black==26.3.1 by @yuecideng in #251

New Features

• Added atomic action abstraction layer for embodied AI motion generation by @yuecideng in #239
• Added RL benchmark suites, metrics, plotting, reporting, and runtime support by @matafela in #231 and
  #233
• Added kinematic solver and workspace analyzer benchmark scripts by @matafela in #233
• Added multi-version documentation build support by @yuecideng in #234
• Added synthesized data generation tutorial by @yvvonie in #238
• Added multi-asset loading with spacing in the preview script by @yuecideng in #224

Improvements

• Adapted simulation, sensors, objects, materials, examples, and tests to DexSim v0.4.0 by @yuecideng in
  #226
• Enhanced workspace analyzer computational efficiency by @yuecideng in #230
• Refined URDF assembly component prefixes and name casing policy by @chase6305 in #236
• Improved API reference detail and coverage by @yuecideng in #252
• Updated PyTorch kinematic solver benchmark parameters by @matafela in #240
• Updated CobotMagic arm asset and default safe margin by @matafela in #228 and #235
• Updated skills structure and roadmap docs by @yuecideng in #257

Bug Fixes

• Fixed PyPI release workflow by @yuecideng in #259
• Fixed PyTorch solver qpos mapping by @matafela in #253
• Fixed plan_trajectory behavior by @yhnsu in #242
• Fixed demo action shape normalization and control-part mapping by @yuecideng in #237
• Fixed robot forward kinematics by @matafela in #223
• Fixed semantic mask robot user_ids indexing by @yuecideng in #222
• Fixed OPW solver behavior by @matafela in #229
• Fixed crash when no grasp pose is found by @matafela in #232
• Fixed README and website documentation links by @yuecideng in #246 and @yangchen73 in #249

Documentation

• Rewrote install guide and dependency guidance by @yuecideng in #227
• Added NVIDIA driver guide and mesh loading tutorial by @XuanchaoPENG in #245
• Added README badge upgrades and docs auto-sync by @yuecideng in #250
• Added AI coding agent skill cross-references by @yuecideng in #247
• Removed AI coding agent skill references by @yuecideng in #254
• Added uv installation support and updated dependencies by @yuecideng in #258

All Commits Since v0.1.3

• Fix PyPI release workflow by @yuecideng in #259
• Add uv installation support and update dependencies by @yuecideng in #258
• Adapt DexSim v0.4.0 by @yuecideng in #226
• Update skills structure and roadmap docs by @yuecideng in #257
• Add data generation tutorial for synthesized data pipeline by @yvvonie in #238
• Remove AI coding agent skills references by @yuecideng in #254
• Fix PyTorch solver qpos mapping by @matafela in #253
• Add atomic action abstraction layer for embodied AI motion generation by @yuecideng in #239
• Improve API reference detail and coverage by @yuecideng in #252
• Upgrade Black to 26.3.1 and fix code style by @yuecideng in #251
• Upgrade PyTorch kinematics by @matafela in #244
• Upgrade README badges and add auto-sync to introduction docs by @yuecideng in #250
• Correct website link by @yangchen73 in #249
• Add AI coding agent skills and documentation cross-references by @yuecideng in #247
• Fix README docs link by @yuecideng in #246
• Add multi-version documentation build support by @yuecideng in #234
• Add NVIDIA driver guide and mesh loading tutorial by @XuanchaoPENG in #245
• Fix plan_trajectory by @yhnsu in #242
• Update PyTorch kinematic solver benchmark parameters by @matafela in #240
• Refactor benchmark tooling by @matafela in #233
• Refine URDF assembly component prefixes and name casing policy by @chase6305 in #236
• Fix demo action shape normalization and control-part mapping by @yuecideng in #237
• Update CobotMagic default safe margin by @matafela in #235
• Enhance workspace analyzer computational efficiency by @yuecideng in #230
• Add RL benchmark by @matafela in #231
• Fix crash when no grasp pose is found by @matafela in #232
• Fix OPW solver by @matafela in #229
• Update CobotMagic arm asset by @matafela in #228
• Rewrite install guide and make LeRobot a required dependency by @yuecideng in #227
• Add multi-asset loading with spacing in preview script by @yuecideng in #224
• Fix robot forward kinematics by @matafela in #223
• Fix semantic mask robot user_ids indexing by @yuecideng in #222

---

Full Changelog: v0.1.3...v0.2.0

v0.1.3

EmbodiChain v0.1.3 Release Notes

Highlights

This release introduces multi-GPU RL training support, new simulation object types (cloth, grasp annotator), enhanced physics and action functor APIs, CLI improvements, and many bug fixes. For detail, please see below.

Breaking Changes

• Articulation Default Drive: Changed articulation default drive type to none by @matafela in #210
• Action Functor Modes: Added support for pre and post mode in action functor execution by @yuecideng in #183

New Features

• Multi-GPU RL Training: Added support for multi-GPU training for reinforcement learning by @yangchen73 in #188
• Cloth Object: Added cloth object simulation support by @matafela in #198
• Grasp Annotator: Added grasp annotator for annotation pipeline by @matafela in #196
• Mass Randomization: Added articulation mass randomization functor by @yuecideng in #219
• Extra Observations: Added support for extra observations from gym_config by @yuecideng in #205
• Asset Preview CLI: Added standalone asset preview script by @yuecideng in #207
• CLI Entry Points: Added top-level CLI entry points for preview_asset and run_env by @yuecideng in #214
• Configurable Data Roots: Added configurable data roots and asset download CLI by @yuecideng in #209

Bug Fixes

• Aligned Pose: Fixed using bugs of the get_aligned_pose and _prepare_warpping function by @wuxinxin27 in #217
• Gizmo API: Updated to new Gizmo API by @yuecideng in #213
• Camera Recording: Fixed camera recording save on episode reset by @yuecideng in #208
• Observation Manager: Fixed KeyError when add mode not present in observation_manager.active_functors by @yuecideng in #200
• Video Saving: Fixed video saving bug for the first frame in data generation by @yuecideng in #199

Improvements

• Physics Attributes: Enhanced physics attributes APIs and related functors by @yuecideng in #193
• Contact Sensor: Improved contact sensor data buffer and added support for saving data to LeRobot dataset by @yuecideng in #197
• Task Imports: Moved task env imports to tasks/__init__.py by @yuecideng in #212
• ARX5 Asset: Updated ARX5 robot assets by @matafela in #192

Documentation

• Functor Docs: Added hyperlinks and JSON config examples to functor docs by @yuecideng in #220
• Contribution Guide: Updated env and robot contribution guide by @yuecideng in #194
• PR Skill: Added PR skill for creating pull requests by @yuecideng in #201
• Contributor: Added Lizhe Chen as a contributor by @ChenlizheMe in #195

---

Full Changelog: v0.1.2...v0.1.3

v0.1.2

EmbodiChain v0.1.2 Release Notes

Highlights

This release introduces significant improvements to the RL training pipeline, new gripper support, USD interoperability, Online Data Streaming implementation and mang bug fixed. For detail, please see below.

Breaking Changes

• RL Rollout Refactoring: Refactored RL rollout with collector/buffer separation and shared TensorDict by @yangchen73 in
  #175
• Action Manager: Implemented an action manager to process action and replaced RLEnv with EmbodiEnv by @yangchen73 in #164

New Features

• SDF Collision: Added support for SDF collision for rigid bodies by @MahooX in #185
• Arena Space: Added arena_space argument to environment configuration by @yuecideng in #187
• Online Data Streaming: Added support for online data streaming engine by @yuecideng in #160
• USD Import: Added import USD functionality by @MahooX in #127
• USD Export: Added export USD functionality by @MahooX in #168
• GRPO Algorithm: Implemented GRPO (Group Relative Policy Optimization) algorithm by @yangchen73 in #146
• GRPO Camera: Added camera to record GRPO training by @yangchen73 in #146
• Gripper Support: Added Robotiq 85 140 DH AG95 gripper support by @matafela in #177
• Semantic Mask: Split robot left/right index in semantic mask computation by @dexscutai in #142
• Indirect Light: Added support for indirect light setting for environment by @yuecideng in #161
• Planner Enhancement: Added planner grid cell sampler by @yuecideng in #147
• Light Control: Added light keyboard control for runtime adjustment by @yuecideng in #140

Bug Fixes

• URDF Assembly: Fixed URDF assembly for copying mtl file by @yuecideng in #184
• Dataset Saving: Fixed dataset saving for single episode in run_env by @yuecideng in #178
• Contact Sensor: Fixed contact sensor visualization by @yangchen73 in #170
• Observation Manager: Fixed observation manager issue for adding new keys by @yuecideng in #169
• OPW Solver: Fixed OPW solver issues by @matafela in #152
• Robot IK: Fixed [xyz, quat] format pose IK computation for robots by @yuecideng in #149
• Trainer Config: Fixed num_envs override from trainer config by @yangchen73 in #148
• Cartpole Training: Fixed cartpole training by @yangchen73 in #155
• UR5 Asset: Updated UR5 assets by @yuecideng in #167

Improvements

• Functor Tests: Added unit tests for all functors by @yuecideng in #181
• RigidObject Velocity: Added set velocity for RigidObject by @yuecideng in #179
• Manager Docs: Enhanced manager documentation for action, reward, and dataset by @yuecideng in #176
• Dataset Filter: Added filter dataset saving for environment by @yuecideng in #143
• Cached Property: Updated cached property implementation by @yhnsu in #141
• Trajectory Interp: Added interpolated trajectory with segment numbers by @matafela in #151
• Physics Step: Updated physics step for spatial functors by @yuecideng in #144
• MotionGenerator: Refactor Planner and MotionGenerator by @matafela in #173

Documentation

• URDF Assembly Tool: Added documentation for URDF Assembly Tool by @yuecideng in #174
• New Robot: Added documentation for adding new robots by @yuecideng in #172
• RL Module: Updated RL module API documentation by @yuecideng in #171
• AGENTS.md: Added AGENTS.md for developer reference by @yuecideng in #165
• CLAUDE.md: Added CLAUDE.md with AI assistant guidelines by @yuecideng in #162

---

Full Changelog: v0.1.1...v0.1.2

---

This release contains 39 commits from 6 contributors: @Chen Yang, @Chen Jian, @yueci Deng, @MahooX, @yhnsu , and
@dexscutai.

v0.1.1

🚀 EmbodiChain v0.1.1 Release Notes

Highlights

This patch release focuses on stability and usability: environment observation space correctness, more accurate camera pose reporting in the keyboard control tool, and cleaner environment launch workflows—plus documentation updates to streamline installation and onboarding.

What’s Changed

✨New Features

• Add cartpole RL env by @matafela in #100
• Support detach assets from env automatic reset by @yuecideng in #121
• Add Franka FR3 by @matafela in #123
• Add ARX5 robot by @matafela in #124
• Add keyborad control to adjust camera sensor pose in simulation by @yuecideng in #128
• Add gizmo helper tool for robot control by @yuecideng in #132

🐛 Bug Fixes

• Fix observation space issue for BaseEnv by @yuecideng in #125
• Fix camera pose output from keyboard control tool by @yuecideng in #137
• Fix observation space issue for BaseEnv by @yuecideng in #125
• Fix setting articulation joint drive none by @matafela in #136

🔧 Improvements

• Refactor env launcher scripts by @yuecideng in #126
• Change visual material setter behavior for RigidObjectGroup by @yuecideng in #107
• Support shared visual material instances by @yuecideng in #110
• Enable get_sensor_pose_in_robot_frame for stereo camera by @dexscutai in #111
• Save success status before reset @yhnsu #119
• Suport camera group ids for batch rendering on RT backend by @yuecideng in #108
• Porf: modify URDF path handling: Pass from urdf_cfg to solver_cfg by @chase6305 in #135

📚 Documentation

• Update docs for installation from pypi by @yuecideng in #104
• Improve documentation by @yuecideng in #115
• Add sim docs @liguilong256 in #117

New Contributors

• @dexscutai — first contribution: #111

Full Changelog: v0.1.0...v0.1.1

v0.1.0

🚀 EmbodiChain v0.1.0 Release Notes

EmbodiChain is an end-to-end, GPU-accelerated, and modular platform for building generalized Embodied Intelligence. We are excited to announce this early release, which includes the basic project structure, the simulaiton components, the basic gym components for robot learning, the simple RL algorithm with examples, the simulation examples and the documentations.

The project website can be found here: https://dexforce.com/embodichain/index.html#/.
For more amazing features (coming in the future), please check the roadmap: https://dexforce.github.io/EmbodiChain/resources/roadmap.html

---

🌟The Contributors

We are grateful for our contributors in this release:

• @LihaoZheng-git
• @ZhaoRunyi
• @Jasonxu1225
• @yhnsu
• @chase6305
• @kerwan-zeng
• @matafela
• @yangchen73
• @liguilong256

Thank you for your contributions!