Notice: This repository is a fork of the main Isaac Lab repo built on top of upstream commit 84d0ff05. Modifications for heterogeneous training are layered directly on top of the core framework, maintaining a clean Git history.
Notice: This project is different from IsaacLab-HARL which focuses on heterogeneous multi-agent learning in Isaac Lab. In this project, we focus on heterogeneous single-robot learning, where each environment contains a single robot but the robots across environments are different. This allows us to train morphology-agnostic policies across multiple quadrupedal and humanoid/bipedal robots simultaneously.
Isaac Lab is a GPU-accelerated, open-source framework designed to unify and simplify robotics research workflows, such as reinforcement learning, imitation learning, and motion planning. Built on NVIDIA Isaac Sim, it combines fast and accurate physics and sensor simulation, making it an ideal choice for sim-to-real transfer in robotics.
This fork introduces Hetero-IsaacLab, a specialized architecture for training morphology-agnostic locomotion policies across multiple heterogeneous quadrupedal and humanoid/bipedal robotic environments simultaneously.
Most physics simulators and RL frameworks assume homogeneity, making it difficult to train universal controllers. This repository bridges that gap, providing concrete advantages:
- Morphology-Agnostic Feature Learning: The policy is forced to learn fundamental locomotion principles that transcend specific hardware morphologies rather than memorizing robot-specific quirks.
- Efficient Multi-Platform Deployment: Training 8 quadruped or 5 humanoid robot types heterogeneously uses the same compute as training 1 robot type, eliminating the need to maintain separate codebases and models.
- Better Exploration: Different morphologies explore different regions of the state-action space naturally (e.g., varying leg lengths, weights, and joint actuation limits discover robust, versatile gaits).
- Dual-Paradigm Heterogeneous Environments: Supports both Manager-Based environments and high-throughput Direct RL workflows (
HeteroQuadrupedEnvandHeteroHumanoidEnv). - Quadruped Joint Unification: Enforces an "ANYmal Joint-Major" format, mapping diverse quadruped joint orders (e.g., Spot, Unitree, ANYmal) to a standard policy format.
- Native Humanoid Joints: Each humanoid keeps its own joints in its own order, with no padding: observations and actions are per robot (
{robot_name: tensor}), sized by that robot's joints (from 12-DOF biped legs to 30+ DOF full-body humanoids with arms and hands). - Index Mapping System: Efficient zero-copy conversion between global environment IDs and robot-local indices.
- Comprehensive Domain Randomization: Handles extreme morphological quirks with per-robot overrides for mass, center of mass, ground friction, velocity command ranges, and external disturbance pushes.
For a comprehensive breakdown of these experimental results, including the detailed methodology and the complete set of performance plots, the full report is available in this WandB report.
Our documentation page provides everything you need to get started with this framework.
Installation is quite similar to the original Isaac Lab's default installation, cloning this repository instead of the original one.
# Clone Hetero-IsaacLab
git clone https://github.com/modanesh/Hetero-IsaacLab.git
cd Hetero-IsaacLab
# Install dependencies (and training modules, eg rsl_rl)
./isaaclab.sh --install rsl_rlTo train on a specific subset of quadrupeds, pass the --quadrupeds flag from anymal_d,anymal_c,anymal_b,unitree_a1,unitree_go1,unitree_go2,unitree_b2,spot:
# Train on all 8 quadruped robots with 4096 environments
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py \
--task=Isaac-Velocity-Flat-HeteroQuadruped-v0 \
--quadrupeds anymal_d,anymal_c,anymal_b,unitree_a1,unitree_go1,unitree_go2,unitree_b2,spotTo train on heterogeneous humanoid/biped robots (cassie,digit,g1,h1,h2), pass the --humanoids flag:
# Train on all 5 humanoid robots with 4096 environments
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py \
--task=Isaac-Velocity-Flat-HeteroHumanoid-v0 \
--humanoids cassie,digit,g1,h1,h2You can also train on the standalone Unitree H2 full-body humanoid (31-DOF) via the Manager-Based RL workflow:
# Download official Unitree H2 USD assets (if needed)
python scripts/tools/download_h2_assets.py
# Train Unitree H2 on flat terrain
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py \
--task=Isaac-Velocity-Flat-H2-v0 \
--num_envs=4096
# Play and visualize trained H2 policy
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/play.py \
--task=Isaac-Velocity-Flat-H2-Play-v0 \
--num_envs=16To evaluate your trained policies and generate high-resolution cinematic videos of all your heterogeneous robots, you can use the play.py script. By default, Isaac Lab spans robots widely across the terrain. There is a custom --cluster_robots flag that automatically disables the curriculum, clusters all robots into a tight dynamic grid, and overrides the camera to track the pack in 1080p resolution.
# Evaluate 14 robots clustered tightly on the 'boxes' terrain in 1080p
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/play.py \
--task=Isaac-Velocity-Rough-HeteroQuadruped-v0 \
--quadrupeds anymal_d,anymal_c,anymal_b,unitree_a1,unitree_go1,unitree_go2,unitree_b2 \
--load_run ".*5.1_rough.*" \
--num_envs 14 \
--video --video_length 1000 --headless \
--cluster_robotsNote: You can easily change the specific terrain block the robots cluster on by modifying the target_difficulty_row and target_terrain_col variables directly in play.py.
Isaac Lab is built on top of Isaac Sim and requires specific versions of Isaac Sim that are compatible with each release of Isaac Lab. Below, we outline the recent Isaac Lab releases and GitHub branches and their corresponding dependency versions for Isaac Sim.
| Isaac Lab Version | Isaac Sim Version |
|---|---|
main branch |
Isaac Sim 4.5 / 5.0 / 5.1 |
v2.3.X |
Isaac Sim 4.5 / 5.0 / 5.1 |
v2.2.X |
Isaac Sim 4.5 / 5.0 |
v2.1.X |
Isaac Sim 4.5 |
v2.0.X |
Isaac Sim 4.5 |
We wholeheartedly welcome contributions from the community to make this framework mature and useful for everyone. These may happen as bug reports, feature requests, or code contributions. For details, please check our contribution guidelines.
- Please see the troubleshooting section for common fixes or submit an issue.
- For issues related to Isaac Sim, we recommend checking its documentation or opening a question on its forums.
- Please use GitHub Discussions for discussing ideas, asking questions, and requests for new features.
The Isaac Lab framework is released under BSD-3 License. The isaaclab_mimic extension and its corresponding standalone scripts are released under Apache 2.0. The license files of its dependencies and assets are present in the docs/licenses directory.
If you use Isaac Lab in your research, please cite their technical report:
@article{mittal2025isaaclab,
title={Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning},
author={Mayank Mittal and Pascal Roth and James Tigue and Antoine Richard and Octi Zhang and Peter Du and Antonio Serrano-Muñoz and Xinjie Yao and René Zurbrügg and Nikita Rudin and Lukasz Wawrzyniak and Milad Rakhsha and Alain Denzler and Eric Heiden and Ales Borovicka and Ossama Ahmed and Iretiayo Akinola and Abrar Anwar and Mark T. Carlson and Ji Yuan Feng and Animesh Garg and Renato Gasoto and Lionel Gulich and Yijie Guo and M. Gussert and Alex Hansen and Mihir Kulkarni and Chenran Li and Wei Liu and Viktor Makoviychuk and Grzegorz Malczyk and Hammad Mazhar and Masoud Moghani and Adithyavairavan Murali and Michael Noseworthy and Alexander Poddubny and Nathan Ratliff and Welf Rehberg and Clemens Schwarke and Ritvik Singh and James Latham Smith and Bingjie Tang and Ruchik Thaker and Matthew Trepte and Karl Van Wyk and Fangzhou Yu and Alex Millane and Vikram Ramasamy and Remo Steiner and Sangeeta Subramanian and Clemens Volk and CY Chen and Neel Jawale and Ashwin Varghese Kuruttukulam and Michael A. Lin and Ajay Mandlekar and Karsten Patzwaldt and John Welsh and Huihua Zhao and Fatima Anes and Jean-Francois Lafleche and Nicolas Moënne-Loccoz and Soowan Park and Rob Stepinski and Dirk Van Gelder and Chris Amevor and Jan Carius and Jumyung Chang and Anka He Chen and Pablo de Heras Ciechomski and Gilles Daviet and Mohammad Mohajerani and Julia von Muralt and Viktor Reutskyy and Michael Sauter and Simon Schirm and Eric L. Shi and Pierre Terdiman and Kenny Vilella and Tobias Widmer and Gordon Yeoman and Tiffany Chen and Sergey Grizan and Cathy Li and Lotus Li and Connor Smith and Rafael Wiltz and Kostas Alexis and Yan Chang and David Chu and Linxi "Jim" Fan and Farbod Farshidian and Ankur Handa and Spencer Huang and Marco Hutter and Yashraj Narang and Soha Pouya and Shiwei Sheng and Yuke Zhu and Miles Macklin and Adam Moravanszky and Philipp Reist and Yunrong Guo and David Hoeller and Gavriel State},
journal={arXiv preprint arXiv:2511.04831},
year={2025},
url={https://arxiv.org/abs/2511.04831}
}In addition to Isaac Lab's acknowledgements and citations, please consider citing the Hetero-Isaac repository and technical blog post if you use it in your research:
@software{heteroisaac,
author = {Danesh, Mohamad H.},
title = {Hetero-Isaac: Heterogeneous Quadrupedal Simulation built atop Isaac Lab},
year = {2026},
url = {https://github.com/modanesh/Hetero-IsaacLab},
license = {BSD-3-Clause},
doi = {10.5281/zenodo.19488668}
}
@misc{danesh2026heterogeneous,
title = {Heterogeneous Environments in Isaac Lab},
author = {Danesh, Mohamad H.},
year = {2026},
howpublished = {Technical Blog Post},
url = {https://modanesh.github.io/blog/hetero-isaaclab},
doi = {10.17605/OSF.IO/M4DGU}
}