Quadrrl is a training and benchmarking library for both legged and wheeled-legged quadruped robots. It includes deep reinforcement learning based locomotion training tasks across flat and rough terrains.
The table below shows a list of all the supported quadruped robots:
| Category | Robot Model | Environment Name | Image |
|---|---|---|---|
| Legged | Anymal C | Quadrrl-Velocity-Rough-Anymal-C-Direct-v0 |
 |
| Anymal D | Quadrrl-Velocity-Rough-Anymal-D-v0 |
 |
|
| Boston Dynamics Spot | Quadrrl-Velocity-Rough-Spot-v0 |
 |
|
| Unitree Go2 | Quadrrl-Velocity-Rough-Unitree-Go2-v0 |
 |
|
| Unitree B2 | Quadrrl-Velocity-Rough-Unitree-B2-v0 |
 |
|
| Deeprobotics Lite3 | Quadrrl-Velocity-Rough-Deeprobotics-Lite3-v0 |
 |
|
| Zsibot ZSL1 | Quadrrl-Velocity-Rough-Zsibot-ZSL1-v0 |
 |
|
| Legged-Wheeled | Unitree Go2W | Quadrrl-Velocity-Rough-Unitree-Go2W-v0 |
 |
| Unitree B2W | Quadrrl-Velocity-Rough-Unitree-B2W-v0 |
 |
|
| Deeprobotics M20 | Quadrrl-Velocity-Rough-Deeprobotics-M20-v0 |
 |
|
| Zsibot ZSL1W | Quadrrl-Velocity-Rough-ZSIBot-ZSL1W-v0 |
 |
- GPU: NVIDIA GPU with CUDA support (RTX 3060 or better recommended)
- CPU: Multi-core processor (8+ cores recommended)
- RAM: 16 GB minimum, 32 GB recommended
- OS: Linux (Ubuntu 20.04+) or Windows 10/11 (64-bit)
- Isaac Lab: Installed per official guide
- Python: 3.10 or newer
- Installation Guide - Setup instructions
- Getting Started - Quick start guide
- Simulation Videos - Video demonstrations
- Project Structure - Code organization
- Training Guide - Single-agent and multi-agent RL training
- Scripts Documentation - Available scripts and utilities
- Tasks Documentation - Task architecture
git clone https://github.com/lbnmahs/quadrrl.git
cd quadrrl
conda activate isaaclab
python -m pip install -e source/quadrrl
python scripts/list_envs.py
python scripts/reinforcement_learning/rsl_rl/train.py \
--task=Quadrrl-Velocity-Flat-Anymal-C-v0 \
--num_envs=4096See Installation Guide and Getting Started Guide for details.
- RSL-RL - Default framework with PPO support
- RL Games - NVIDIA's RL framework
- SKRL - Scikit-learn compatible RL library
- HARL - Multi-agent RL framework (customized for Isaac Lab)
We welcome contributions! Please see CONTRIBUTING.md for guidelines on how to contribute to Quadrrl.
- Fork the repository, create feature branches, and open pull requests with clear descriptions.
- Run
pre-commit run --all-filesbefore submitting changes. - Add tests or evaluation scripts when introducing new environments or reward structures.
- Update documentation when you add new tasks or major capabilities.
If you use Quadrrl in your research, please cite:
@software{quadrrl2026,
title={Quadrrl: Isaac Lab-Based Multi-Quadruped Locomotion Training and Performance Evaluation Suite},
author={Mahihu, Laban Njoroge},
year={2026},
url={https://github.com/lbnmahs/quadrrl}
}This project is licensed under the BSD-3-Clause License - see the LICENSE file for details.
Special thanks to Dr. Manal Helal for her guidance and advice. See ACKNOWLEDGMENTS.md for a complete list of acknowledgments.