robot_lab (RL extension library based on Isaac Lab)

An open-source library extending Isaac Lab with ready-made reinforcement-learning training environments for many legged and humanoid robots.

robot_lab is an open-source project (Apache 2.0 license) maintained by developer Ziqi Fan on GitHub, positioned as an extension library for Isaac Lab (NVIDIA's Isaac Sim-based robot-learning framework): rather than modifying Isaac Lab itself, it keeps robot models and training tasks in a separate directory so they can be developed independently. It ships with more than 25 robots, covering quadrupeds such as ANYmal D, Unitree Go2/A1/B2, and DEEP Robotics Lite3; wheeled-leg robots such as Go2W and B2W; and humanoids such as Unitree G1/H1, Fourier GR1, and Booster T1. Training mainly uses RSL-RL (ETH's open-source PPO implementation), with experimental extras such as AMP motion imitation and BeyondMimic motion tracking. As of September 2026, the latest release is v2.3.2, matching Isaac Lab 2.3. It suits newcomers who want to train locomotion policies for an off-the-shelf robot directly on Isaac Lab without writing an environment from scratch.

ExampleAfter installing Isaac Lab, pick robot_lab's velocity-tracking task for the Unitree Go2, train a walking policy with RSL-RL across thousands of parallel environments, then export the policy for sim-to-sim validation or deployment to the real robot.

Related

NVIDIA Isaac Lab · legged_gym · unitree_rl_gym (Unitree RL Gym) · rsl_rl · RL-based Locomotion Control · Massively Parallel Reinforcement Learning

Sources

GitHub - fan-ziqi/robot_lab
robot_lab Releases

As of

2026-09