NVIDIA Isaac Lab
NVIDIA Isaac™ Lab is an open-source unified framework for robot learning to train robot policies. Isaac Lab is built on top of NVIDIA Isaac Sim™, providing high-fidelity physics simulation using NVIDIA PhysX® and photo-realistic rendering. It bridges the gap between high-fidelity simulation and perception-only robot training, helping developers and researchers more efficiently build intelligent, adaptable robots with robust, perception-enabled, simulation-trained policies.
How Isaac Lab Works
Its modular architecture and GPU-based parallelization make Isaac Lab ideal for building robot policies that cover a wide range of embodiments, including humanoid robots, robotic arms, and autonomous mobile robots (AMRs).
This is a comprehensive platform for robot learning—from environment setup to policy training and deployment—giving you the flexibility to customize and extend its capabilities with various physics engines, including NVIDIA PhysX, Warp, and MuJoCo. NVIDIA Isaac Lab is also used for robot foundation model training by the NVIDIA Project GR00T research team.
Teach Robots New Skills
Create more robust, efficient, and capable robotic systems by teaching robots new skills in simulation. Robot learning in simulation helps reduce the need for extensive hardware expenses and time-intensive policy training iterations.
Quadruped Locomotion Policy Training
Boston Dynamics Spot quadruped locomotion policy training using Isaac Lab.
Teaching a Robot to Climb
Lightweight Berkeley Humanoid training in Isaac Lab to quickly climb the staircase.
Robot Learns in a Real-World Scenario
Fourier humanoid imitating a human in doing kitchen tasks via reinforcement learning.
Training a Humanoid Robot
MenteeBot humanoid robot training in a virtual warehouse setting to push a cart.
Key Benefits
Flexible Robot Learning
Customize workflows with robot training environments, tasks, learning techniques, and the ability to integrate custom libraries (e.g,. skrl, RLLib, rl_games, and more).
Reduced Sim-to-Real Gap
The GPU-accelerated PhysX version provides accurate, high-fidelity physics simulations. These include support for deformables that allows for more realistic modeling of robot interactions with the environment.
