NVIDIA Isaac Lab

NVIDIA's open-source, GPU-parallel robot learning framework for training policies at scale in simulation.

Isaac Lab is an open-source robot learning framework (BSD-3 licensed) led by NVIDIA, evolved from the Orbit framework published in 2023, and NVIDIA's officially recommended replacement for Isaac Gym, which is no longer supported. It's built on top of Isaac Sim, using PhysX for physics and RTX for rendering, and can run large numbers of parallel environments simultaneously on the GPU; it comes with actuator models, a variety of sensors, domain randomization, and data-collection tools, and supports reinforcement learning, imitation learning, and motion planning. Many humanoid and quadruped reinforcement-learning locomotion policies are trained here. As of September 2026, version 3.0 is in early access and has moved to a multi-physics-backend architecture that can plug in the Newton physics engine, with some workflows no longer requiring Isaac Sim to be installed at all.

ExampleTo train a walking policy for the Unitree G1, open a few thousand parallel environments in Isaac Lab with domain randomization applied, then deploy the trained policy to the real robot.

Also called

Orbit, IsaacLab

Related

NVIDIA Isaac Sim · Isaac Gym · Orbit / IsaacGymEnvs / OmniIsaacGymEnvs (predecessors of Isaac Lab) · GPU-Accelerated Parallel Simulation · Newton Physics Engine · RL-based Locomotion Control

Sources

Isaac Lab Documentation
Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning (arXiv 2511.04831)
isaac-sim/IsaacLab releases (GitHub)
NVIDIA Developer: Isaac Gym - Preview Release (now deprecated, no longer supported)

As of

2026-09