NVIDIA Isaac Lab
developer.nvidia.com
- Free plan
- Windows
- Linux
- Self-hosted
7.3 Club score Visit developer.nvidia.com
Plans
Every plan and priceOpen source
Free
Free plan
What the maker publishes
- Cloud
- Isaac Automator deploys Isaac Lab to AWS, GCP, Azure, and Alibaba Cloud as a remote workstation. source
- Deployment
- LEAPP packages a policy compute graph and metadata into a portable bundle for real hardware. source
- Foundation-model post-training
- Isaac RL Post Training refines pretrained models such as Isaac GR00T with reinforcement learning and parallel simulation. source
- Included robots and tasks
- Version 3.0.0 ships more than 16 robot models and more than 30 environments, including manipulators, quadrupeds, and humanoids. source
- Learning integrations
- Environments can be trained with RSL RL, SKRL, RL Games, or Stable Baselines, and multi-agent reinforcement learning is supported. source
- Newton physics
- Newton can mix MJWarp, MPM, VBD, and Kamino to simulate rigid bodies, particles, deformables, and closed chains. source
- Notable limits
- Multi-GPU and multi-node training needs Linux and NVIDIA NCCL and is not supported on Windows; some Isaac Sim workflows use proprietary terms, while kit-less Newton workflows do not. source
- Platforms and install
- The default install uses Newton without Isaac Sim and runs on Ubuntu 22.04+ x86_64, Linux aarch64 including DGX Spark, or Windows 11 x86_64, with Python 3.12. source
- Rendering and scene tools
- Rendering ranges from high throughput to photorealism with Newton and OVRTX, including 3D Gaussian scenes and Cosmos 3-Nano Transfer. source
- Sensors
- Sensor simulation includes RTX cameras, LIDAR, and contact sensors. source
- Simulation backends
- Users choose physics, rendering, and visualization backends and can train and evaluate across them. source
- Support
- Questions and feature requests go to GitHub Discussions; GitHub Issues are for scoped bugs, docs, and features. Recurring office hours are also offered. source
- Training scale
- Training can scale from one GPU to a cluster, including perception-in-the-loop workloads. source
- What it does
- Isaac Lab is an open-source framework for training and refining robot policies with reinforcement learning and imitation learning. source
- Who it is for
- It is aimed at robot-policy work across humanoids, manipulators, and autonomous mobile robots. source
Checked against the maker's pages on 14 Nov 2026.