Research Notes
Isaac Sim's USD-native scene representation runs from asset import through deployment in one shared runtime, pairing switchable PhysX and Newton physics backends with RTX and physics-based sensor simulation in the same platform. Controllable synthetic data generation includes RGB, bounding-box, and segmentation annotators with COCO and KITTI export.
NVIDIA extends the platform with companion open-source tooling: Isaac Lab for robot learning, Isaac TeleOp for demonstration capture, NuRec neural reconstruction for closing the sim-to-real gap, and Cosmos world models for data augmentation. Named robotics developers including Boston Dynamics and Figure AI have announced adoption of the Isaac and Omniverse stack.
Isaac Sim carries a high hardware floor: official requirements start at 16 GB of VRAM on an RTX-class GPU, with 48 GB recommended for complex scenes, and NVIDIA warns that smaller GPUs may be insufficient for large renders. Supported platforms are limited to Ubuntu 22.04 or 24.04 and Windows 11, with Windows 10 no longer supported.
The simulator is heavier than minimal alternatives, with first-launch extension caching that can take several minutes and a learning curve reviewers describe as steeper than simpler simulators. While Isaac Sim itself is free under Apache 2.0, redistributing the bundled Omniverse Kit inside a commercial product requires a separate Omniverse Enterprise license.
Isaac Sim grew out of NVIDIA's robotics simulation research, with its first public release, Isaac Sim 2019.1, announced on the NVIDIA developer forum in April 2019 alongside the Isaac SDK. The line expanded through subsequent generations: version 4.0 was announced in 2024, and the 5.0 generation reached general availability at SIGGRAPH in August 2025.
The 5.0 release also open-sourced the codebase under Apache 2.0 on GitHub, marking the shift to open development. Releases have continued on a rapid cadence since, with 5.1.0 in October 2025, 6.0.0 in June 2026, and 6.1.0 in September 2026, and a 7.0 alpha preview in progress.
NVIDIA Isaac Sim is an open-source robotics simulation application built on NVIDIA Omniverse libraries and OpenUSD, used to design, test, and train AI-driven robots in virtual environments. It is provided free under the Apache 2.0 license and distributed through GitHub, NGC containers, pip and conda wheels, and cloud marketplaces.
The simulator combines GPU-accelerated physics with RTX sensor rendering and supports end-to-end workflows for synthetic data generation, reinforcement learning with the companion Isaac Lab framework, ROS and ROS 2 integration, and digital-twin simulation. Robot models and environments are imported from CAD, URDF, MJCF, and real-world captures into a single shared USD scene.
