Isaac Sim and Isaac Lab Are Now Available for Early Developer Preview

NVIDIA today released developer previews of NVIDIA Isaac Sim and NVIDIA Isaac Lab — reference robotics simulation and learning frameworks. Now available on GitHub, these releases offer early access to cutting edge capabilities for building, training, and testing AI-powered robots in physics-based simulation environments.

What’s new in Isaac Sim

An image of a simulated factory scene with boxes, ladders and a forklift, as generated by Isaac Sim. — *Figure 1: A simulated factory scene with a forklift in Isaac Sim*

Isaac Sim is a reference application built on NVIDIA Omniverse that enables users to develop, simulate, and test AI-driven robots in physically based simulation environments. Isaac Sim 5.0’s major updates include:

Open source availability

Isaac Sim-specific extensions are being open sourced in a new public Github repository, which will allow users to build and run Isaac Sim. This repository will serve as early access that’s available to the community, providing access to new features through a public development branch. Note that components of Omniverse Kit remain closed source, and external contributions are not accepted at this time.

This public development branch will be continuously updated with bug fixes. Binary releases will be provided for the official 5.0 release once all components have passed NVIDIA’s quality assurance standards.

Advanced synthetic data-generation pipelines

Isaac Sim introduces several new extensions that expand synthetic data generation capabilities for training, testing, and validating AI-powered robots

MobilityGen will be available as extensions in Isaac Sim, enabling diverse physics-based data and perception model training data generation, such as occupancy maps, robot states, poses, velocities, and images.
Grasp data generation will be available as a new tutorial introducing a workflow enabling automated generation of grasp candidates, simulating each grasp attempt, and recording success metrics for training and evaluation.
A new writer for NVIDIA Omniverse Replicator is optimized for NVIDIA Cosmos Transfer input, for users to easily generate and export high-quality synthetic data for ingestion. It supports standalone workflows and the Script Editor, and can be seamlessly integrated into existing Isaac Sim synthetic data generation scripts.
New workflows in Isaac Sim now support synthetic data generation for large-scale physical environments. These include Incident Simulation, which triggers diverse real-world scenarios, and Caption Generation, which automatically annotates incident data to enrich training datasets. Together, they enhance scene understanding for Vision AI model development. Additionally, existing Actor and Object Simulation workflows have been refined to further improve synthetic data quality and efficiency.

New robot models

Isaac Sim 5.0 introduces new robot models and import tools that make simulation setup faster, more consistent, and closer to real-world behavior.

A new robot schema standardizes robot definitions in OpenUSD, now adopted across NVIDIA’s expanded robot asset library featuring new robot models. Robot importers now use the new robot schema, and the new Robot Import Wizard offers a streamlined, step-by-step process for rigging and importing robots for simulation.
The new schema in Isaac Sim 5.0 also supports a new joint friction model defined through an OpenUSD schema, including actuator and friction parameters, developed in collaboration with Hexagon Robotics and maxon. These models ensure that the actuation of joints and motors in simulation closely mirrors real-world system dynamics, reducing the simulation-to-real gap.

Improved sensor simulation

Isaac Sim now offers a major leap in sensor simulation, making it easier than ever to define and test sensor models with greater realism and control.

A new OmniSensor USD schema will define RTX sensors directly within USD, alongside a new depth sensor model that simulates realistic stereo depth with detailed disparity artifacts. Additionally, the asset library has been expanded to include new sensor models.

Standardized ROS 2 interfaces and ZMQ bridge:

Isaac Sim 5.0 offers full support for ROS 2 Jazzy Jalisco, compatibility with standardized ROS 2 simulation interfaces, and an updated MoveIt 2 Tutorial for motion planning workflows. Additionally, a new ZeroMQ bridge facilitates communication between Isaac Sim and external systems. It also serves as a reference example for developing custom middleware bridges.

What’s new in Isaac Lab

An image of a robot standing in front of a table, representing a bi-manipulation task of pipe sorting using a fine-tuned GR00T N1 model with GR00T-mimic data in Isaac Lab. — *Figure 2: Isaac GR00T N1 deployed in Isaac Lab on a Fourier GR1 humanoid robot for a bi-manipulation task.*

Isaac Lab is an open-source framework purpose-built for training and evaluating robot learning policies. The new updates to Isaac Lab 2.2 enable:

Enhanced synthetic motion generation: This amplifies collected data for robot manipulation using NVIDIA Isaac GR00T-Mimic’s bi-manual training environment.
Omniverse Fabric integration: Isaac Lab now supports Fabric for faster load times and more efficient execution of physics simulations and sensor data collection.
Tensorized suction cup gripper: This provides flexible, dynamic gripping simulation, essential for accurate reinforcement learning and manipulation tasks.