NVIDIA released Isaac ROS 5.0 on Tuesday at the ROSCon conference in Toronto, Canada, introducing agentic development workflows and expanded platform support for robotics developers. NVIDIA said the GPU-accelerated software collection builds on Open Robotics' open-source ROS framework to serve an ecosystem of nearly 1.3 million ROS users.
The release adds support for ROS Lyrical and Ubuntu 24.04. Working with the Open Source Robotics Alliance, NVIDIA contributed a standard data-handling interface to ROS Lyrical designed to help robotics applications run efficiently across varied computing hardware, using CUDA as an implementation for GPU acceleration.
Isaac ROS 5.0 incorporates new Isaac skills for setup and manipulation, accompanied by agent-ready documentation to help AI agents process development workflows. A new FoundationStereo fine-tuning skill allows an AI agent to adapt stereo perception models to specific camera configurations and operational environments. NVIDIA also updated its FoundationPose model with an inference library that tracks object positions and orientations up to 5.5 times faster, while separating pick-and-place perception pipelines into a standalone skill.
Ecosystem integrations
Hardware and software developers have begun adopting the release across industrial robotics pipelines. Ekumen uses the isaac_ros_cumotion package on a GPU to calculate collision-free motion paths for warehouse arms in two to five milliseconds. Intrinsic integrated FoundationPose into its Open Machine Tending Solution within the Intrinsic Core suite. RealSense sponsored the open-source AgenticROS project to connect Isaac ROS with NVIDIA Nemotron open models and NemoClaw blueprints, while optimizing its D585 Pro stereo depth cameras for the Jetson Thor platform.
Seeed Studio paired Isaac ROS with its reBot Arm on Jetson Thor to handle perception, spatial understanding, and motion planning. Automotive supplier Magna integrated the software as a modular perception base for synchronized data collection and Isaac GR00T deployment, pairing it with Isaac Sim hardware-in-the-loop testing. Third-party tooling integrations include Prefix.dev's Pixi package manager for environment configuration, Ouster's integration of Stereolabs ZED stereo cameras, and Foxglove's desktop and web tools for visualizing live ROS data.
Hardware deployment
The platform runs across NVIDIA's edge computing hardware, spanning the entry-level Jetson Orin Nano up to Jetson Thor systems. Universal Robots incorporated Isaac ROS into its AI Accelerator software development kit to help arms adapt to unaligned parts without rigid fixtures. ROBOTIS deployed Isaac ROS CuMotion into its AI Worker platform for vision-guided picking, placing, and alignment, while Flexiv connected the software to its Rizon 4 adaptive robot.
Mentee Robotics adopted Isaac ROS as the primary perception stack for its MenteeBot humanoid across Jetson Orin and Thor hardware. FieldAI integrated the software to run autonomous robot foundation models directly at the edge without cloud connectivity, while Noble Machines deployed it for industrial general-purpose robots. NVIDIA published Isaac ROS 5.0 as free, open-source software available directly on GitHub.
