NVIDIA has released Isaac ROS 5.0, updating its GPU-accelerated robotics software with AI coding-agent skills and a change to how robot applications exchange GPU-backed data. The official release notes date version 5.0.0 to September 21 and list packages tagged v5.0.0 on GitHub.

The release moves to ROS 2 Lyrical Luth and adds an apt repository for its ecosystem packages on Ubuntu 24.04. Its acceleration layer now uses Lyrical's rosidl::Buffer interface with a CUDA backend, allowing standard ROS messages to carry array fields backed by GPU memory.

That change has a concrete migration cost. NVIDIA removed legacy NITROS packages and says applications that call their APIs or types directly need source-level changes. A retained ROS 2 bridge is deprecated and scheduled for removal in a future release.

Agent assistance, with upgrade caveats

New workflows use the open Agent Skills format. The Isaac ROS CLI includes a skill for activating development environments and an early-access skill for migrating nodes away from CUDA-based NITROS APIs. NVIDIA also describes skills for stereo-perception fine-tuning and pick-and-place development in its broader Isaac tooling.

The release adds GPU partitioning that allocates portions of a GPU's streaming multiprocessors to ROS 2 processes. NVIDIA explicitly says this does not partition GPU memory or isolate workloads.

Developers still face compatibility limits: RealSense cameras are supported only in Docker mode, and some motion-planning examples may fail with robot-vendor packages not yet certified for Lyrical. For existing deployments, the immediate task is checking API migration and hardware compatibility before adopting the new agent-assisted workflows.