The most substantive change in this version is embedding AI agents into the development workflow. NVIDIA introduced reusable "skills" for environment setup and manipulation, for developers and AI agents to complete robot development tasks; it also provides agent-oriented documentation so AI can more easily understand Isaac ROS tools and processes, thereby turning developer intent into runnable applications faster.
Some skills go beyond single-point coding assistance. The new FoundationStereo fine-tuning skill lets AI agents help adapt stereo perception models to developers' own cameras, environments, and robot applications, to achieve more accurate perception under specific sensor configurations. The common pick-and-place workflow—chaining detection, depth estimation, and pose output—is now offered as a standalone, agent-oriented skill, which NVIDIA says gives robot developers more flexibility beyond Isaac ROS itself.
On performance, NVIDIA says its FoundationPose foundation model for object pose estimation and tracking now provides an agent-oriented inference library that lets robots perceive and track object position and orientation up to 5.5x faster. To be clear: 5.5x is NVIDIA's own claimed speedup, and the source gives no benchmark configuration, comparison target, or third-party reproduction, so it should be treated as a vendor claim rather than confirmed fact.