CoreWeave Introduces Physical AI Templates for Robotics Teams

CoreWeave Physical AI Launchables gives robotics and autonomous vehicle developers tested templates for running AI workloads on CoreWeave’s cloud, reducing the setup work before a job can begin.

CoreWeave unveiled the product Oct. 21 at NVIDIA GTC Berlin. Its engineers built and tested the templates so teams can start a job with one command. The templates reduce preparation involving computing environments, storage and job execution.

The initial templates support workloads using NVIDIA Isaac Sim, CARLA, MuJoCo, NVIDIA AlpaSim and NVIDIA Cosmos 3. They cover data generation, model evaluation and policy training for physical AI development.

The product also includes Cosmos Data Factory agent skills for enhancing, labeling and organizing video, image and text data. Skills for Claude Code, Codex and other coding agents can start jobs, monitor progress and retrieve results.

CoreWeave is working with Cadence, Convergent Science, Encord, nTop, PhysicsX and Voxel51 to bring their physical AI software to its cloud infrastructure.