CoreWeave launches one-command Physical AI templates for robotics and autonomous-vehicle workloads

What is changing?

CoreWeave has introduced a Physical AI Launchables catalog that lets robotics and autonomous-vehicle teams start complex AI workloads with a single command. The company says the templates are pre-tested to reduce infrastructure setup and cover simulation, synthetic data generation, training, and evaluation workflows.

The initial catalog includes ready-to-run workflows for NVIDIA Isaac Sim, CARLA, MuJoCo, NVIDIA AlpaSim, and NVIDIA Cosmos 3. Examples cover robotic pick-and-place tasks, CARLA scenario matrices, MuJoCo reinforcement-learning experiments, AlpaSim closed-loop autonomous-driving evaluations, and rare driving-scenario generation with Cosmos 3. Each template includes a manifest defining its inputs and outputs.

Users can modify settings such as scenarios, duration, parallelism, and models through plain text or name-value pairs. CoreWeave also says custom workloads can be created from blank templates and that the feature is available today through account-managed access. This does not guarantee that every template will produce identical results on every hardware setup.