CoreWeave launches Physical AI Launchables

CoreWeave launched Physical AI Launchables, a catalog of tested templates for robotics and autonomous vehicle teams to run common workloads without rebuilding cloud environments.

“robotics and autonomous vehicle teams are doing some of the most demanding engineering in this industry”

Templates support Nvidia tools

At launch, Physical AI Launchables supports Nvidia Isaac Sim, CARLA, MuJoCo, Nvidia AlpaSim and Nvidia Cosmos 3 for data generation and policy training.

Templates include coding agents

Each template includes agent skills for Claude Code, Codex and other coding agents to adjust settings and monitor jobs.

Reduces setup work

The templates reduce setup work by managing resource limits, gpu use, parallel jobs and results for robotics and autonomous vehicle teams.

Works with existing infrastructure

Physical AI Launchables integrates with CoreWeave’s infrastructure, monitoring tools and Forge platform for developing models and agents.

Partners include Cadence

CoreWeave partners with Cadence, Convergent Science, Encord, nTop, PhysicsX and Voxel51 to run physical ai software on its infrastructure.

Templates handle data generation

Templates cover data generation, evaluation, grading and policy training with auto-labeling and data curation for ai workflows.

What the templates do

The templates reduce infrastructure work by reusing environments, limiting gpu use, running jobs in parallel, automating runs and centralizing results.