Benchmark Cosmos Policy on simulated robot tasks.
Runs simulated robot-manipulation evaluations and inference profiling for NVIDIA Cosmos Policy on LIBERO and RoboCasa.
When to use it
Use when evaluating Cosmos Policy benchmarks, profiling inference performance, or configuring headless GPU rendering for robot simulation.
Give it a Cosmos Policy evaluation request; it sets up the environment, runs simulation benchmarks, and validates the resulting evaluation logs.
The compute table specifies one A40/A100 GPU, with approximately 16 GB VRAM for LIBERO or 18 GB for RoboCasa.
The setup workflow requires building and entering the supported Docker container.
This skill
~/.libero/config.yaml
Writes your LIBERO configuration (irreversible)
GitHub
Downloads evaluation source from GitHub