Skip to content

Command line

Use tower --help, tower smoke --help, tower run --help and tower score --help in the installed evaluator environment. tower smoke is asset-free; tower run accepts a TOML configuration. Benchmark physics commands are documented in the public runtime guide.

Multi-scene benchmark

tools/simulator/benchmark_run.py run evaluates a task with the Isaac interpreter. Supply --task, --assets, --package, --policy-url, --ik-python, --output and --accept-eula; --data-root, --gpu and shader-cache flags match the portable runner. --scenes defaults to 0-49 and accepts comma-separated ids and ranges. --batch-size defaults to 1 and is limited to 10. --keep-going continues after failed batches without retrying. --pilot-review is optional maintainer qualification.

benchmark_run.py aggregate --runs DIR... --output DIR combines disjoint runs without Isaac. Duplicate scenes and mixed identities or qualifications are errors.