OmniLink reference robot assessment source add-on, 2026-10-29
Apache-2.0. Add to a fresh writable public OmniSim v9.2.7 root.
This adds assessment code and specs; the public reference controllers,
robot models and Go2 ONNX policy remain in the OmniSim installation.
Run python -m omnisim doctor. A working engine/physics runtime and
controller Python with numpy + onnxruntime are required. psutil is required
by the runner. The prepared Windows delivery includes these dependencies.
python assessment.py legged-run --spec projects/assessments/legged/examples/go2-before.json --out .tmp/go2-before-01
python assessment.py legged-run --spec projects/assessments/legged/examples/go2-after.json --out .tmp/go2-after-01
Use g1-before.json / g1-after.json for the standing assessment.
python legged-frozen-suite.py --robot go2 --out .tmp/go2-frozen-01
This repeats every published condition. legged-campaign starts the original
discovery template and does not include the later prospective challenges.
Exit 0 PASS, 1 valid task FAIL, 2 invalid/error. Always use fresh output directories.
G1 here is standing only; G1 walking demonstrations use a weight-bearing harness.
These cases have no support harness or external push. One-time joint-pose
initialization is shared across configurations. No hardware/OEM firmware validation.
Different builds/runtime versions can produce different measurements.

Go2 v1.1.0: 17 frozen conditions. Compare the previous 0.25 rad support guard with a 0.12 rad level-body guard at the same 15 s request. Both retain the same 1-tick blend, model, policy, gains and physics. Seven previously exposed timings are regressions; four new fractional timings were declared before testing. G1 remains the earlier 14-condition standing study.
