Institute for Physical AI @ BMI · The Charlot Lab · TR-2026-17

RPO-Bench.

An open, deterministic benchmark for the two hard links of on-orbit logistics: non-cooperative capture (control) and monocular 6-DOF pose (perception). Fixed scenarios, fixed seeds, bit-identical every run — in a browser or headless in Node — so anyone can reproduce a score and try to beat it.

Capture · 40 non-cooperative rendezvous

controllersuccessmean Δvmean stepsreproduced

Pose · 30 tumbling-target tracks · monocular · SPEED metric

estimatorrotation errtranslation errSPEED scorereproduced

The published scores are computed headless with node bench/rpo_bench.mjs. Press the button to re-run the fast baselines live, in this tab — the numbers are bit-identical (the same seeds, the same math), which is the point: a benchmark you can verify, not trust.

Code, scenarios, and baselines → github.com/dcharlot-physicalai-bmi/orbital-logistics (bench/)
Live instruments → physicalai-bmi.org/research/charlot-lab/space-logistics · Learned policy → Hugging Face