Positioning — why this exists
Real factories already pair robots that were never trained together. A 500 Hz industrial arm next to a 50 Hz mobile base; a vision-only manipulator next to a force-feedback one; a vendor-A controller next to a vendor-B controller under binding ISO 10218-2:2025. At deployment time, your robot's teammate is opaque (no policy access), heterogeneous (different morphology and action frequency), and ad hoc (no prior joint training). Hospitals and warehouses are the same picture.
Most multi-robot benchmarks assume identical embodiments and shared training. The few that don't focus on planning or navigation, not on contact-rich physical manipulation. The intersection of Heterogeneity × Black-box partner × Safety × Manipulation is empty in the published literature. CHAMBER is built to fill it, and CONCERTO is the first method designed against this four-aspect contract.
How we sit relative to the closest prior work
Every prior precedent covers at most three of the four aspects. The table lists the closest precedent for each pair of aspects; no published row hits all four.
| Method | Heterogeneous | Black-box partner | Safety bound | Contact-rich manipulation |
|---|---|---|---|---|
| Liu 2024 RSS (LLM-AHT) | ✓ | ✓ | ||
| COHERENT (LLM-MR planning) | ✓ | ✓ | ||
| Huriot & Sibai 2025 (conformal CBF) | ✓ | ✓ | ||
| HetGPPO / HARL (heterogeneous MARL) | ✓ | |||
| Wang et al. 2017 (multi-robot CBFs) | ✓ | ✓ | ||
| RoCoBench (multi-robot manipulation) | ✓ | ✓ | ||
| SafeBimanual (safe bimanual manip.) | ✓ | ✓ | ||
| CONCERTO + CHAMBER | ✓ | ✓ | ✓ | ✓ |
Read the table by columns to see what each aspect covers in isolation, and by rows to see what no single line of work has yet combined. Contact-rich manipulation appears with multi-robot coordination (RoCoBench) and with safety (SafeBimanual), but never with black-box ad-hoc partners under explicit safety assumptions at the same time. Heterogeneous here is the four-aspect literature-gap level; CHAMBER's six measurable sub-axes decompose it further (below). See Why heterogeneous ad-hoc teamwork? for the long-form argument.
The six heterogeneity axes
| Axis | Symbol | What it varies | Validity (ADR-026) | Where the priors come from |
|---|---|---|---|---|
| Action space | AS | 7-DOF arm vs 2-DOF mobile base on shared task | construct-invalid on pick-place → retained as control; null on co-carry | HARL, HetGPPO |
| Observation modality | OM | vision-only vs vision + force/torque + proprioception | construct-invalid on pick-place → retained as control | Visual-tactile peg-in-hole literature |
| Control rate | CR | 500 Hz arm vs 50 Hz base, chunk size held constant | untested | RTC, A2C2, FAVLA |
| Communication | CM | latency 1–100 ms, jitter µs–10 ms, drop 10⁻⁶–10⁻² | untested | 3GPP R17, URLLC |
| Partner familiarity | PF | trained-with vs frozen-novel partner, mid-episode swap | untested | FCP, MEP |
| Safety | SA | mixed-vendor force-limit / SIL-PL pairs, contact-rich | untested | ISO 10218-2:2025 |
A preregistered ≥20 percentage-point homogeneous-vs-heterogeneous gap is necessary but not sufficient: an axis ships in the benchmark only on a task meeting the coupling-validity criterion (ADR-026), under the CHAMBER-Bench v1.0 protocol (ADR-027). The historical staged spike protocol (Stage 1: AS + OM, Stage 2: CR + CM, Stage 3: PF + SA) is ADR-007.
Who this is for
Multi-robot RL researchers — CHAMBER scores ad-hoc teamwork at the manipulation tier with verifiable, preregistered result bundles. Start with the five-minute quickstart.
Safe-control researchers — CONCERTO's safety module is a reference implementation of the exponential-CBF + conformal + OSCBF stack with a hard braking fallback. The unresolved theoretical question (average-loss → per-step bound) is documented in ADR-004.
Robotics practitioners and integrators — CHAMBER's communication profiles are anchored to 3GPP Release-17 URLLC and 5G-TSN industrial-trial data, and the safety axis references ISO 10218-2:2025 directly. See the threat model.
Non-goals
CHAMBER is not a navigation, planning, or generic reinforcement- learning benchmark; the four-aspect intersection requires contact-rich physical manipulation. CONCERTO is not a certified safety product — it is a research-grade reference implementation and is not a substitute for safety engineering in production deployments.
Roadmap
The project advances in phases. Phase 0 locked the design contract (the ADRs) and ran the staged heterogeneity-axis spikes; its outcome was the coupling-validity reinterpretation (ADR-026). Phase 1 shipped the CHAMBER-Bench v1.0 protocol (ADR-027), the admission campaign, the versioned partner zoo, and the populated per-task leaderboard. Ahead: heterogeneity axes that survive coupling-valid operationalization (control rate and communication are the next candidates), additional Tier-2 task admissions from the Tier-3 candidate list, and — later — a real-robot demonstration platform.
Day-to-day progress: the CHANGELOG and the issues board.