RobotGov
RobotGov qualifies physical-AI evidence, binds authority to exact proposed actions, and requires downstream verification before a bounded simulator consequence is recorded. It preserves the difference between observation, permission, application, and independently observed outcome.
Replay Boundary
This page describes bounded review and Gazebo sandbox evidence. It is not physical robot control, production isolation, functional-safety certification, cybersecurity certification, deployment approval, or an incident-causality report.
Partner-Reviewed Panda Pilot
An existing ROS 2 Jazzy, MoveIt 2.12.4, and Gazebo Harmonic Panda workflow was integrated with separate proposal and authority processes, current pre-state evidence, one-shot stage authority, a private receiver, and independent final-state observation.
| Exact authorized workflow | WORKFLOW_COMPLETED |
| Unsigned access | UNSIGNED_ACCESS_REJECTED |
| Material mutation | MUTATION_REJECTED |
| Replay | REPLAY_REJECTED_NO_SECOND_EFFECT |
| Receiver outage | RECEIVER_OUTAGE_NO_RECEIPT |
| Observer outage | OBSERVER_OUTAGE_NO_COMPLETION |
Profile Matrix
| Profile | Purpose | Min Human Distance | HSR Max Speed | Jackal Max Speed |
|---|---|---|---|---|
industrial_strict | Warehouse, factory, and high-throughput motion envelopes. | 1.2m | 0.8m/s | 1.2m/s |
collaborative_lab | Lower-speed human-robot proximity in lab or controlled collaboration settings. | 0.5m | 0.6m/s | 0.8m/s |
handover_mode | Explicit close interaction with tighter speed authority. | 0.35m | 0.4m/s | 0.5m/s |
Dataset-Backed Summary
| Dataset | NavWareSet social-navigation pose traces |
| Scenes | 10 |
| Profile cases | 30 |
| Observed replay rows | 74,934 |
| Verdict mix | HOLD 21 / CLAMP 9 |
| Row-level allow rate | 0.8971 |
| Row-level gate rate | 0.1029 |
| Expected intervention capture rate | 1.0 |
| LLM calls in enforcement path | 0 |
| Report SHA-256 | c490751a590ddc851399c4640c6744f0d7c8ce075d6f5373c9773d488f0fcacc |
Phase 0 Fit Screen
For a qualified physical-AI evaluation, send sanitized clean and hazardous telemetry. We map it into action envelopes, replay it under your operating profiles, and return a SHA-anchored report showing where actions would be allowed, clamped, held, or escalated for human review.