RobotGov — physical AI authorization

Motion authority.
Not conversations.

RobotGov extends the same deterministic control-plane pattern to embodied AI: mobile robots, manipulators, humanoids, UAVs, and UGVs. Physical action envelopes are evaluated before motion authority is granted — deterministically, with no LLM in the enforcement path.

74,934
Authorization rows evaluated
89.71%
Row allow rate
10.29%
HOLD / CLAMP gate rate
0
LLM calls in enforcement path
Request evaluation View replay evidence
robotgov — live authorization replay ALLOW
Robot
Human
CLAMP zone
HOLD zone
Profile: collaborative_lab
Operating profile matrix

Three profiles. One control plane.

Each operating profile defines the physical envelope parameters for a deployment context. Profile selection is a policy decision, not a runtime judgment call.

industrial_strict
1.2m
Minimum human distance. Warehouse, factory, and high-throughput motion envelopes.
HSR 0.8m/s · Jackal 1.2m/s
collaborative_lab
0.5m
Minimum human distance. Lower-speed human-robot proximity in lab or controlled collaboration settings.
HSR 0.6m/s · Jackal 0.8m/s
handover_mode
0.35m
Minimum human distance. Explicit close interaction with tighter speed authority for handover tasks.
HSR 0.4m/s · Jackal 0.5m/s
Profile Purpose Min human dist HSR max speed Jackal max speed
industrial_strict Warehouse, factory, high-throughput motion 1.2m 0.8m/s 1.2m/s
collaborative_lab Human-robot proximity, lab or controlled settings 0.5m 0.6m/s 0.8m/s
handover_mode Explicit close interaction, handover tasks 0.35m 0.4m/s 0.5m/s
Dataset-backed replay

Public traces. Deterministic envelopes.

We ran 10 public NavWareSet social-navigation pose traces through three operating profiles. 74,934 authorization windows evaluated. Every verdict is reproducible from the same inputs.

robotgov_replay_summary.json
datasetNavWareSet social-nav traces
scenes10
profile_cases30
authorization_rows74,934
row_allow_rate0.8971
row_gate_rate0.1029
verdict_mixHOLD 21 / CLAMP 9
expected_capture_rate1.0
llm_calls_in_path0
report SHA-256
c490751a590ddc851399c4640c6744f0d7c8ce075d6f5373c9773d488f0fcacc
Verdict breakdown
67,174
ALLOW rows
89.71%
HOLD × 21
Hard boundary enforced
Motion authority denied
CLAMP × 9
Speed reduction triggered
Kinematic gate
1.0
Expected capture rate
All interventions caught
What HOLD and CLAMP mean:
CLAMP — speed reduced to profile limit, motion continues under constraint
HOLD — hard boundary enforced, motion authority denied until distance clears
ALLOW — envelope within policy, motion authority granted without constraint
Action envelope schema

What goes into an authorization window.

Every authorization row is a structured physical action envelope. Policy evaluates it deterministically — no natural language, no model judgment.

physical_action_envelope.json — sample window
robot_id: jackal_03
profile: collaborative_lab
timestamp_ns: 1748901234000000
motion_state
speed_ms: 0.72
heading_deg: 214.3
mode: autonomous_nav
proximity
nearest_human_m: 0.48
humans_in_2m: 2
verdict
result: HOLD
reason: PROXIMITY_WITHIN_HOLD_ZONE
speed_limit_ms: 0.6
audit_hash: b3ac30bc…
profile
Active operating profile. Determines distance thresholds and speed limits. Set by policy, not by the robot at runtime.
industrial_strict · collaborative_lab · handover_mode
nearest_human_m
Euclidean distance to nearest detected human in the scene. Primary input to proximity-based authorization decisions.
float — derived from pose trace, not from robot self-report
speed_ms
Current robot speed in m/s. Compared against profile speed limit for HOLD evaluation.
float — compared against profile.max_speed
verdict + reason + audit_hash
Deterministic output. Same inputs, same profile → same verdict, same hash, every evaluation.
ALLOW · HOLD · CLAMP — no LLM in path
Claim boundary
This preview uses public NavWareSet social-navigation pose traces. It is proximity-boundary replay evidence — not functional-safety certification, flight certification, production deployment approval, an incident log, or a claim that RobotGov is sufficient for any specific deployment context.

Profile parameters, distance thresholds, and speed limits must be validated for your specific robot platform, operating environment, and regulatory context before production deployment.
Request a Physical AI evaluation
Send sanitized clean and hazardous telemetry. We map it into action envelopes, replay under your operating profiles, and return a SHA-anchored report showing where motion would be allowed, clamped, held, or escalated.
Request evaluation Back to Ark Sovereign