Physical AI · the authority layer · in research
Verification at the edge. Before the actuator moves.
For robots, cobots, AGVs, and drones. The design: classify a structured agent goal, mint a sub-second Authority Token bound to robot identity, verify locally at the edge, fall back safely if anything is wrong — layered on top of your existing safety PLC, not in place of it. The edge verifier (Rust + TypeScript) and offline geofence checks are built and conformance-tested; this work is in research and has not been deployed on a live robot fleet. We’re looking for design partners.
02 · What we own / what you own
We sit above the PLC. You keep the safety stack.
The authority decision
Intent classification, conformance scoring, policy evaluation, token mint, audit chain — engineered for a sub-second authority decision in the control loop.
The safety envelope
Your existing safety PLC, e-stops, light curtains, and certified motion limits remain authoritative. We never override a safety stop. We refuse to authorize anything outside the declared envelope.
We are not the safety system
We don’t replace functional safety. We don’t certify your hardware. We don’t make joint-level motion decisions. We govern at the goal/task level.
One audit chain per fleet
Every goal accepted or refused — across every robot, AGV, drone, and cobot — is hash-linked into the same immutable audit chain. Replayable in slow time after the fact.
03 · How it’s designed to ship
The edge verifier verifies its RS256-signed edge-authority token fully offline (cached tenant JWKS, zero network on the hot path) and enforces an offline signed geofence. This is research/reference-grade today — it has not been deployed on a live fleet.