Autonomy that regulators can certify and engineers can debug.
A physics-grounded driving stack that replaces end-to-end neural nets with partial differential equations, HD maps, and a regulatory-grade safety supervisor. Same input, same output, every time, auditable line by line, small enough to run on a Raspberry Pi.
Neural-network autonomy cannot be certified.
Every major AV company has hit the same wall: regulators, OEMs, and insurers cannot approve a system whose decisions cannot be reproduced, explained, or bounded. This is why L3+ deployment has stalled.
Same scene, different decision.
Stochastic inference means the same frame can produce different outputs. Safety cases require determinism, you cannot certify variance.
You can't trace why it braked.
A trained model is a black box. When something goes wrong, the investigation ends at "the network said so." That answer will not survive an NTSB hearing.
No operational design domain.
Networks degrade silently outside their training distribution. Without an explicit ODD, you cannot define where the system is safe, and regulators demand you define it.
Build autonomy out of math that already satisfies a safety case.
Partial differential equations have a century of certification precedent in aerospace, nuclear, and automotive. They are reproducible, explainable, and bounded by construction. We use them to drive a car.
One deterministic pipeline.
Perception to actuation.
Five stages, all auditable, all deterministic. Every decision in the pipeline is a function of its inputs, no hidden state, no learned bias, no inference variance.
Perception
Camera + YOLO-class detector + Kalman tracker → object tracks with velocity.
HD Map
Lanelet2 / OSM map, Frenet projection, route planning with lane-change cost.
Prediction
1.5 s motion horizon. Map-aware lane-following, CV/CTRV kinematic fallback.
PDE Field
Continuity + velocity + potential PDEs on 256×64 grid. 0.2 ms solve.
Control
Pure-pursuit + field-blend steering, speed PID, SocketCAN at 20 Hz.
A real Minimum-Risk Maneuver, not a panic brake.
When perception drops, the ODD is violated, or a collision becomes imminent, the FieldSpace MRM supervisor executes a regulatory-grade controlled stop: comfort-decelerate, hold lane, drift to shoulder, park. UNECE-R157 / ISO 22737 shaped.
What ships today. What comes next. No hype.
We do not sell a self-driving car. We sell the deterministic decision layer that makes one certifiable, plus the extensions to the L4 ODDs where the unit economics work today.
L2+ hands-on assist
L3/L4-ready architecture. Production-grade software, pilot-ready for bounded ODDs.
- ✓Deterministic PDE traffic field
- ✓HD-map localization (Lanelet2 / OSM)
- ✓Short-horizon motion prediction
- ✓Minimum-Risk Maneuver supervisor
- ✓ODD geofence enforcement
- ✓Runs on Raspberry Pi 4 at 20 Hz
L4 in bounded ODDs
Pilot-funded. Closed-campus, ports, yards, defense off-road.
- →Radar + camera sensor fusion
- →Redundant compute (Pi + Jetson)
- →Fleet telemetry + remote assist
- →ODD survey tooling
- →ISO 21448 SOTIF evidence package
Tier-1 OEM / L3 highway
OEM-funded. Certification program alongside integrator.
- ○ISO 26262 ASIL-D evidence
- ○UNECE R157 submission (L3 traffic jam)
- ○NHTSA exemption petition
- ○Per-vehicle royalty production license
- ○Multi-million mile validation
L5 unsupervised autonomy does not exist in production anywhere on earth. Any vendor who claims it is selling you a slide deck. We ship code.
Safety Suite v1 results.
Five safety-critical scenarios run in CARLA with synthetic ground truth. Every scenario: earlier hazard detection, zero false positives, all braking margins met.
| SCENARIO | BASELINE LEAD | FIELDSPACE LEAD | GAIN |
|---|---|---|---|
| Falling Debris | -0.50s | +0.20s | +0.70s |
| Sudden Cut-In | -0.10s | +0.20s | +0.30s |
| Occluded Pedestrian | -0.30s | +0.20s | +0.50s |
| Stopped Vehicle | -0.40s | +0.20s | +0.60s |
| Sliding Cargo | -0.80s | +0.20s | +1.00s |
Three deployment paths where the economics already work.
Closed-campus autonomy
Warehouse yards, ports, airport ground ops. No FMVSS, no NHTSA, no public road. Deploy in 6-9 months with your current sensor set.
Defense off-road
No NHTSA gate. ITAR-friendly. Deterministic, auditable code is the only thing a safety board will clear for an unmanned ground vehicle.
Tier-1 OEM supplier
License the stack to an OEM who brings the certification budget. You are the algorithm provider; they are the integrator. Per-vehicle royalty.
Two ways to deploy.
Monitor in parallel with zero control authority, or run the full deterministic pipeline. REST + WebSocket API or ROS 2 native.
Safety Observer
Runs alongside your existing stack. Zero control authority. Ingests fused objects, outputs hazard alerts with full explainability traces.
End-to-end deterministic pipeline
Perception through actuation. PDE field, prediction, MRM, pure-pursuit + PID. 20 Hz on a Pi. SocketCAN out.
The industry hit a wall.
We have the ladder.
Deterministic. Explainable. Certifiable. The only autonomy stack a safety board can approve and a regulator can audit.