Physical Intelligence · Ground
Every road, reasoned.
Jean evaluated 5,490 live route nodes across a real Oakland road network with zero LLM calls in the cognitive path — pure deterministic reasoning, no model trained on driving demonstrations.
The result
Navigation without a learned policy.
Most autonomous navigation leans on models trained over millions of driving miles. Jean reasons over the road graph directly — evaluating routes as a deterministic search, not a probability drawn from training.
"A model trained on driving data can tell you what usually works. Deterministic reasoning over the actual road graph tells you what is true for this route, right now — and can show its work."
How it reasons
The road graph, evaluated directly.
Deterministic navigation over live network data
| Property | Trained navigation models | Jean |
|---|---|---|
| Route decision | Probability from training | Deterministic graph search |
| Explainability | Opaque — a learned weight | Every node evaluation traceable |
| Cognitive-path LLM calls | Often required | Zero |
| Training data required | Millions of miles | None |
5,490 nodes across a real Oakland road network were evaluated in the cognitive path with no LLM inference — the reasoning is the search, not a call to a model. This is corroborating evidence for the same claim the arm and the ROV make: one architecture, many substrates.
Why it matters
The same mind that holds the deep finds the road.
Ground navigation isn't a separate product with a separate model. It's the same reasoning architecture that threads a needle with a robotic arm and holds a vehicle steady underwater — pointed at a road graph instead of a workspace or a water column.
Who this is for
Built for ground autonomy that has to explain itself.
Fleet & Logistics
Routing you can audit
Every route decision is a traceable evaluation over the real road graph — not a probability from a model you can't inspect. When a route is chosen, you can see why.
Defense & Ground Systems
No training-data dependency
Deterministic navigation reasons over the network it's given — no need to have driven the terrain a million times first. Critical where the map is new and the data is thin.
Researchers
Reasoning, not black boxes
Zero LLM calls in the cognitive path means the navigation logic is inspectable end to end — a deterministic search you can audit, not a weight you have to trust.
Investors
A third domain, same engine
Arm, ocean, road — three physical domains from one architecture and one small team. Each new domain reusing the last is the whole thesis, made concrete.
Start a conversation
The navigation methodology is available.
The navigation approach, the Oakland run data, and the deterministic reasoning path are documented. If you want to verify the method or talk about your ground-autonomy problem, write to us.
Write to us directly
Tell us your vehicle and your environment. We respond personally — no sales process, no auto-responder.
Write to us →contactus@myasolutions.org · full phase records available on request