Platform

MIM. The Motion Intelligence Model.

A foundation model for physical motion. It lets any machine know where it is, where everything around it is, and what is about to happen. In real time, on the edge.

±2 CM GROUND TRUTH, TOP THREE AUTO MANUFACTURERS · 130/135 MINES MARKED, U.S. ARMY

Six capabilities

One model. Six capabilities.

Positioning is the entry point. The same model then predicts, detects, coordinates, simulates, and reasons.

01Lo
Localize±10 cm with no GPS and no pre-built map. ±2 cm with higher-grade sensing.U.S. ARMY · TOP THREE AUTO MANUFACTURERS
02Pr
PredictAnticipates where things will be, not just where they are, from learned motion patterns.
03De
DetectA vehicle off pattern, a machine failing, a person where no person should be.
04Co
CoordinateConvoys hold formation. Swarms divide work. One operator directs many bodies.DARPA · ~100-VEHICLE FORMATIONS
05Si
SimulateA live digital twin of everything in motion, built from the mesh itself.
06Re
ReasonUnderstands motion well enough to answer questions about it, and act on the answers.
See how the mesh computes position

Where it runs

Five ways to lose GPS. One answer.

Every environment defeats satellites differently. Ranging has held in every one deployed so far.

Environment Why GPS fails there What the mesh does
Logistics campusesPACCAR Metal containers and trailers reflect the signal. Position jumps between rows. ~10 cm
Ports and minesBHP · MIOS Pit walls block half the sky. Airborne dust degrades what is left. Collision avoidance, degraded GNSS
Airports and transitNJ TRANSIT Underground there is no sky at all. GPS does not degrade, it stops. Station anchors only
Smart citiesU.S. DOT Buildings bounce the signal down the street. Raw GPS was wrong by eight meters. 8 m → 10 cm
Data centersNEXT Indoors, with dense metal racks. No satellite reaches the floor. Not yet deployed

Same model in every deployment · no product fork

See the deployments

Why defensible

Four reasons this is hard to copy.

The corner no one else is in.

200 Hz on a sub-$50 processor. No GPU.

Every benchmark in the record was achieved on commodity silicon. Buying the same accuracy and reflex rate with GPU-class hardware cannot reach the bill of materials a truck, a railcar or a haul truck can carry.

EVERY DEPLOYMENT
A city corridor in about two hours.

Underground rail carried by anchors in the stations alone. Cost scales with stations, not miles of track.

U.S. DOT · NJ TRANSIT
1,800 meters, through obstructions.

Peers range to each other without line of sight, which is why the accuracy survives jamming, tunnels and mine pits. Army convoys degraded gracefully under active RF jamming.

PEER RANGING · U.S. ARMY
Cancer cells, same company.

The company’s perception and inference work was applied to automating cervical cytology for Hologic, working with Google. What it does is not confined to vehicles.

HOLOGIC

See what your machines are missing.