Automotive · Frontier

Teach-the-Truck — Self-Discovered Vehicle State Axes

Cycle the liftgate a few times — the truck watches it forever.

Frontier Frontier

Plausible physics, early work — shown honestly as an experiment.

Every mechanical state — liftgate up, ramp out, spare wheel present, one door ajar — bends the signal in its own repeatable way. Cycling the state a few times teaches the system exactly which bend belongs to it. From then on, the mesh reports that state with no new sensor ever installed.

In the physics: Every mechanical configuration change moves conductive/dielectric structure and thus deterministically reshapes the multipath fingerprint (specific CIR taps appear/vanish, amplitudes shift). Controlled repetition isolates the covariance direction attributable to that state from ambient variation.
Preparing the recorded fieldLoading real LatentField pipeline output…

This is real pipeline output for Automotive, replayed. The exact reading depends on your room.

Bring this into the real roomCabin Presence Node1 sensing nodes · ~$40 parts · about 45 min
Go deeper

How the field becomes an answer

Axis discovery via poke-and-observe (repeat state toggles) supervised contrast on CIR + CSI features to extract the state's discriminant axis project live data onto learned axes per-axis CUSUM for silent state changes (e.g., liftgate creeping open on highway).

How we prove it

Own-hardware: van with 6 togglable states, 10 teach cycles each, measure held-out state-classification accuracy and teach-time; CSI-Bench task structure as protocol template for repeatability.

Who it is for

A programmable sensor: every fleet defines its own alarms without new hardware or firmware — sold to fleet telematics platform PMs as an extensible sensing SDK, not a fixed-function device.

Complete first-room estimateCabin Presence Node1 nodes · ~$40 parts · 45 min Request kit, subscribe, or set up →
Compatible DIY hardware options
Prosumer DIY$18–35

45 min · intermediate

← All Automotive capabilities