Robots can plan and repeat a trajectory. The hard part is adapting the final millimetres of contact, as friction, pose and compliance change. Senxe targets that bounded sensorimotor layer.
Cultured neurons — living cells, not a network named after them — propose; a deterministic safety layer decides. The biological side never holds hardware authority — that is a property of the architecture, not a setting.
[ ↗ ARCHITECTURE BRIEF ]Official Unitree Isaac Lab scene, base-fixed G1 + Dex3. Two deterministic runs produced the same result: all five phases completed, and a 100 g cylinder lifted 50.583 mm. 182 automated tests pass alongside it.
What the scene reports is contact sensing and simulator object pose — not calibrated physical force, and not a learned detector.
Isaac integration evidence
Cultured neurons read the contact state and return a candidate correction. Stochastic, unverified, never wired straight to the actuators.
Every proposal is accepted, clamped, or discarded against fixed limits. This side holds all hardware authority until the living side beats matched controls.
A full pre-hardware stack — natural-language task, planner, neural interface contract, robot boundary — running and tested end to end, with no institutional support so far.
The next phase puts living neurons on a multi-electrode array, and that is not a phase to run alone. I am looking for an academic sponsor — a PI or lab to oversee biosafety, culture workflow and experimental design — and for shared instrument access ahead of any purchase. Introductions to neuroscience, electrophysiology or robotics groups are just as useful.
[ ↗ [email protected] ]© 2026 Senxe