Edge Case

Edge
Case

Continual learning to turn rookie robots into vets.First lesson: touch.

The problem

Robots stop learning when they ship.

The real world doesn't stop changing.

Train Ship fibra en degeneraciónNot deployed
a wet carton a new gripper a surface not in the dataset
Months without learning · every edge case fails the same way, forever
Why touch

Most physical tasks are force-aware. Without it, robots become unreliable and unreliable robots don't get deployed. Adding touch today means buying tactile data that does not include all possible permutations and retraining a model the team spent months stabilizing.

Inspired by biology

Your body doesn't send everything to your brain.

1,000,000,000
10 Bits per second the brain runs on
Bits per second the body takes in · skin alone sends about 1,000,000
The gap · almost all of it is handled before the cortex is asked
Our view

Physical AI should be as efficient as the human brain. Edge Case sits where the periphery sits, and learns the way memory does.

Source: Zheng & Meister, 2024
piel
Our solution

Nature has deployed more embodied intelligence than anyone ever will.

We're just learning from it.

Edge Case
raíz posterior g. raíz anterior m. médula espinal, corte transversal Edge Case · peripheral layer To brain · only when it matters In · tactile Out · reflex, ms