Leo

"clean this" while pointing at a spill is the demo. the interesting part is everything that has to happen for it to work. the robot has to locate your arm, project where it's aimed, and match that ray against objects it already holds in memory. speech is the easy half. grounding is the part that kept voice assistants stuck on context free commands for a decade. Matic runs all of it onboard: > 5 RGB-IR cameras, no lidar > an NVIDIA Jetson doing the inference locally > a 3D map of the home that updates while it cleans > raw footage discarded in real time, nothing leaves the house 75 languages, no app, no account. calling this voice control undersells it. it's a machine that can see what you're pointing at.