How robots learn to hold fragile cups without crushing them

Holding a full plastic cup requires exquisite force control—too little and it slips, too much and it crumples irreversibly. TactileReflex solves this by calibrating grip reflexes directly from sensor noise characteristics (a 30-second static hold), eliminating manual tuning and material-specific models. Three parallel feedback loops—slip suppression, weight adaptation, and force limits—run at 12 Hz on vision-tactile sensor data. On dynamic pouring, the system succeeds 9/10 times while baseline fixed-force grippers fail completely.