VisualForce uses gripper deformation to guide robot actions without force sensors at deployment

The VisualForce team says it trains a model to estimate force from a soft gripper’s visible deformation, then samples 32 action–force pairs and chooses the action whose predicted force is closest to a target from demonstrations. In 15 trials per task and method, it reports 73.3% success on plug insertion versus 26.7% for a baseline that executes one sampled action. The system needs no force or tactile sensors at deployment.