Ensuring Force Safety in Vision-Guided Robotic Manipulation via Implicit Tactile Calibration
The model uses tactile feedback to recalibrate future states and reduce harmful force during door opening.
Real-world adaptation used 20% of 110 demonstrations and was tested on unseen doors and disturbances.
Replicate across drawers, windows, assembly, and dexterous hands; measure peak force, damage rate, recovery, and hardware wear over long runs.
The model uses tactile feedback to recalibrate future states and reduce harmful force during door opening.
Abstract; p. 8, ConclusionsReal-world adaptation used 20% of 110 demonstrations and was tested on unseen doors and disturbances.
p. 8, Q1-Q3Experiments cover only door opening because data collection is costly.
The setup uses a gripper rather than a dexterous hand.
No general safety certification, long-run hardware wear, or multi-task deployment evidence is provided.
Force-aware control reduces one technical risk but does not establish general humanoid safety or economics.
Thematic