Velocity-Based Variable Thresholds for Improving Collision Detection in Manipulators

In absence of tactile and visual sensors, manipulators rely on dynamic modeling and force/torque sensors to detect collisions. Predicated upon the understanding that collision detection will be improved by removing the effects of torque sensor error and friction compensation error, we propose an approach to measure these undesired effects in external torque residual signals without resorting to remodeling of the robot. Compensation of these effects leads to the proposed velocity-based variable thresholds which are shown to detect collisions with better efficacy than uncompensated methods. Implementation of the proposed velocity-based variable thresholds on a torque sensor equipped robot validates our methodology.