Whole Body Human-Robot Collision Detection Using Base-sensor Neuroadaptive Interaction

Abstract

Conventional methods to detect collisions or physical interaction between robots and human users and/or the environment consist of torque sensing at joints. In the case of non-collaborative robots, collision detection can be accomplished by wrist Force-Torque sensing at the end-effector, or covering the robot with pressure sensitive skin sensors. In this paper we present a novel approach to detect whole-body collisions with a robot manipulator equipped with a base force-torque sensor (BFTS), instead of a wrist force-torque sensor (WFTS). Our approach is summarized in the Base-sensor Assisted Physical Interaction (BAPI) controller described here. Although several other studies have investigated the advantages of this sensing configuration in conjunction with classical model-based computed torque controllers, here we make use of a Neuro-Adaptive controller (NAC) that can estimate the robot dynamic parameters on-line, for high performance interaction. The NAC requires no prior physical knowledge of the robot model parameters, and it offers Lyapunov stability and tracking performance guarantees. We offer the theoretical basis of the BAPI control algorithm and present experimental results with a 6 degrees of freedom (DOF) robot arm demonstrating the effectiveness of our approach.