A Boosted Decision Tree Approach for a Safe Human-Robot Collaboration in Quasi-static Impact Situations

Abstract

According to the ISO/TS 15066, human safety in quasi-static impact situations in human-robot collaboration is assessed first by identifying all high-risk impact situations and then by measuring maximal and steady-state values the impact force and pressure at these possibly critical situations. This means that if something is changed in a collaborative application, the ISO/TS 15066 requires that the risk analysis and the force measurements must be redone, which severely limits the flexibility of a robotic system. In this paper, a physics guided boosted decision tree is proposed as a tool to assess human safety. The basic hypothesis is that a physics guided boosted decision tree can be trained to estimates the peak impact force for a given impact velocity, robot configuration, an impact point on the robot and a human body part. Based on experimental measurements done with the Universal Robots UR10e and on a simple mathematical model of an impact between a point on a robot and a point on a human body part, a feature vector is generated as an input to the boosted decision tree. After the training using Matlab’s Least-squares boosting algorithm, the boosted decision tree can predict the measured peak impact force with a relative error of less than 9% thus supporting the basic hypothesis. However, the predictions of the trained boosted decision tree are valid only for the case of a quasi-static impact in a vertical direction between a robot’s end-effector and a back of human’s non-dominant hand.