Physics Guided Machine Learning Approach to Safe Quasi-Static Impact Situations in Human-Robot Collaboration Following the Power and Force Limiting Method of the ISO/TS 15066 Standard

Abstract

Abstract Following the performance and force limitation method of the ISO/TS 15066 standard, human safety is evaluated in a quasi-static impact situation in a human-robot collaboration task. To achieve this, all potentially critical impact situations are first identified and then the corresponding impact forces and pressures are experimentally measured and compared with their limit values specified by ISO/TS 15066. If a collaborative application is evaluated as safe, once something is changed in the collaborative workspace, to obtain certainty again, one has to re-assess the safety properties and to re-measure the corresponding forces and pressures. This measurement-based safety assessment severely limits the flexibility of a collaborative process. In this paper, as a method for overcoming the flexibility problem, a physics guided machine learning ensemble for prediction of peak impact forces, within predefined modification dimensions of the collaborative application, is proposed. The input features for the ensemble are obtained using a simplified mathematical model of an impact between a robot and a human. Based on experimentally measured peak impact forces performed with the Universal Robots UR10e, peak impact forces covering an almost whole range of data distribution of interest defined by modification dimensions are obtained using interpolation. As an example of the presented methodology, a generic pick and place task with two modification dimensions is considered. As the most important application of the presented methodology, a map showing the maximal safe impact velocity for a grid of points in a part of a collaborative workspace of interest is shown.