Collision Mass Map for Safe and Efficient Human-Robot Interaction

Abstract

Efficient and safe integration of robots into human workspaces remains a significant challenge. The ISO 10218-2 standard defines permissible force thresholds that a robot is allowed to exert on humans, along with a simple model to estimate the impact force based on the impact velocity, the involved human body part, and the effective mass of the robot. In this work, we experimentally demonstrate that state-of-the-art approaches fail to compute the effective robot mass accurately, leading to unsafe or overly-restrictive robot behavior. We address this shortcoming by presenting a data-driven collision mass map that accurately predicts the effective mass perceived at the end effector for a given collision location for the entire workspace. These maps are trained using a limited set of impact data selected by our proposed measurement procedure and can serve as valuable references for safety-critical applications. We validate our method on two robots, demonstrating accurate force predictions in compliance with ISO 10218-2. In our experiments, we show that our approach greatly reduces the required force measurements compared to state-of-the-art data-driven methods for risk assessment. Furthermore, our approach allows one to easily integrate different payloads, making it highly adaptable to various collaborative tasks. The proposed collision mass map can be standardized and deployed for any collaborative robot, enabling simple integration of robots for safe and more efficient human-robot interaction.