Minimum Distance Calculation Using Approximate Convex Decomposition for Safe Human-Robot Interaction

Abstract

In human–robot interaction (HRI), maintaining a safe distance is the primary requirement. The flexibility and safety of HRI are often limited by the efficiency and accuracy of distance algorithms. This study proposes a fast and accurate method based on the Gilbert-Johnson-Keerthi (GJK) algorithm by optimizing the geometric representations of both humans and robots. To model the human accurately and stably, its point cloud is captured by a depth camera and decomposed into convex polyhedra using the approximate convex decomposition method with a novel area concavity. Meanwhile, the support function of the capsule is optimized to represent the robot more efficiently. A comprehensive metric is designed to jointly evaluate both the efficiency and accuracy of the algorithms. Experiments demonstrate that the area concavity remains stable under the perspective rule. Moreover, the proposed minimum distance algorithm exhibits optimal overall performance in both efficiency and accuracy while effectively preventing false alarms.