Effects of Human-Like Characteristics in Sampling-Based Motion Planning on the Legibility of Robot Arm Motions

Abstract

Conveying the intended goal of a robot arm motion has been shown to increase the quality of human–robot collaboration drastically. To this end, optimization-based approaches have been proposed that optimize the legibility of a robot’s motion. However, they are limited in two ways. First, they are typically not validated in environments with obstacles and narrow passages that require collision-free motion planning. Second, they do not consider the influence of the anthropomorphization process that might be caused by a human-like motion or appearance of the arm. This leads to the question of to what extent the legibility of motions is influenced by these factors. In this work, we study the influence of our previously proposed human-likeness function on the legibility of robot arm motions in the context of sampling-based motion planning. We evaluate it against three other motions: a functional motion, a recorded expert motion, and a legible motion based on a heuristic for the observer’s prediction. For this, we conduct an extensive user study with 94 participants. In contrast to other works, we manipulate the robot’s appearance and the complexity of the environment. We thus provide insights into how the legibility of robot motions is influenced by human-like characteristics in motion, appearance and restricting workspace conditions. The complete stimulus material, raw data and all evaluation scripts used in this work are provided at https://mytuc.org/zpvt .