Effect of efficiently computed explanations for robot motion planning failures in human–robot interaction

Abstract

Transparent interaction between an operator and a robotic system is essential for successful task completion. This requires a mutual understanding of decisions and processes in order to provide accurate diagnoses and troubleshooting suggestions in the event of an error. In the motion planning domain, a deep understanding of the decisions made by the system is essential for the successful navigation of a robot in dynamic environments. Due to inaccuracies in the environment perception or in the configuration of the motion planner, robot motion planning incidents can occur that are difficult for the operator to understand. In this work, we present a method that is able to quickly provide explanations for motion planning failures. In the context of optimization-based planners, failures related to planning constraints can be identified using an adaptation of the diagnosis algorithm FastDiag. It is able to provide a preferred minimal diagnosis in logarithmic time, even for large sets of constraints. To evaluate the potential of the proposed method, we conduct a user study to investigate the impact of the provided explanations of motion planning failures on the operator’s performance, trust, and workload. The results show that quickly providing additional explanations for failed motion planning improves task completion time, overall trust in the system, and reduces the number of interactions required. However, no effect was found on perceived workload.