Mobile Robot Fault Detection based on Redundant Information Statistics

Abstract

Detecting and reacting to faults (i.e., abnormal situations) are essential skills for robots to safely and autonomously perform tasks in human-populated environments. This paper presents a fault detection algorithm that statistically monitors robot motion execution. The algorithm does not model possible motion faults, but it instead uses a model of normal execution to detect anomalies. The model of normal execution is based on comparisons between redundant sources of information; specifically, wheel encoder readings and localization algorithm output are used as the redundant sources of displacement information. The algorithm was implemented on a service robot that often navigates in a human-populated environment without supervision. Experimental results show that the algorithm can detect even minor motion faults and stop execution immediately to guarantee safety to the humans around the robot.