Unknown Fault Detection for Mobile Robots Based on Particle Filters

Abstract

An improved particle filter was presented to simultaneously detect unknown faults and diagnose known faults for mobile robots. Firstly, the kinematics and fault models of the monitored mobile robot were given. Secondly, two parameters were extracted from sample-based expression for a posteriori probability density: sum of sample weights, and reliability of a posteriori belief state. These features were used to detect whether the estimation given by particle filter was believable or not. Unbelievable estimation indicates that the true state was not in the current state space, i.e. it is a novel state (or a unknown fault). This method preserves the advantages of particle filters and can diagnose known faults as well as detect unknown fault. The method is testified on a real mobile robot.