Non-Parametric Fault Identification for Space Rovers

Abstract

Autonomous fault detection is prerequisite for autonomous system repair which is of great value for spacecraft where human intervention is expensive, slow, unreliable, and occasionally impossible. We present a method which successfully achieves autonomous fault detection in simulation. The approach uses a nonparametric estimate of the system state updated based upon sensor measurements. Our system does state estimation using a decision-theoretic generalization of particle filters which takes into account the difference in utility of fault detection vs. fault nondetection.