Efficient Monitoring for Planetary Rovers

Abstract

Planetary rovers operate in environments where human intervention is expensive, slow, unreliable, or impossible. It is therefore essential to monitor the behavior of these robots so that contingencies may be addressed before they result in catastrophic failures. This monitoring needs to be efficient since there is limited computational power available on rovers. We propose an efficient particle filter for monitoring faults that combines the Unscented Kalman Filter (UKF) and the Variable Resolution Particle Filter (VRPF). We use the UKF to obtain an improved proposal distribution that takes into account the predictive likelihood. This requires computing a UKF for every possible transition to a fault or nominal state at each instance in time. Particles are then generated from the state transition matrix weighted by the predictive likelihood. Since there are potentially a large number of faults that may occur at any instance, this approach is not very scalable. We use the VRPF to address this concern. The VRPF introduced the notion of abstract states that may represent sets of states. There are fewer transitions between states when they are repented in abstraction. We show that the VRPF in conjunction with a UKF proposal improves performance and may potentially be used in large state spaces. Experimental results show a significant improvement in efficiency.