Risk Sensitive Particle Filters

Paper at NIPS 2001 by Sebastian Thrun, John Langford, and Vandi Verma

Presented by Vandi Verma

Abstract

We propose a particle filter that incorporates a model of cost when generating particles. The approach is motivated by the observation that the cost of accidentally not tracking hypotheses might be significant in some areas of state space, and irrelevant in others. By incorporating a cost model into particle filtering, states that are more critical to the system performance are more likely to be tracked. Automatic calculation of the cost model is implemented using an MDP value function calculation that estimates the value of tracking a particular state. Experiments in two mobile robot domains illustrate the appropriateness of the approach.

Back to the Main Page