ADDRESSING COMPLEXITY ISSUES IN A REAL-TIME PARTICLE FILTER FOR ROBOT LOCALIZATION

Abstract

Exploiting a particle filter for robot localization requires expensive filter computations to be performed at the rate of incoming sensor data. These high computational requirements prevent exploitation of advanced localization techniques in many robot navigation settings. The Real-Time Particle Filter (RTPF) provides a tradeoff between sensor management and filter performance by adopting a mixture representation for the set of samples. In this paper, we propose two main improvements in the design of a RTPF for robot localization. First, we describe a novel solution for computing mixture parameters relying on the notion of effective sample size. Second, we illustrate a library for RTPF design based on generic programming and providing both flexibility in the customization of RTPF modules and efficiency in filter computation. In the paper, we also report results comparing the localization performance of the proposed extension and of the original RTPF