Clustering Sensor Data for Terrain Identification using a Windowless Algorithm

In this paper we are interested in autonomous systems that can automatically develop terrain classifiers without human interaction or feedback. A key issue is clustering of sensor data from the same terrain. In this context, we present a novel offline windowless clustering algorithm exploiting time-dependency between samples. In terrain coverage, sets of sensory measurements are returned that are spatially, and hence temporally correlated. Our algorithm works by finding a set of parameter values for a user-specified classifier that minimize a cost function. This cost function is related to change in classifier probability outputs over time. The main advantage over other existing methods is its ability to cluster data for fast-switching systems that either have high process or observation noise, or complex distributions that cannot be properly characterized within the average duration of a state. The algorithm was evaluated using three different classifiers (linear separator, mixture of Gaussians and k-NEAREST NEIGHBOR), over both synthetic data sets and mobile robot contact feedback sensor data, with success.