System Identification, Using Vision-Based Localisation, for a Hexapod Robot

Abstract

This Masters Thesis consists of two broad sections. The first documents the vision-based localisation algorithms that I have implemented. The second section describes the process and the results of system identification for RHex. All of the localisation algorithms are feature-based schemes which detect artificial landmarks using a fast colour segementation process. The main difference between the algorithms is the type of artificial landmark used. The line-based localisation scheme returns a two degree of freedom location, where as the point (or sphere) and cube based schemes can give full SE(3) poses. These algorithms were specifically designed for use with the vision system on the hexapod robot RHex, but they are general enough to be used on any similar infrastructure. The system identification was implemented using classical parameter-based methods, where RHex's motion model is approximated by linear ARMAX models. The analysis showed that the faster the gait used, the greater the order of the ARMAX polynomials required to give a reasonable model representation. A novel aspect of the system identification problem is that RHex is not a rigid body (as its legs are compliant). This is in contrast to the majority of (mobile) robots, which can be reasonably approximated by rigid bodies.