Experiments in Decentralized Multivehicle Localization Using Ultra-Wideband Transceivers

Abstract

This thesis presents a study of a decentralized solution for collaborative decentralized multivehicle localization in GNSS-denied environments using an unscented Kalman filter (UKF). By placing two ultra-wideband (UWB) transceivers on each uncrewed ground vehicle and having vehicles communicate only odometry information, robots are able to act as mobile landmarks for each other. A linear proof-of-concept is first presented and tested in simulation. A more realistic nonlinear version is then developed, with both a world-frame and a relative-frame formulation. Both versions were tested in simulation using Gazebo, and in real-world experiments using DWM1001 UWB transceivers and three Clearpath Husky robots. The world-frame filter was found to outperform dead reckoning in most instances, however in some cases poor odometry in one robot resulted in reduced performance for other robots in the system. Results for the relative formulation consistently demonstrated superior performance over dead reckoning, with an average improvement of 71.8% in RMSE. Performance was best when some robots remained stationary to act as more reliable landmarks for the vehicle in motion, but the improvement was still significant when all robots were in motion.