Slip-aware localisation for wheeled vehicles

Abstract

Wheeled robots are subject to slip which may cause errors in position estimation if not correctly observed. Wheel slip is typically considered only by a path-following controller. An offline-calibrated slip value can also compensate for wheel odometry (WO). This thesis proposes a new method of fusing visual odometry (VO) and WO with integrated slip estimation using the Extended Kalman Filter. The approach handles the correlation between the slip and the rover’s position estimation and occasional errors in any VO or WO measurements. Furthermore, it is possible to tune the model to emphasise WO (e.g. when no slip is expected and thus reduces the number of VO measurements) or rely more on VO (high slippage variability).

An adaptive filter is introduced to tune both process and measurement noises automatically based on available measurements. It enables the assessment of noise statistics at each filter’s iteration, leading to the design of a reactive VO scheduling algorithm. The solution does not rely on prior knowledge of the environment and offers a route to conserve computational resources while maintaining good navigational accuracy.

The localisation system can be further improved if the wheel slip is known in advance. This prior information can be exploited by investigating how the wheel slip predictions, derived, for example, from forward-facing vision or motor current, can be fused within the model using different slip prediction schemes. The solutions provide improved localisation accuracy and hint at the intriguing possibility of slip-based SLAM.

The above research scope was reduced to one dimension to lower the complexity and focus on the principles of new algorithms. However, to advance some elements described in this thesis, the adaptive filter with integrated slip estimation and reactive VO scheduling is expanded into two dimensions. The results demonstrate improved navigational performance but highlight the need for more work in the future.

Date of Award	6 Dec 2022
Original language	English
Awarding Institution	University of Bristol
Sponsors	CGI IT UK
Supervisor	Arthur G Richards (Supervisor), Tom S Richardson (Supervisor) & Mark Woods (Supervisor)