Kinematic Analysis of a Small-Scale Off-Road Autonomous Vehicle Using GPS and Wheel Encoder Data

open_access
Open Access
Author:
Adu, Isaiah
Area of Honors:
Mechanical Engineering
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Sean N Brennan, Thesis Supervisor
Jean-Michel Mongeau, Thesis Honors Advisor
Keywords:
State Estimation
Sensor Fusion
Autonomous Vehicles
Kalman Filtering
Dead Reckoning
Kinematic Models
GPS
Encoders
Abstract:
Autonomous vehicles require accurate state estimation for safe navigation and control. This thesis develops methods for estimating vehicle position, velocity, heading, and yaw rate by fusing Global Positioning System (GPS) and wheel encoder measurements. The methods were validated using experimental data from a 1/5th scale RC vehicle equipped with RTK-GPS and high-resolution encoders. Vehicle heading was estimated by GPS using position-based and velocity-based approaches. The velocity-based method produced significantly less noise and was adopted for subsequent analyses. A regression-based method simultaneously estimated tire radii and track width, eliminating systematic errors in encoder velocity calculations during turns. Tire slip analysis revealed correlations between slip events and remaining deviations in radius estimates. Dead reckoning was compared using kinematic unicycle and kinematic bicycle models with GPS and encoder sensor combinations. The kinematic unicycle model achieved lower position errors across all combinations. GPS heading prevented drift accumulation while encoder heading exhibited integration errors. Kalman filters with bias-augmented state estimation were implemented for both process models. The filters estimated encoder biases online, enabling fusion of all measurements despite systematic errors. Fused estimates achieved smoother velocity profiles than GPS alone while maintaining bounded position errors. Sensor dropout analysis demonstrated limited degradation during GPS outages with linear error growth and successful recovery when measurements resumed. The implemented methods provide a foundation for robust state estimation in autonomous vehicles operating with imperfect sensors and intermittent GPS coverage.