Dynamic Testing and Calibration of Gaussian Processes for Vehicle Attitude Estimation

Abstract

A method of estimating a vehicle's attitude in relation to the road surface using only light detection and ranging (lidar) measurements is presented. Gaussian processes, a machine learning technique, is used to relate the measurements of the road surface to the pitch and roll of the vehicle. Testing was performed under normal driving conditions on a test track as well as under high dynamic maneuvers on a skid-pad to assess performance of the algorithm. On-vehicle results show that the attitude calculations are capable of being implemented in a real-time system and have been compared against a multi-antenna GPS attitude measurement for accuracy.