Authors: Hao Wang, Yaxin Ren, Noboru Noguchi
Published in: CSBE-SCGAB Technical Conferences » 5th CIGR and AGM Quebec City 2021 » Regular Sessions
Download RAW file:
Description: To control an autonomous vehicle to perform applications such as path following or obstacle avoidance requires standard robotics knowledge of position, kinematics, and dynamics. The integration of the IMU?s gyroscopes for orientation, coupled with wheel velocities and parameters measured by GPS is able to estimate the pose of the vehicle more accurately than by any of the single sensors. Combining the estimation of vehicle states by motion models and the measurements by sensors is meaningful for motion prediction and risk assessment for robot tractors. However, the state variables of sophisticated vehicle models are difficult to measure. In addition, many leading methods in machine learning are limited their usage in the online identification of a model, because of limited data, system disturbance, and measurement noise. In this research, we adopted linear regression methods to identify the vehicle dynamic model and to improve the orientation measurement. In addition, a data-driven steering model derived from the unicycle model is developed in this research for headland turns. Experiments were conducted by using a 4-wheel robot tractor with the speed ranging from 1.5 m/s to 3.5 m/s. Fusing the estimation of the vehicle model and sensor data can deduct the pose of the tractor within centimeters accuracy during the GPS outages.
Conference name: 5th CIGR International Conference and CSBE-SCGAB AGM 2021, Quebec City,QC, 11-14 May 2021.
Session name: Machinery and Robotic Systems 3
Publication type: Presentation
Language 1: en
Rights: Canadian Society for Bioengineering