A Robust Visual-IMU-Wheel Odometry Using PID Controller for Autonomous Soccer Robots

Abstract

Accurate and reliable autonomous navigation systems are important for self-localization in robot soccer game fields. Visual-IMU-odometry methods potentially achieve higher accuracy than GNSS-based methods due to their superior resolution and a higher rate of data sampling, especially in indoor scenarios. However, existing approaches fail to achieve good accuracy in ever-changing realistic match environments. This study presents a state-of-the-art odometry method by fusing visual sensor, inertial measurement unit (IMU) sensor, and wheel speed sensor in the autonomous soccer robot (ASR) context. First, a computer vision (CV) method is employed to track moving objects and generate the target location where the robot needs to be directed. Next, a motion estimation process is introduced into the conventional localization process to generate the robot movement using a proportional-integral-derivative (PID) controller. Finally, the robot's response to dynamic objects from the 360-degree point of view is tuned by monitoring the rapid convergence of the motion estimation process and applying Ziegler-Nichols tuning strategy. The research demonstrates increased accuracy in the proposed method, achieving high accuracy in auto-positioning with max error values below 2% for the robot coordinate and 4.4% for the angle of robot direction. The mean error on the x-axis, y-axis, and the angle of robot direction is 0.63 %, 0.47 %, and 1.19%, respectively.