Steering-Angle Prediction and Controller Design Based on Improved YOLOv5 for Steering-by-Wire System

Area of Science:

• Robotics and Artificial Intelligence
• Control Systems Engineering
• Computer Vision

Background:

• Steering-angle prediction is vital for autonomous vehicle (AV) control, but traditional methods like YOLOv5 face limitations in accuracy and efficiency.
• Controlling steering angles is challenging due to difficulties in measuring angular velocity and external factors like disturbances and friction.

Purpose of the Study:

• To propose a lightweight steering angle prediction network, YOLOv5Ms, enhancing computational efficiency and prediction accuracy for AVs.
• To develop an adaptive output feedback control scheme with output constraints for effective regulation of predicted steering angles.
• To address the lack of diversity in existing lane-line datasets by creating a novel dataset from real-world road images.

Main Methods:

• Developed YOLOv5Ms, a lightweight model based on YOLOv5, incorporating Shape-IoU_Loss for improved bounding box regression.
• Created a new lane dataset from real roads to train the network model.
• Designed an adaptive neural network output feedback controller using backstepping and Lyapunov barrier functions for constrained output control.

Main Results:

• YOLOv5Ms achieved a 30.34% reduction in weight storage and a 7.38% increase in accuracy over YOLOv5s.
• The proposed control scheme demonstrated a 39.16% better root mean squared error (RMSE) than traditional backstepping control.
• The system showed effective estimation of angles, angular velocity, and unknown disturbances, ensuring closed-loop system stability.

Conclusions:

• The YOLOv5Ms model offers a significant improvement in efficiency and accuracy for steering angle prediction in autonomous vehicles.
• The adaptive output feedback control scheme effectively regulates steering angles, outperforming traditional methods in accuracy and robustness.
• The integrated approach ensures stable and reliable control for autonomous driving systems.