Lane following with obstacle avoidance for unmanned tracked vehicles using monocular vision and active disturbance rejection control

This paper presents an integrated system for autonomous lane following with obstacle avoidance in unmanned tracked vehicles (UTVs), combining monocular vision and Active Disturbance Rejection Control (ADRC). A vision-based guidance system is developed using deep learning models: YOLOPv2 for lane segmentation and YOLOv8 for obstacle detection within a dynamic region of interest. Novel lane processing algorithms address partial detections and generate aligned lane boundaries, while a computationally efficient virtual lane generation mechanism enables path planning around obstacles without requiring dedicated depth sensors. To follow the path defined by this guidance system, an ADRC controller is designed for the UTV's lateral control channel based on a kinematic model, incorporating disturbance estimation via an extended state observer, ensuring robust regulation of lateral path error. The system's effectiveness is demonstrated through comprehensive experimental validation on a physical UTV platform in two distinct environments: an indoor track with static obstacles and an outdoor setting with both static and dynamic obstacles. Outdoor trials confirm the system's robustness against real-world challenges, including sloped terrain, varying natural lighting, and multi-colored lane markings. Furthermore, the system successfully navigated around obstacles and critically validated its fail-safe stop logic when the path was fully blocked. Comparative tests against a conventional PID controller quantitatively demonstrate the ADRC's superior tracking accuracy and disturbance rejection capabilities, highlighting its enhanced robustness in both controlled indoor and unstructured outdoor environments. These results confirm the feasibility of achieving robust lane following and effective obstacle avoidance in UTVs using cost-efficient monocular vision. Supplementary material: https://youtu.be/9aKGugeYmfw?si=qiBCTzi7hYUvwUW6