Trailer-referenced autonomous navigation of agricultural tractor–trailer systems

摘要

The advancement of autonomous navigation for tractor–trailer systems is pivotal to realizing smart agriculture, especially in addressing the unique challenges of orchard environments. This paper introduces a novel framework for tractor–trailer autonomous navigation, integrating a stereo vision-based two-stage 3D obstacle detection system and a trailer-referenced Frenet Optimal Trajectory (FOT) path planning algorithm. The 3D obstacle detection combines semantic insights from RGB images with spatial data from stereo cameras, incorporating a distance-enhanced module to improve detection accuracy, particularly for small obstacles. The trailer-referenced FOT algorithm simultaneously optimizes tractor and trailer paths in the Frenet frame, ensuring seamless alignment, robust obstacle avoidance, and enhanced trajectory tracking. By integrating these components with the Stanley controller, the framework delivers simplified yet effective control strategies, bridging model-based and model-free approaches. Rigorous validation through simulations and real-world experiments demonstrates the framework’s efficacy in executing precise path tracking and obstacle avoidance in complex orchard scenarios. This work offers a significant contribution to agricultural robotics, enabling cost-effective, reliable, and scalable solutions for autonomous navigation in precision farming. The demo video is available online: link . • A framework for autonomous tractor–trailer navigation. • A vision-based, two-stage 3D object detection model for obstacle detection. • A trailer-referenced Frenet Optimal Trajectory algorithm for local path planning. • Integrate with the Stanley controller for simulations and real-world experiments.