Scholarlink

<p>Automation in agriculture is gaining significant popularity due to its potential to enhance productivity and reduce operational costs. In palm tree-clustered environments, autonomous navigation is essential for tasks such as cleaning fallen dates. However, autonomous navigation in such environments poses challenges in localization, path planning, and control: GPS signals degrade under dense foliage, planning must be fast enough while handling narrow passages, and control must adapt to slippery terrains. This paper presents an integrated navigation system that unifies localization, planning, and control for real-time operation. Localization uses a night-enhanced You Only Look Once (YOLO) v4 detector trained with augmented data, achieving 98.8% accuracy under varied lighting. Detected landmarks are fused with ultrasonic and odometry data through a confidence-weighted algorithm for improved accuracy. A combined path planning control framework is proposed using an Artificial Potential Field (APF)-based approach that directly maps potential values to velocity commands, eliminating the computational bottleneck between separate planning and control modules. The proposed system pre-builds a potential field representation of the entire environment, enabling efficient handling of frequent localization updates by simply repositioning the search window around corrected robot positions without requiring complete field recalculation. Moreover, Gaussian functions are employed instead of standard inverse-distance models to represent obstacles, creating smoother potential gradients that facilitate escape from local traps. The strengths of attractive and repulsive forces are adaptively chosen to allow the robot to navigate through narrow corridors. Experimental results demonstrate that the proposed framework achieves the shortest paths with substantially lower control effort and a six- to twelve-fold reduction in computation time compared to state-of-the-art planners, while maintaining smooth and collision-free motion in cluttered environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-time autonomous navigation in palm tree environments using vision-based landmark detection and an integrated path planning and control framework

  • 期刊论文

  • 发表日期：

    2026年1月19日

查看全文
  • Mohammed Baziyad,
  • Bilal Arain,
  • Tamer Rabie,
  • Ibrahim Kamel

摘要

Automation in agriculture is gaining significant popularity due to its potential to enhance productivity and reduce operational costs. In palm tree-clustered environments, autonomous navigation is essential for tasks such as cleaning fallen dates. However, autonomous navigation in such environments poses challenges in localization, path planning, and control: GPS signals degrade under dense foliage, planning must be fast enough while handling narrow passages, and control must adapt to slippery terrains. This paper presents an integrated navigation system that unifies localization, planning, and control for real-time operation. Localization uses a night-enhanced You Only Look Once (YOLO) v4 detector trained with augmented data, achieving 98.8% accuracy under varied lighting. Detected landmarks are fused with ultrasonic and odometry data through a confidence-weighted algorithm for improved accuracy. A combined path planning control framework is proposed using an Artificial Potential Field (APF)-based approach that directly maps potential values to velocity commands, eliminating the computational bottleneck between separate planning and control modules. The proposed system pre-builds a potential field representation of the entire environment, enabling efficient handling of frequent localization updates by simply repositioning the search window around corrected robot positions without requiring complete field recalculation. Moreover, Gaussian functions are employed instead of standard inverse-distance models to represent obstacles, creating smoother potential gradients that facilitate escape from local traps. The strengths of attractive and repulsive forces are adaptively chosen to allow the robot to navigate through narrow corridors. Experimental results demonstrate that the proposed framework achieves the shortest paths with substantially lower control effort and a six- to twelve-fold reduction in computation time compared to state-of-the-art planners, while maintaining smooth and collision-free motion in cluttered environments.