AgriPath: A Robust Multi-Objective Path Planning Framework for Agricultural Robots in Dynamic Field Environments

Chenghan Yang¹Dingkun Zheng¹Siming Chen²Madina mansurova¹Baurzhan Belgibaev¹Baidong Zhao^1*

• ¹Al-Farabi Kazakh National University, Almaty, Kazakhstan
• ²Fudan University, Shanghai, China

Robot path planning is a cornerstone of precision agriculture, enabling safe and efficient operations for agricultural robots. However, the complex field environment—characterized by static and dynamic obstacles, dense vegetation, and unstructured terrain—poses significant challenges to effective path planning. Conventional methods, such as A*, Dijkstra, and Rapidly-exploring Random Trees (RRT), exhibit limitations in efficiency and adaptability to dynamic conditions. To address these challenges, this study introduces AgriPath, a robust multi-objective path planning framework that integrates an improved Convolutional Neural Network (CNN), an improved A* algorithm, and an Improved Whale Optimization Algorithm (IWOA) to optimize pathfinding, convergence efficiency, and obstacle avoidance in complex agricultural settings. Key innovations include: an Improved CNN leveraging causal convolution and multi-head self-attention mechanisms to improve temporal modeling for short-term trajectory prediction, augmented by Gaussian perturbations to enhance initial solution diversity; an improved A* algorithm incorporating dynamic heuristic functions based on Normalized Difference Vegetation Index (NDVI), combined with Kalman filtering, to bolster global path adaptability; IWOA employing nonlinear convergence factors and differential evolution mechanisms to dynamically balance path length, smoothness, and planning time; and an Improved Douglas-Peucker algorithm paired with cubic B-spline smoothing and navigation command modules to ensure path simplification and real-time execution. Experiments conducted in the Modern Agricultural Demonstration Zone at Chengdu, Sichuan Province, China, across Simple, Moderate, and Complex scenarios, demonstrate that AgriPath outperforms advanced algorithms—SBREA*, Ant Colony A*, Orchard A*, and Greedy A*—in path length, smoothness, planning time, and dynamic obstacle avoidance success rate, indicative of superior multi-objective optimization balance. This study significantly enhances the efficiency and robustness of agricultural robot path planning, offering a more adaptive solution for autonomous navigation in precision agriculture while providing new theoretical and practical directions for the field of path planning.