Global Feasible Path Planning for Pest Monitoring Robots in Unstructured Agricultural Environments

Yipeng Shao¹Fazhan Tao²Pengju Si²Baofeng Ji²Meng Yang Li^3*

• ¹Henan University of Science and Technology, Luoyang, China
• ²Longmen Laboratory, Luoyang, China
• ³Luoyang Normal University, Luoyang, China

In the context of sustainable agriculture and precision agriculture, autonomous mobile robots play a pivotal role in intelligent plant pest and disease detection. However, navigation within complex agricultural environments—characterized by narrow crop rows and irregular obstacles— remains a persistent challenge. To address these limitations, this study presents a Goal-Oriented Adaptive Bidirectional RRT* (GABi-RRT*) algorithm. This algorithm addresses the problems of low sampling efficiency and unstable path quality of the RRT* algorithm in narrow spaces. Firstly, during the sampling phase, to mitigate the low efficiency associated with the random sampling of the conventional RRT* algorithm, an innovative dynamic goal-oriented probabilistic sampling strategy is presented. This strategy adaptively adjusts the sampling probability throughout the planning process while accounting for the obstacle distribution between the current node in the random tree and the target, thereby enhancing the algorithm's sampling efficiency. Secondly, in the RRT* expansion phase, to augment the algorithm's exploratory capability across diverse environments, this paper integrates the RRT* expansion mechanism with an enhanced Artificial Potential Field (APF) and introduces an adaptive step-size strategy to improve the algorithm's search efficiency. Subsequently, in the path optimization phase, to further refine path quality and reduce the turning frequency of monitoring robots, a hybrid approach combining pruning optimization and cubic B-spline curve fitting is employed. This method serves to eliminate redundant nodes and smooth the generated path. Finally, comparative experiments in a continuous narrow environment simulating crop rows reveal that the proposed GABi-RRT* algorithm significantly outperforms Bias-RRT*, Informed-RRT*, and P-RRT*; it shortens the average running time by 45.39%, 49.71%, and 71.52% respectively, and the path length by 5.70%, 3.97%, and 1.60%, demonstrating superior capabilities in terms of path quality, stability, and search efficiency.