Objective The study is aimed to address the issue of lengthy global path planning of agricultural robot in complex agricultural environment and the path solution is not optimal.

Method A fast marching tree algorithm based on an improved artificial potential field method (APF-FMT*) was proposed. Firstly, relative distance was introduced in the gravitational potential field, adjusting the strength of attraction based on the distance from the target point. This overcomed the issue of excessive attraction force in the artificial potential field method when the distance to the target point was too far. The FMT* algorithm was combined with the improved artificial potential field method, and a third order B-spline curve was used to smooth the path. Finally, three agricultural working maps were created for simulation experiments.

Result APF-FMT* was compared with FMT*, RRT*, and Informed-RRT* algorithms. The simulation results demonstrated that in maps Map1 and Map2, APF-FMT* consistently found good solutions quickly, and the path solutions were improved with an increasing number of samples. The search time reduced by over 45% compared with the other three algorithms. In Map3 with narrow channels, the search times of APF-FMT* and FMT* reduced by more than 75% compared with RRT* and Informed RRT*, and better solutions were obtained.

Conclusion The proposed APF-FMT* algorithm based on the improved artificial potential field method not only overcomes the issue of redundant exploration in the FMT* algorithm, but also effectively solves the problem of unreachable target points in the artificial potential field method. This algorithm improves the efficiency and safety of path planning for agricultural robots.