Real-time turn path planning with obstacle avoidance for autonomous agricultural field machines

Turn path planning is an essential component of autonomous agricultural field operations, where nonholonomic field machines must navigate between work swaths while respecting spatial constraints such as field boundaries and obstacles. This article presents a real-time turn path planning algorithm for autonomous agricultural field machines. The algorithm was designed to operate as part of a cooperative coverage path planning system in which multiple autonomous tractors perform sequentially dependent tasks simultaneously on the same field. The proposed algorithm is based on Hybrid A*, combining analytical Reeds-Shepp and Dubins path computation with a four-dimensional A* search, followed by gradient descent smoothing to achieve path feasibility. A state machine architecture was employed to meet the real-time constraint of a maximum of 100 ms per function call and 3 s of total computation time per turn path. The algorithm was evaluated through four experiments which include real-world field experiments with autonomous tractors. In the simulated obstacle-free experiment, the average computation time was 118 ms, which corresponds to approximately 4 % of the permitted maximum. The generated paths were feasible, with curvatures averaging approximately 60 % of the maximum permitted by the minimum turning radius. In the field experiment, the algorithm successfully generated all required turn paths in real time as part of a complete autonomous coverage operation. The results demonstrate that the proposed algorithm is capable of generating feasible turn paths in real time for typical agricultural field operations with spatial constraints.