Path Planning Algorithms for Agricultural Machines

Abstract

If the field plot shape is not rectangular and if it contains obstacles, the coverage path planning problem is hard to solve for a non-omnidirectional machine. Scientists have developed several algorithms to solve this coverage path planning problem, but all of them have pros and cons. If the machines were omnidirectional and turning times were decreased to insignificant, the problem would be quite easy to solve using known robotic path planning methods. Traditional agricultural machines, like tractors, tractor-trailer combinations, self-propelled harvesters and other man-driven machines are slow to turn at headlands. This is the most differentiating property of the problem formulation compared to traditional robotic coverage path planning, which has dealt mainly with omnidirectional kinematics. In this article two different algorithms are presented to solve the coverage path planning problem for agricultural machines. The first algorithm is a higher level algorithm to split a complex shaped field plot to smaller parts is presented. The higher level splitting algorithm is presented in detail in this article. The algorithm can handle any field, including obstacles. The algorithm is based on trapezoidal split, merge and search. The algorithm is suited to any kind of vehicle, which is described with a few parameters, like working width and turning time function. In the latest version, the required headlands are generated automatically and there is also a possibility to define regional restrictions as forbidden driving directions. With this formulation it is possible to take into consideration the previous operations, under drains and steep gradients. The second algorithm utilizes bottom-to-top approach. It is designed for real-time usage and it solves the problem recursively: the operated area is removed from the field and the algorithm is repeated until the whole field plot is completed. In the development phase of algorithm, a simulator has been utilized. The underlying idea is to calculate the efficiency for all possibilities to make one trip around the field and to select the best one. It is assumed that every new swath is side-by-side to the some previous one or to the boundary of the field plot. However, even if the underlying idea is simple, the search space explodes when the number of corners of the field plot raises and heuristics is needed in order to restrict the number of possibilities without losing optimality. The algorithm is suited to any kind of vehicle, which is described with a few parameters like working width and minimum turning radius. Preliminary results are very promising and are presented in the article.