Cooperative working strategy for agricultural robot groups based on farmland environment

Abstract: This study aims to propose a collaborative control strategy with the task assignment and whole area coverage for heterogeneous groups of agricultural robots under the complex environment of farmland. The new system was also utilized to reasonably allocate the workload and work area of each robot in the collaborative operation of agricultural robot groups, and thereby improving the overall work efficiency in precision agriculture. Various performance parameters were considered, including the energy consumption, failure rate, historical workload, and service quality of agricultural robots. Taking the overall efficiency of a robot team as the optimization goal, the agricultural robots were selected to effectively perform the specific task, where the workload of each participating robot was determined in the collaboration control framework. The complex working environment of agricultural robots was set, according to the actual characteristics in the digital ecological circular farm of the Shandong University of Technology and Zibo Hefeng Seed Company, China. A field experiment was carried out on the whole area coverage strategy of robot groups. The farmland separated by the intertwining road was taken as a primary partitioning, according to the actual farm working environment. The primary partitions were rasterized for environmental modeling operations, and thereby the common working range of robot groups was set to the unit length of cells in the grid. Specifically, the irregular obstacle was treated by the expansion operation in binary morphology of image processing, when the edge line of the obstacle cannot align to the edge line of the raster in the simulation. In the modeling of rasterized environment and the treatment of obstacle expansion, the raster partitioning at the second level was established inside the first level partitioning. The merging operation between the raster partitions was also conducted to reduce the number of partitionings in the raster diagram. The operations expansion and reduction were carried out for the enlarged obstacles after the partition merging. The Genetic Algorithm(GA) and Hybrid Particle Swarm Optimization (HPSO) were combined to improve the traditional GA crossover operation in computer-assisted support systems. The genetic diversity and chromosome population structure were utilized to enhance the convergence speed, where the chromosome was crossed with the chromosome of real-time optimal fitness value in an iterative process. The GA concept of chromosome population diversity was established to consider the differences of fitness value for the chromosome in GA, and the species diversity in the different phases, thereby setting the adaptive crossover probability and mutation probability. The improved GA was then used to solve the problem of traversal sequence in a depth-first search algorithm when traversing over the partitions of the first and the second level. The path search rules were set in the second level partition, where the starting point and the end point of a traversal were determined in the depth-first search algorithm. The work area of a robot was allotted according to the workload of the robot using the depth-first search algorithm when traversing in the raster chart. The traversal rules of the robot in the working area were set by A* algorithm and eight-neighborhood search algorithm, further to realize the complete coverage of robot groups on the whole region. The simulation results showed that the path length of traversing each partition, iteration times, and the converge time to the optimal solution in the improved GA were 2.8%, 69.5%, and 64.2% less than those of the traditional GA, while, 9.3%, 19.0%, and 9.9% less than those of the simulated annealing algorithm. When the total area of obstacles remained unchanged, the total area repetition rate of the robot population covering the whole area increased, as the number of obstacles increased. The total area repetition rate of the obstacles with the highest dispersion in the map was 23.4% in the simulation experiment. Four types of special farmland were set, where there were different shapes but the same total amount of workload. 7 obstacles with the same positions, shapes, sizes and quantities were set in each specially-shaped farmland. The total area repetition rates of robot were 16.7%, 13.1%, 11.9%, and 6.7% when traversing the four farmlands. The experimental results showed that the task assignment and regional coverage scheme can achieve the robot group at work full coverage of work area within the prescribed time. The work strategy of the robot groups can provide theoretical support for the collaborative operation of agricultural robot groups on the complex environment in modern agriculture.