Research On Path Planning Of Mobile Robot Swarms For Orchards

The autonomous driving operation of orchard mobile robot can reduce the dependence of agricultural production on labor and improve the production efficiency in orchard.MultiAgent Path Finding(MAPF),as an important cornerstone for autonomous driving of mobile robot cluster,has always been the main content of researchers at home and abroad.Traditional path Planning methods have low quality of solving paths and poor adaptability of solving algorithms.In view of these problems,the specific research contents of this paper are as follows:(1)In order to solve the problem of low path quality and poor adaptability by traditional path planning algorithms,this paper first proposes a multi-agent path planning algorithm set,which includes various types of algorithms to enrich the adaptability of the model.Secondly,in order to automatically select the algorithm from the algorithm set according to the scene,this paper proposes a Res Net-based convolutional neural network model to complete the automatic selection of the algorithm.The shortest path of a single agent is embedded in the map and the manually labeled MAPF features are combined and input into the input layer of the model,which expands the difference between different path solving instances,and uses transfer learning to improve the efficiency of model learning and the model selection algorithm.The accuracy and coverage of the multi-agent path planning algorithm are verified by experiments at last.(2)For multi-agent local path planning,this paper proposes a multi-agent local path planning algorithm based on PPO-IL reinforcement Learning.Imitation Learning(IL)is added on the basis of PPO reinforcement Learning.Agents in reinforcement Learning can be guided by expert demonstration in Imitation Learning.Thus,it can move towards high quality action-behavior space and improve the learning efficiency of the agent.Through experimental verification,compared with the original PPO algorithm,PPO-IL with imitation learning has a certain improvement in convergence speed of multi-agent local path Planning algorithm,and the success rate and solving time of path Planning have a certain improvement.(3)Aiming at the path planning of multi-mobile robots in orchards,due to the unstructured characteristics of orchards,this paper designs and implements a path planning system for mobile robots in orchards.The system integrates the multi-agent path planning algorithms in the above two points.According to the different orchard environments,it can select the corresponding path solving algorithm,which meets the needs of users for refinement.