Research On Multi-agent Cooperation Based On Task Prediction In RoboCup Rescue Simulation

RoboCup Rescue Simulation System(RCRSS)is a typical multi-agent system(MAS).Heterogeneous agents need to complete the rescue task under the limitation of time and communication conditions.This paper studies the multi-agent cooperation strategy based on RCRSS.In order to solve the problems of information exchange between multi-agents,slow task delivery and uneven distribution of tasks,the city map is partitioned to allocate multi-agent exploration areas.Agents make autonomous task predictions based on the collected information,and the decision center gathers the prediction information and makes task assignments.The main work of this paper are as follows:First,as to quickly build a map model in an unknown environment,this paper proposed a map partitioning method based on the cuckoo search algorithm based on clustering degree.This method determines the initial partition center point by the aggregation degree of buildings,and then updates the partition center point by cuckoo search algorithm.The buildings are clustered and partitioned,and the agents are distributed in the map partition for exploration,so as to build a global map model faster.Experiments prove the effectiveness of this method.Besides,in order to improve the efficiency of single agent processing tasks,this paper proposed a task prediction model based on Markov chain.This method constructs corresponding task prediction models for different agents.The fire agent predicts the spread of fire through the combustion of the building,and the medical agent judges the specific location of the rescue mission by the environmental information.Agents execute the task according to the autonomous prediction task.This method can reduce the dependence of agent on the decision center and improve the speed of single agents to execute tasks.Simulation experiments were carried out on five maps such as Berlin and Mexico,and the results proved the reliability of the prediction task.Finally,for improving the efficiency of multi-agent cooperation,this paper proposed a multiagent cooperation strategy based on the decision tree model on the basis of the above two methods.This method simplifies the data information needed by the decision center into several characteristics.The decision tree is generated according to the ID3 algorithm,and the over-fitted branches are pruned to obtain a specific decision tree model,which is used to classify the information of task prediction by decision center,as well as helping agents coorperate with each other.Experiments prove that this method can effectively improve the working efficiency of multi-agent system.