Research On Swarm Intelligence Based Algorithm For Distributed Search And Collective Cleanup

With the rapid development of robot technology and application in recent years,the research from single robot to multi-robot becomes hotspot at home and abroad. Multi-robot task performing in unknown region becomes more and more popular. The key problem in multi-robot task performing is multi-robot cooperation.Traditional cooperation for multi-robot includes centralized method of distribution or the central coordination. The performance of these algorithms will decrease dramatically by the number of robots increased quickly. And the communication between robots and control terminal are susceptible to interference in case of bad environment.The topic comes from National Natural Science Foundation program" The network environment based on swarm intelligence approach virtual machine Collaborative Platform" .The program hopes to explore a new parallel computing model to control the large-scale group collaboration and verify algorithm performance through multi-robot distribute search and collective cleanup scenario. The project scenarios are: in a unknown region were littered with a number of targets need to be cleaned, some of the targets are too big to cleaned by only one robot, need two or more robots cooperation to clean it .Robots should search and clean-up all the targets through robots'collaboration and interaction with the environment, for larger targets, robot will work together to complete clean-up. The algorithm is fully distributed and the whole process does not require centralize control, it also does not produce a significant increase in traffic bottlenecks when a substantial increase in node. Because it's fully distributed, the communication requirements are greatly reduced.The main research topics include the following:(1) Secondary search algorithm based on probability: we divided the map into a number of small sub-region and using a secondary search algorithm based on probability to speed up the search speed. To avoid duplication detection, each robot is equipped with a buffer to record the path of the current sub-region.(2) Particle swarm optimization strategy for balance: use the particle swarm algorithm for decision-making when target is detected and balance the tradeoff of exploration and exploitation to optimize the whole system operation.(3) Algorithm Simulation: In order to verify the feasibility and efficiency of the algorithm, we chosen a entity robot simulator which developed by the University of Southern California. Through comparison with other method to verify the efficiency of the algorithm.The simulation results prove that the algorithm can solve the distribute search and collective clean-up problem in unknown environment. In addition, under the same experimental results of different algorithms optimized contrast also reflects the more efficient of our algorithm. Algorithm is fully distributed, no self-node, on features such as low communication requirements could be applied to large-scale production applications.