Local Perception Based Global Collective Behavioral Decision-making Of Swarm Robots Under Restricted Communication

Swarm robotic system has great potential in countermeasure environments due to its efficiency and robustness.However,individuals in a swarm may only communicate with their neighbors under communication restrictions,therefore it is hard for swarm robots to obtain global information.Moreover,some robots maintain radio silence considering the concealment,which means only onboard sensors are enabled to collect information.Thus,it is significant to research the global coordinated behaviors of swarm robots based on local information.Regarding the related research of the global coordinated behavior of swarm robots based on local information,there are still some issues exist.For instance,the global information irregularly lacks because of the unstable communication links,which resulting in high computational complexity of the existing decision-making methods.Besides,it is difficult to execute complex decision-making methods in real time considering the weak computing capability of the individuals in swarm robots.Furthermore,when the communication is not available,most scholars artificially design rules under convergence assumptions,in which the coordinated behaviors of the swarm robots are emerged after a long period of semi-random motion.The artificial rules based method often leads to low efficiency and the ability of direct decision-making is still lacking.Under the background of reconnaissance tasks in countermeasure environments,the mobile swarm robots are taken as the research object.The communication restriction is divided into two states: the communication connection is available and unavailable.Then the study of global coordinated behavior decision-making based on local information is conducted.The main research is summarized as following sections.(1)A computing resource scheduling algorithm with communication restriction is proposed to improve the performance of complex decision-making methods.On account of the weak computing capability of the individuals in swarm robots,the computing resources of neighbor robots are utilized to reduce running time.The availability of the computing resources is affected by packet loss and delay of the restricted communication links.Therefore,a utility of computation bandwidth allocation algorithm is proposed to maximize computing resources per unit transmission rate.Generally,the complex decision-making methods can be divided into sub-computing tasks that allocated to suitable neighbor robots.In this way,the total running time is shortened by parallel computing.The research results show that the robots with insufficient computing resources are able to perform complex decision-making algorithm in real time using the proposed computing resource scheduling approach.Furthermore,it is proved that the proposed approach is more effective than traditional round-robin and Max-SNR methods.(2)A visual perception approach for communication restriction decision-making is designed.Specifically,a visual multi-object-tracking perception method is designed for radio silence robots to obtain trajectories of the surrounding objects.In order to reduce the influence of accumulated error on decision-making,a spatio-temporal fusion tracking algorithm is proposed.The tracking accuracy is improved by fusing the temporal tracking results and the intra-frame spatial structure.The star-shaped structure is constructed by taking high confidence targets as root nodes.In the fusion process,spatial relative position relationship is regarded as input of observation function,and temporal tracking result is regarded as input of prediction function.The research results show that it is easier to realize global cognition by using the designed local trajectory information than discrete information.(3)A fully distributed decision-making method for global coordinated behaviors based on local trajectory information is proposed.For the robots with radio silence,a distributed strategy function is constructed based on the recurrent network.This function extends the input information to historical trajectories,which weakens the Markov assumption.Therefore,the accuracy of the partial information based decisionmaking method is improved.Furthermore,A priori model of state transition distribution is proposed.The physical properties of the robot are incorporated into the inference process through converting the state transition difference between the actual environment and the simulator into a probability distribution,so that the inference strategy obtained in the simulator can be directly applied to the actual swarm robots.The research results show that the proposed method can make near-optimal global coordinated behaviors for robots with radio silence,and the efficiency is much higher than traditional virtual physical forces.(4)An experimental platform of swarm robots with restricted communication is built,and the proposed method is verified.For the robots with available communication links,the experiments of computing resource scheduling for complex decision-making methods are carried out.The results show that more computing resources can be obtained from the designed local mobile ad hoc cloud,i.e.,the robots nearby,when the communication link is unstable.The running time of the complex decision-making method is greatly reduced in approaching reconnaissance tasks.For the robots with radio silence,the experiments of visual perception based decision-making are conducted.The results show that the local trajectory information can be stably obtained by the designed visual perception system,and then the near-optimal coordinated behaviors can be inferred using the proposed recurrent strategy function.Finally,experiments where both communication robots and radio silence robots exist are conducted.The results show that the global behaviors are achieved by treating each other as obstacles,which validates that the proposed method of global coordinated behavior based on partial information.