Research On Connectivity And Autonomous Control Of Multi-robot Systems

With the development of big data processing technology and artificial intelligence,the robot's perception and processing capabilities have been continuously enhanced.Robots are being widely applied to various fields such as industry,agriculture and daily life.Compared with single-robot system,multi-robot systems(MRS)can achieve better performance through cooperation and coordination.Autonomous control is the key to realize the application of MRS.And maintaining communication connectivity is the foundation of cooperation and coordination.However,most of the existing autonomous control methods are designed with an assumption of a connected robots team,leading poor applicabilities.Therefore,it is of great significance to research on the connectivity and autonomous control of MRS.Based on the analysis of the theory and characters on connectivity of MRS,this thesis solves the problem of autonomous control with connectivity under different scenarios.The main content and innovations include:· Researching and analyzing the basic theories and characters on the connectivity of MRS(Chapter 2)This thesis systematically expounds the basic theories on connectivity of MRS for the first time.Based on the communication topology graph,we give the necessary and sufficient condition for a connected MRS,introduce the concepts of algebra connectivity,vertex connectivity,edge connectivity of MRS and related properties in detail.Then we analyze the characters on connectivity of MRS from three aspects: space-time,task and survivability,and propose the concepts of implicit connectivity and kernel connectivity.These results can provide theoretical guidance for solving the problems of autonomous control with connectivity,and support the design of autonomous control methods.· Proposing a method for autonomous control with inner-connectivity of MRS(Chapter 3)Considering the multi-robot navigation scenario,this thesis constructs a model for the problem of the autonomous control with inner-connectivity of MRS,and designs a multi-layered control framework for the multi-robot navigation.Based on a distributed navigation method,we add a path planning module,design and im-plement a hybrid approach called RRT-DNF.The effectiveness and performance of the RRT-DNF method are verified by typical simulation cases.The results show that the RRT-DNF method can complete the multi-robot navigation task and guarantee the inner-connectivity of MRS.Compared with the distributed navigation method,the RRT-DNF method improves the task execution efficiency in the obstacle-free environment by 20.17%-63.94%,and in the obstacle environment is improved by7.77%-29.08%.· Proposing a method for autonomous control with MRS-BS connectivity(Chapter4)This thesis constructs a model for the problem of autonomous control with MRSBS connectivity,and proposes a role-based control framework MRSA,which can decide relay robots in a fully distributed way.An AMRD algorithm based on LAP decision is proposed.MRSA-based control methods are designed for multi-robot systems with different models.We verify the effectiveness and performance of MRSA-based control methods through a series of simulations.The results show that the MRSA-based control methods can maintain the continuous connectivity between the multi-robot system and the base station,and the control traffic load is reduced by 90% compared with the market-based method.· Proposing a method for autonomous control with connectivity of model-unknown MRS(Chapter 5)This thesis firstly uses the deep reinforcement learning method to realize autonomous control with connectivity of MRS.Considering the leader-follower scenario,we construct a model for the problem of autonomous control with connectivity of model-unknown MRS.We design a discrete control learning framework based on deep Q-network,and a continuous control learning framework based on deep deterministic policy gradient.We design and implement a leader-follower simulator,which provides an environmental model for MRS learning.The experimental results show that both learning frameworks can effectively complete a variety of leader-follower missions with different settings.