Research On Task Coordination For Heterogeneous Robot Swarms Under Real-time Communication Constraints

Heterogeneous robot swarms are widely used in complex scenarios such as multi-UAV cooperative search-attack and post-disaster search and rescue.Due to the lack of infrastructure support,heterogeneous robot swarms usually communicate through ad hoc networks.However,the mobility of robots leads to rapid changes in network topology and frequent interruption of links,which hinders the information exchange of robots and seriously reduces coordination efficiency.Besides,the complexity and uncontrollability of the environment bring conditional time constraints and uncertain time constraints to the tasks,increasing the difficulty of task coordination.To address the above problems,this paper focuses on the joint optimization of communication and task coordination,and the task coordination under time constraints.Considering the problem of communication interruption in task coordination for heterogeneous robot swarms under limited communication constraints,a distributed information consensus-based communication and coordination algorithm is proposed.We redefine the task from the perspective of task coordination and communication.Relay task and extand search task are added to ensure the stability of communication link and strengthen the search ability of robot swarms at the same time.In order to avoid the execution conflict caused by inconsistent information in distributed coordination,this paper proposes a distributed information consensus algorithm,which makes the whole group decision-making consistent according to the information exchange rules.To adapt to the real-time dynamic scenario,a fast convergent distributed Hungarian algorithm is used for task allocation.Task list update policy is used periodically,to delete undesirable tasks.In order to verify the effectiveness of the algorithm,six different maps are designed for simulation experiments.The results show that the proposed algorithm has less task completion time and average waiting time in most scenarios compared with classical task coordination algorithms such as CBBA.Focusing on the complex task coordination problem of heterogeneous robot swarms under conditional time constraints and uncertain time constraints,a coalition-based task coordination algorithm with deep reinforcement learning is proposed.Deep reinforcement learning is used to predict the task execution effect,and the parameters are continuously learned and adjusted according to the execution results.In order to reduce the early training process,transfer learning is introduced to transfer the model parameters trained from the offline data set to the current scenario.In this paper,the conditional simple time network with uncertainty(CSTNU)is designed to represent time constraints,and the conflict of time constraints is handled by a dynamic controllability check algorithm.Moreover,based on the predicted output of transfer deep reinforcement learning,appropriate robots are selected to form a coalition and complete tasks collaboratively.In order to verify the effectiveness of the algorithm,simulation experiments are carried out under different time constraints and different scenarios.The results show that the proposed algorithm achieves higher task completion rate and lower average waiting time in most scenarios,showing stable performance.