The Design And Application Of Multi-agent Test-bed For Coordinated Control

Multi-Agent System (MAS) is a hot research topic in the communication and control communities. The system is composed by a number of agents, which could coordinate the behavior of agents for completing a task or solving a problem. Multi-agent technology can be applied in many areas, such as intelligent robot system, traffic control, manufacturing, and distributed decision system.The researches on collaboration strategy of MAS are focus on path planning, cooperative localization and cooperative obstacle avoidance. The proposed theories and methods are usually verified by simulation and the performance of them in actual system is not considered. Therefore, this paper develops a practical Multi-Agent Test-Bed for coordinated control from the view of the whole.The Test-Bed integrates basic modules required in applications of MAS. The software for different collaboration strategy can be developed fast based on the modules library, and the practical performance of MAS algorithm can be tested on the platform. The Test-Bed is universal and easy to redevelop with a clear architecture. The main application of it contains two aspects. One is testing the practical performance of coordinated control algorithm in theory research. The other is simulating the implementation of application program based on the requirements analysis to test the feasibility and effects of application program.In order to improve the feasibility of the Test-Bed, this paper designs three applications and implements them on the Test-Bed. The three applications are: (1) Multiple agents collaboratively completing a real-time regional monitoring. (2) Formation control of Multi-Agent. (3) Tracking control of Multi-Agent.Furthermore, the Test-Bed is applied to verify the practical performance of collaboration strategies of MAS. A prediction-based collision avoidance algorithm in multi-agent system is proposed in this paper. The algorithm uses a collision prediction method to reduce the frequency and time of stopping action effectively which agents take for collision avoidance. And it is deployed on the platform for testing its practical performance. Through the analysis of experiment result, the algorithm is improved and therefore is more suitable for real system.