Research On Coordination Control For Networked Multi-agent Systems

Integrated with development achievements in many research areas, the agent technology represents the fore-edge of high-end technology developments. During recent decades, the agent system has become a hot topic of the international robotics and artificial intelligence research, more and more widely applied to various fields of industry, transport, services, aerospace and defense. However, accompanied by the growing complexity of the task environment, a single agent plays the very extremely limited performance. The usages of agents perform as from a single platform of the gradual development to the more flexible coordination multi-agent systems. Multi-agent systems have the following characteristics:the complexity of task, the distribution of spatial and temporal, functionable, distributionable, the distribution of perception and low reliability. The coordination control of multi-agent systems needs to take the perception, implementation, communications, and environmental dynamic changes in non-ideal situation fully into account. From a practical point of departure, this paper discusses three main issues in the coordination and control:consensus, swarm intelligence and formation control.The main work and research results are as follows:1. Focus on the situations with error sequence, packet-drop, etc.; the networked consensus with time-delay is researched. Combined with the analysis results about the allowed threshold of time-delay in the network, a consensus algorithm with data buffer is proposed. Through the process of sorting data in the buffer, the algorithm can make sure that agents get an optimal control input during a sampling period. The performance of the entire system gets better under the algorithm.2. The consensus problem under limited communication channel and noise is researched. Firstly, a consensus algorithm with quantizer is proposed. The inner information among agents is dealed with the quantizer, so that the bandwidth using for information transformation is reduced. The consensus can be reached through this limited network bandwidth. Secondly, the performance of system consensus is improved under the noisy environment by embedding Kalman filter into agents.3. Finite-time consensus problems are considered. For solving rapid consensus problem, an algorithm which can make the system reach consensus during finite-time is proposed. By defining symbolic function, the algorithm accelerates the convergence rate. The numerical simulation shows the effectiveness of the proposed algorithm.4. The flocking intelligence in multi-agent systems is researched, including the system with single leader and multi-leaders. Firstly, the concept of virtual force potential is introduced. An effective flocking algorithm which is easy to calculate is proposed, by defining the interaction force among agents. Secondly, the strategies of choosing limited neighbors and the optimal leader are proposed in multi-leader systems, and a flocking algorithm is proposed under the strategies. The algorithm makes sure that all the followers can choose their optimal leader. Agents with the same leader work together and agents with different leaders separated from each other.5. The formation control in multi-agent systems is researched. A new problem named as minimum formation distance problem is proposed. The problem takes the sum of all the agents’ trip lengths into account. The purpose of the problem is to find some kinds of algorithms, so that find the optimal path and reduce the sum of agents’trips. Three effective algorithms have been proposed in this paper, i.e., traverse algorithm, dynamic programming, and ant colony optimal intelligent algorithm. The numerical simulations illustrate the effectiveness of proposed algorithms.