On Consensus And Formation Control Of Multi-agent Systems

Multi-agent systems is an important research field of complex networks. The consensus problem of multi-agent systems has attracted lots of scientists from physics, mathematics, computers, system control, biologies, sociologies and so on due to its wide applications in engineering. The difficulties of consensus problem not only lie in the distributed controller designing, network connectivity preserving and collision avoidance, but also lie in the facts that agents have intrinsic nonlinear dynamics or the systems may subject to input saturations or boundary constraints. This dissertation is to address the above significant problems in multi-agent systems. The main contributions are as follows:A sufficiently large control force is often required to ensure the network connectivity preservation and collision avoidance for a multi-agent system moving in a free space. How-ever, such control force is not realistic in engineering applications. Hence the estimated minimum control force ensuring network connectivity preservation and collision avoidance is given. Moreover, the control force estimation approach has been successfully applied in three typical collective control scenarios including swarming, flocking, and flocking with-out velocity measurement. When the control force is not enough to ensure the network connectivity presetvation and collision avoidance, the multi-agent group will be separated into several clusters. Hereby a bipartite flocking problem is considered under switching ve-locity networks such that the flock eventually separates into two disjoint clusters with all individuals moving with the same velocity magnitude, and every pair of agents in different clusters moves in the opposite directions. The bipartite flocking motion has its references in both natural collective motions and human group behaviors such as predator-prey and panic escaping scenarios.In real engineering multi-agent systems, it is often encountered that only relative dis-placements can be measured in order to save resources and costs for industrial applications. For a class of nonlinear leader-following multi-agent systems with double-integrator dynam-ics, semi-global consensus is achieved with local Lipschitz conditions. More interestingly, the initial formation is asymptotically maintained. In addition, considering the facts that actuators often have physical limits in real engineering systems, a class of linear multi-agent system is considered with input saturations and directed switching proximity networks. By virtue of a distributed observer-based controller, semi-global consensus can be achieved us-ing algebraic graph theory and Lyapunov stability theorems even the control gain is small, hence it is of good potentiality in industrial applications. Considering a multi-agent group moving in a bounded two-dimension space, under jointly connectivity conditions and a properly designed protocol using boundary bouncing force, the velocities of all the agents reach consensus. Moreover, without the guidance of global information, multi-robot cyclical torus formation control algorithm is also proposed. The theoretical designs are successfully implemented in multiple Amigobot robot formation control to achieve a desired flock in a rectangular space with bouncing boundaries, and to form a class of collective behavior in an ellipse motion pattern.Finally, it is the conclusions of the dissertation and the prospects for further research work on consensus and formation control of multi-agent systems.