Observer-Based Consensus Of Multi-Agent Systems

Because of the wide applications in the feld of military and civilian, the dis-tributed control of multi-agent systems has been an active research feld. Consensusis one of fundamental issues in the study of the distributed control of multi-agentsystems and is the base to develop other coordinated control problems. This disser-tation mainly studies consensus problems of second-order multi-agent systems. Forthe case that some information is unmeasured, we further consider the observer-based consensus problems. The main results and contribution of this dissertationare as follows.The observer-based consensus problem of second-order discrete-time multi-agent systems is studied. A distributed full state observer and a distributed controllaw, respectively, are designed, where distributed full state observer is used to esti-mate all states of multi-agent systems. Finally, the sufcient and necessary condi-tions for consensus of second-order discrete-time multi-agent systems are given byusing the matrix analysis method for the cases of fxed and random topology.The observer-based consensus problem of second-order leader-follower multi-agent systems with switching topology where none of velocities of leader and follow-ers can be available is studied. Firstly, under the assumption that the accelerationof the leader is active and known to all followers, this work estimates the dynamicdiferences of the velocities between the leader and the followers by designing dis-tributed observers. A common Lyapunov function is used to analyze the closed-loopstability and sufcient conditions of consensus are established. Secondly, the ideais further extended to two cases where the acceleration of the leader is constant butunknown to all followers and is active and known only to some followers. In thiscase, sufcient conditions of consensus are presented under the relaxed conditions.The sampled-data consensus problem is addressed for second-order leader-following multi-agent systems under fxed undirected/directed interaction. A dis-tributed linear consensus protocol is designed based on sampled data where eachfollower-agent can obtain the positions and velocities only at sampling times. Sev-eral sufcient conditions on the sampling period, the control gain, and the com-munication graph are given to achieve consensus. The quantitative bound of theconsensus errors is also given under, respectively, an undirected interaction topol-ogy and a directed interaction topology. It is shown that the ultimate bound of the tracking errors is proportional to the sampling period.