Point-to-point Iterative Learning Control For Linear Self-agent Systems

As an intelligent control strategy,iterative learning control is widely used in multi-agent systems.In general,the strategy of iterative learning control is to make all agents in the system track the entire reference trajectory in finite batches through interaction and cooperation.However,in the practical application of some multi-agent systems,such as picking and placing heavy objects by multiple robotic arms,the spatial position movement of the drone group and satellite positioning,only part of the reference needs to be tracked.This is the so-called point-to-point iterative learning control(P2PILC).Compared with general iterative learning control,P2 PILC can fully utilize the freedom of non-selected tracking points to optimize system performance,which is of great significance in practical applications.Therefore,this thesis has done related research on the point-to-point iterative learning control of linear self-agent systems.The main contributions of this thesis are as follows:1.In a specific output tracking situation,a two-stage algorithm framework is designed to obtain the optimal time allocation in the sense of the optimal input energy of P2 PILC.In the first stage,norm-optimal ILC is used to solve the optimal control input.It is proved that the tracking error can converge to zero and the limit of the control input sequence exists.In the second stage,the optimal time allocation is obtained by gradient-descent method.Then,the input constraints are introduced,and the optimization problem is extended to the system with input constraints.Afterwards,the robustness of the algorithm with uncertain parameters is proved.Finally,the gantry robot is used for simulation verification.2.In the case of unknown and batch-varying initial state,the design of iterative learning controller for point-to-point tracking is considered.In this case,the initial state learning law is designed to deal with unknown and batch-varying initial states.By combining the initial state learning law with the gradient-type iterative learning control law,the zero-error convergence of the selected tracking points is proved theoretically.A form of initial state that varies according to the iterative learning law is given,and it is compared with a constant initial state and an unknown and bounded initial state that varies with batches.The optimal value of the first tracking time point in the meaning of the optimal input energy under the fixed initial state error is discussed.Finally,through the example of a gantry robot and a single quadrotor,the superiority of the initial state learning law proposed in this thesis is verified,and the switching of the control input at selected tracking points along the batch axis is described.3.In order to realize point-to-point consensus tracking of linear multi-agent systems,a distributed iterative learning controller is designed,in which each agent realizes consensus tracking by learning neighbor information from each other.When the selected tracking points of each subsystem are unknown,the optimal time allocation of P2 PILC for multi-agent systems is solved.When the initial state of each follower agent is unknown and batch-varying,the corresponding initial state learning law is designed to realize the convergence of point-to-point consensus tracking error.For the sake of reducing the burden of data transmission in the networked system,it introduces a quantizer to quantify the point-to-point tracking error.By introducing a logarithmical quantizer,the new distributed point-to-point iterative learning control algorithm realizes the asymptotic convergence of the tracking error.First,compared with the previous method of the alternate direction multiplier method to realize the point-to-point consensus tracking of the multi-agent system,the model considered in this thesis is extended to a linear time-varying model with multiple inputs and multiple outputs.Second,this thesis relaxes the requirements on the topology of the multi-agent system.It is only necessary for the graph to contain a spanning tree,and it is not necessary for the graph to be fully connected.The algorithm proposed in this thesis is less computationally intensive and has a faster convergence speed.4.For a linear multi-agent system,each agent has the same reference trajectory,which switches at a certain batch.Instead of point-to-point consensus tracking,it considers the situation in which each agent independently completes the same subtask.By combining iterative learning control and collective update strategy,it proposes point-to-point collective iterative learning control algorithm.It is proved that the collective dynamic is asymptotically stable and the collective point-to-point tracking error converges monotonically as long as there is an agent of which the point-to-point tracking error is monotonically convergent.In the switching batch,the corresponding switching learning control strategy is designed,and the performance of the algorithm is analyzed.Finally,the effectiveness of the algorithm is proved by the multi-arm picking and placing operation.