Collision Avoidance Control For Multi-agent Systems With Application To Flight Vehicle Swarm

Multi-agent collaboration system has great advantages in flexibility,reconfiguration,scalability and robustness.It has been widely used in formation or cluster flight,multi-mobile robot cooperation,sensor network deployment and energy balance of micro-grid.Flight vehicle swarm,as a key technology of many space and aerial applications,such as new distributed remote sensing,on-orbit service,deep space asteroid exploration,and cooperative monitoring has become the focus of aerospace technology development.Therefore,in the past decades,cooperative multi-agent control has been a research hotspot,and is developing towards more complex and practical direction.Aiming at contributing to safe and effective operation of multi-agent system,with the consideration of collision avoidance requirement,various kinds of perturbations and constraints,limited communication and computing ability,this paper studies robust control under external disturbances and uncertainties,event-driven control and optimal formation control.Furthermore,we expand the results to flight vhicle swarm control,which includes input constrained optimal control and event-driven optimal control.The main research contents and achievements are as follows:1.For multi-agent systems with input constraints,external disturbances and model uncertainties,robust formation control with collision avoidance is studied.Firstly,a robust control law based on robust integral of the sign of the error and anti-windup compensator is designed in the framework of backstepping method,which overcomes the discontinuous input of conventional sliding mode control and addresses the effect of input saturation.Meanwhile,a continuous artificial potential function is introduced to ensure the collision between without local minimization.Then,when considering complex input constraints and polyhedral obstacles,neural networks are used to approximate multiple model uncertainties,while an adaptive robust term is incorporated to compensate for constrained inputs,approximate errors of neural networks and external disturbances.The proposed control algorithm achieves robustness against complex perturbations while avoiding collisions.2.Considering the communication delay between agents and the limited energy of agents,the robust formation control of nonlinear second-order nonlinear multi-agent with heterogeneous communication delays and event-triggered formation control are studied,where collision avoidance constraints are considered compared with the existing studies.The robust controller based on neural network is used to approximate the global disturbances caused by heterogeneous time delays,dynamic uncertainties,external disturbances and observation errors of the derivatives of collision avoidance vectors.The Lyapunov-Krasovskii function method and algebraic graph theory are used to analyze the sufficient conditions for system stability and collision avoidance.Meanwhile,an event-driven distributed collision free formation control law and its corresponding event-triggered conditions are designed,where the event-driven conditions not only depend on the state deviation of the agents,but also depend on the deviation of collision avoidance vector.Using this control law,the requirements of continuous communication among agents and continuous control calculation are avoided,thus reduce the energy consumption of agents efficiently.3.To solve the formation control problem of multi-agent system with various constraints and trajectory optimization,two novel control strategies are proposed,i.e.,model predictive control based on Lyapunov function and adaptive dynamic programming control algorithm.Firstly,the control Lyapunov function(CLF)is introduced into the distributed model predictive control scheme,which combines the strong stability of CLF and formation optimization performance.The feasibility of the system is achieved by introducing a relaxation parameter that makes a trade-off between formation tracking and collision avoidance performance.At the same time,terminal constraints based on the concept of velocity obstacle can eliminate the requirements of long prediction horizon and terminal auxiliary control law,which simplifies the control law's implementation.In order to achieve optimal formation control in infinite horizon,we incorporate he objective function of collision avoidance into the optimization problem's cost function,and then the adaptive dynamic programming method based on single-layer neural network is used to solve the optimization problem.The control law achieves collision avoidance cooperative control,while improving the optimal performance of agents' trajectories and input energy.4.Based on the results above,according to the characteristics of flight vhicle swarm(large quatity,and week performance),we first study an adptive dynamic programming based swarm optimal control method subjected to input constraints through designing a suitable input cost functioin and a novel updating law of the neurual network weights.Furthermore,to reduce both the consumption of flight vehicle's communication,computation and power energy with collision avoidance and input constraints,an event-triggered based optimal swarm control strategy is developed,to the best of my knowledge,at the first time.