Output Resilient Containment Control of Heterogeneous Systems With Active Leaders Using Reinforcement Learning Under Attack Inputs

Qing Li,Lina Xia,Ruizhuo Song

doi:10.1109/access.2019.2947558

Qing Li, Lina Xia + Show 1 more

Open Access

https://doi.org/10.1109/access.2019.2947558

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2019
Citations: 6	License type: CC BY 4.0

Affiliation: University of Science and Technology Beijing

Abstract

The optimal solution to the distributed output containment control problem of heterogeneous multiple-agent systems (MASs) with unknown active leaders under attack inputs by using data-based off-policy reinforcement learning (RL) is proposed. Assume that the control input of each leader is bounded and non-zero. Moreover, followers are vulnerable to attack signals in real-world application. Firstly, distributed observers are designed such that the state and output of observers fall into the convex hull formed by leaders. Then, the output containment problem is converted into $H_{\infty }$ tracking problem by minimizing value function, Algebraic Riccati equations (AREs) are obtained in solving optimal $H_{\infty }$ tracking problem for each follower, which are computed by a data-based off-policy RL algorithm without using agents’ dynamics. At last, the effectiveness of the algorithm is verified by a simulation example.

Highlights

The problem of multi-agent synchronization control has been investigated in many works [1]–[5], among which the problem of containment control has attracted board attention and been successfully applied to many subjects [6], [7]
0, q i=1 βi Problem 1 (Output Containment Control Problem): To solve the output containment control problem with active leaders under attack signal, distributed control law ui is designed for each follower in (2), such that the output of all followers yi keeps them in the convex hull formed by leaders’ output ok in (1), i.e., limt→∞ yi(t) → Co({ok }k∈ ), ∀i
DISTRIBUTED OBSERVER the observer is designed to estimate the trajectory for each follower which is inside the convex hull formed by leaders

Summary

INTRODUCTION

The problem of multi-agent synchronization control has been investigated in many works [1]–[5], among which the problem of containment control has attracted board attention and been successfully applied to many subjects [6], [7]. Q. Li et al.: Output Resilient Containment Control of Heterogeneous Systems With Active Leaders Using RL Under Attack Inputs real applications, the full state information can not always be obtained in the controller design, and the dynamics and dimensions of MASs are different, the output containment control problem needs to be considered. The reference [31] solves the leader-follower output problem for heterogeneous MASs. the containment output control design problem is rarely studied which inspires our research. In heterogeneous MASs, an internal model principle is used to obtain the control protocol for containment control problem [32]–[34] This approach needs all knowledge of the dynamics of agent, and it is difficult to obtain it in some applications.

PROBLEM STATEMENT

OBSERVER PARAMETER DESIGN

IMPLEMENTING THE DATE-BASED OFF-POLICY RL USING NEURAL NETWORKS

SIMULATION

CONCLUSION