Reinforcement learning-based dynamic event-triggered control for nonlinear stochastic multiagent systems under mismatched disturbances

On Optimal Event-Triggered Distributed Control for Stochastic Multi-Agent Systems via Reinforcement Learning

This work proposes a reinforcement learning (RL) based optimal distributed control algorithm for the multi-agent systems (MASs) with stochastic uncertainties that uses the actor-critic-identifier structure and provides a Lyapunov-based stability proof that guarantees all errors are bounded, ensuring precise tracking between the leader and followers.

Event-Triggered Data-Based Reinforcement Learning Control for Optimal Consensus of Heterogeneous Multi-Agent Systems in Graphical Games

This study discusses the optimal consensus issue for heterogeneous multi-agent systems (MASs) defined by partially unknown dynamics within a graphical game framework. Data-driven reinforcement learning has shown efficacy in such systems; however, conventional implementations frequently depend on continuous time data transmission, which puts too much strain on computers and communication systems. This paper proposes a new event-triggered, data-based reinforcement learning control scheme to fix these problems. By adding an event-triggered mechanism (ETM) to the heterogeneous graphical game formulation, the control protocol is only updated when certain error thresholds are crossed. This uses much less resources than time-triggered methods. An off-policy integral reinforcement learning (IRL) algorithm is devised to ascertain the Nash equilibrium solution utilizing quantifiable system data, thereby obviating the necessity for precise knowledge of the internal system matrices. A theoretical analysis using Lyapunov stability theory shows that the suggested event-triggered strategy ensures asymptotic consensus and convergence to the best control weights. Finally, numerical simulation examples show that the proposed approach works well and is better than other methods. These examples show that the controller update frequency and communication load can be greatly reduced while keeping the system stable and optimal.

Optimal Containment of Multiagent Systems With Multistep Policy Gradient Reinforcement Learning.

This article analyzes the optimal containment control problem of discrete-time multiagent systems (MASs). Multistep temporal difference (TD) learning is integrated with policy gradient (PG) reinforcement learning (RL) to form an online off-policy multistep PG (MS-PG) algorithm. The proposed MS-PG algorithm achieves optimal control performance under completely unknown system dynamics and accommodates asynchronous policy updates among agents. The closed-loop system stability and the algorithmic convergence are rigorously established. Furthermore, an actor-critic neural network (NN) architecture is employed to approximate the control policy and the optimal Q-function, respectively, with data-driven weight update laws derived from the proposed algorithm. To improve training efficiency and sample efficiency, an experience replay (ER) mechanism is incorporated, constructing a hybrid learning framework that fully exploits both offline batch data and online operational data. Finally, simulation results verify the effectiveness of the proposed method.

Dynamic Event-Triggered Adaptive Output-Feedback Control for Uncertain Switched Nonlinear Systems via a Prescribed-Time Low-Complexity Approach

Disturbance-Rejecting Consensus Control of Multi-agent Systems Based on Hierarchical Disturbance Observation and Adaptive Event Triggering

RL-Driven Self-Triggered Optimal Stabilization for Unknown Nonlinear Networked Industrial Systems: An Adaptive Fuzzy Identification Design

This article addresses the optimal stabilization of unknown nonlinear networked industrial systems (NISs) with limited communication resources, proposing an innovative self-triggered approximate optimal control framework fused with generalized fuzzy hyperbolic model (GFHM) and reinforcement learning (RL) gradient descent. To handle unknown dynamics without prior model information, a GFHM-based state identifier is constructed, leveraging its universal approximation, flexible structure, and fewer parameters to adaptively reconstruct nonlinear dynamics. A self-triggered mechanism predicts the next control update instant via software, eliminating continuous hardware monitoring and reducing online computing cost, network data transmission overhead, and actuator energy usage. Within the RL framework, a critic neural network (NN) is established, with weights tuned via gradient descent to approximate the Hamilton-Jacobi-Bellman (HJB) equation solution and derive the approximate optimal control policy. Theoretical analysis proves the closed-loop signals are uniformly ultimately bounded (UUB). Numerical simulations validate the scheme's effectiveness in ensuring optimal control performance while saving resources.