Model‐free self‐triggered control based on deep reinforcement learning for unknown nonlinear systems

Abstract

This article proposes a joint learning technique for control inputs and triggering intervals of self‐triggered control nonlinear systems with unknown dynamics. First, deep reinforcement learning is introduced to the self‐triggered control system by considering both the control performance and triggering performance in the reward function. Then, the control inputs and triggering intervals are simultaneously learned by the developed deep deterministic policy gradient approach. Under this strategy, not only the desired control performance is guaranteed for unknown nonlinear systems, but also both the computation and communication occupation for the controlled system are decreased without any triggering thresholds. Finally, simulations for the cart‐pole swing‐up system are illustrated to verify the effectiveness of the proposed scheme.

Controlled descent training

“…Problems in the domain of Control Theory have also had a surge in ANN and ML related works. [6][7][8] The main criticisms of ANNs are their unpredictable training behavior, low output interpretability, and highly hyper-parameter dependent performance.…”

“…ANN architectures have made strides in recent years in solving complicated tasks, like image recognition, 1 natural language processing, 2 artificial image generation, 3 and other engineering tasks 4,5 where classical statistical models might struggle. Problems in the domain of Control Theory have also had a surge in ANN and ML related works 6–8 . The main criticisms of ANNs are their unpredictable training behavior, low output interpretability, and highly hyper‐parameter dependent performance.…”

Data-Driven Self-Triggered Control for Networked Motor Control Systems Using RNNs and Pre-Training: A Hierarchical Reinforcement Learning Framework

“…It is a machine learning approach where an algorithm learns to make decisions by interacting with an environment, using trial and error to discover the best strategy. In the context of model-free, self-triggered control based on reinforcement learning, the controller continuously adjusts its strategy based on the feedback costs from its interaction with the plant, thereby developing a self-triggered control strategy that generates both control actions and triggering intervals [ 17 , 18 , 19 ]. However, these previous studies often focused on designing appropriate rewards for self-triggered control systems, neglecting the hierarchical structure of the self-triggered control strategies.…”