Collaborative Task-Oriented Communication-Control Co-Design for Wireless Networked Control Systems

Abstract

Wireless networked control systems (WNCSs) are increasingly deployed in industrial applications due to their inherent flexibility and scalability. However, achieving high-performance collaborative tasks remains a significant challenge due to limited wireless resources as well as harsh and dynamic industrial environments. To address these challenges, this paper proposes a collaborative task-oriented communication-control co-design (TC4) method for WNCSs via deep reinforcement learning. To this end, a long short-term memory-based estimation model is first trained for each subsystem to provide state compensation when packet losses occur. Building upon the pre-trained estimation models, a Transformer-based collaborative control model for the overall WNCS is designed to capture the dependencies among subsystems. A two-stage curriculum learning strategy is proposed to enhance the training stability of the control model. To efficiently allocate limited wireless resources, we further develop a scalable scheduling model by leveraging a Transformer-based architecture and a task-aware scoring mechanism. Consequently, the synergistic operation of these pre-trained models enables the overall WNCS to effectively execute the designated collaborative task. Extensive simulations on a collaborative transportation task demonstrate that TC4 outperforms baseline methods in terms of trajectory tracking error and formation error.