Co-Design of Control, Computation, and Network Scheduling Based on Reinforcement Learning.

Computationally intensive control tasks, especially the control of image-based mobile controllers, usually require more computational capacity and network transmission resources than traditional control tasks due to the assembly of camera-based visual perception sensors. As a result, the collaborative design of network, computation, and control is of great importance to improve the efficiency of resource usage while ensuring control performance, thus achieving overall system optimality. In this paper, we proposed a synergistic algorithm for network and computational resource scheduling through Deep Deterministic Policy Gradient and control based on self-triggered model predictive control, which provides optimal resource scheduling instructions after observing the system state in real-time, and uses these resources to complete control tasks in the controller with a self-triggering mechanism to achieve the goal of ensuring control performance and reducing energy consumption. Finally, we designed numerical simulation experiments to verify the effectiveness of the algorithm proposed in this paper.