DRL-Based Resource Scheduling for Control System in Heterogeneous Environment

We investigate the scheduling problem of control tasks in heterogeneous environments. We measure the computational resource required for linear quadratic regulators (LQR) and model the scheduling problem as the bin packing problem. We formulate the bin packing problem as a Markov Decision Process. In the Markov Decision Process, we generate two action subspaces within the action space and propose an algorithm called DARL based on deep reinforcement learning. Experimental results demonstrate that the DARL has improved training efficiency and learning performance compared to base-lines, effectively addressing the challenges of resource scheduling of control tasks in heterogeneous environments.