Reinforcement Learning-based Optimal Control and Software Rejuvenation for Safe and Efficient UAV Navigation

Unmanned autonomous vehicles (UAVs) rely on effective path planning and tracking control to accomplish complex tasks in various domains. Reinforcement Learning (RL) methods are becoming increasingly popular in control applications, as they can learn from data and deal with unmodelled dynamics. Cyber-physical systems (CPSs), such as UAVs, integrate sensing, network communication, control, and computation to solve challenging problems. In this context, Software Rejuvenation (SR) is a protection mechanism that refreshes the control software to mitigate cyber-attacks, but it can affect the tracking controller's performance due to discrepancies between the control software and the physical system state. Traditional approaches to mitigate this effect are conservative, hindering the overall system performance. In this paper, we propose a novel approach that incorporates Deep Reinforcement Learning (Deep RL) into SR to design a safe and high-performing tracking controller. Our approach optimizes safety and performance, and we demonstrate its effectiveness during UAV simulations. We compare our approach with traditional methods and show that it improves the system's performance while maintaining safety constraints. Our approach takes 10 seconds less to reach the goal and we interpret this enhancement through a p-norm analysis.