A novel approach to tune Fuzzy-PID membership function based on reinforcement learning

Designing and tuning the membership function parameters of a Fuzzy-PID controller can be challenging in practice, as they are often adjusted through trial and error. In this paper, we propose an adaptive Fuzzy-PID controller for UAVs that integrates fuzzy logic with reinforcement learning algorithms. Fuzzy systems are well known for their ability to handle nonlinear, complex, and uncertain dynamics, while reinforcement learning algorithms have demonstrated strong performance in managing nonlinear systems with uncertainties and enhancing overall model efficiency. Unlike traditional methods that rely on heuristic adjustments or expert knowledge to determine fuzzy system parameters, our approach exploits the complementary advantages of fuzzy logic and reinforcement learning to automatically optimize membership function parameters without trial and error. Furthermore, to overcome the challenge of slow learning processes, the number of variables under investigation is reduced, thereby decreasing computational complexity. The proposed approach significantly improves the performance of the fuzzy inference system, achieving superior control accuracy with minimal control effort.