Adaptive PID Tuning for Quadcopters in Dynamic Environments Using Advantage Actor-Critic Reinforcement Learning

Achieving precise and adaptive control of quadcopter systems is essential for aerial robotics, autonomous navigation, and dynamic environment interaction. This study investigates the implementation of the Advantage Actor-Critic (A2C) algorithm, a prominent deep reinforcement learning (DRL) technique, for the online optimization of Proportional-Integral-Derivative (PID) controllers. The proposed framework autonomously adjusts PID gains through real-time interaction with a simulated environment, facilitating adaptive control under highly dynamic and uncertain conditions. Rigorous validation was conducted using simulated flight paths defined by discrete waypoints, designed to emulate complex operational scenarios. These scenarios incorporated varying payloads, external disturbances such as environmental noise, and significant mass uncertainties, illustrating the robustness of the approach. By utilizing a continuous state-action space, the A2C algorithm ensures effective adaptation and stability within the multi-input multi-output (MIMO) dynamics inherent to quadcopter control. The experimental results underscore the potential of DRL-enhanced control mechanisms for real-time, noise-resilient operation, offering a pathway toward increased autonomy, reliability, and efficiency in aerial systems.