Self-Tuning PID Controller for an Arm-Angle Adjustable Quadrotor Using an Actor-Critic-Based Neural Network

The increasing prevalence of multirotors necessitates the development of new configurations for specific applications. Innovative morphing of multirotors has enhanced their stability, maneuverability, agility, and overall performance. The utilization of Artificial Intelligence (AI) in robotics has enabled robots to exhibit autonomous behaviors in areas such as control, path planning, state estimation, system identification, and image processing. This paper introduces a novel approach to improving the control system of an adjustable arm-angle quadrotor through the integration of a self-tuning Proportional-Integral-Derivative (PID) controller, utilizing an Actor-Critic-based Neural Network (NN). Arm-angle adjustable quadrotors, with their complex and variable dynamics, present challenges in achieving optimal control performance across different arm angles. Traditional PID controllers often struggle to adapt efficiently to these dynamics. To address this issue, the Actor-Critic framework from reinforcement learning is employed to develop a self-tuning mechanism for the PID controller. The Actor network learns the optimal control actions, while the Critic network evaluates the performance of the control system, enabling continuous improvement. Extensive Software-in-the-Loop (SIL) and Hardware-in-the-Loop (HIL) tests demonstrate the effectiveness and feasibility of the proposed approach in achieving robust and adaptive control for tilt-arm quadrotors under diverse operating conditions. The results highlight the superiority of the Actor-Critic-based self-tuning PID controller over conventional methods, showcasing its potential for real-world devlopment in autonomous aerial systems.