Deep Reinforcement Learning-Guided Bio-Inspired Active Flow Control of a Flapping-Wing Drone for Real-Time Disturbance Suppression

Flapping-wing drones (FWDs), owing to their compact size and operation in cluttered and unsteady airflow environments, encounter significant aerodynamic and stability challenges. Studies of avian flight reveal that falcons and other raptors actively deflect their covert feathers to mitigate gusts and maintain stable flight. Drawing inspiration from this mechanism, this study presents a peregrine falcon-inspired Active Flow Control Unit (AFCU) integrated with a Deep Deterministic Policy Gradient (DDPG)-based deep reinforcement learning (DRL) controller for real-time disturbance attenuation. The AFCU employs mechanical covert feathers (MCFs) that actuate to dissipate gust loads during high wind conditions. A reduced-order bond graph model that encapsulates the nonlinear interaction between the primary wing and the feather-based active flow control surfaces is created which is used as a dynamic training environment for the DDPG agent. Utilizing closed-loop interactions, the successfully obtained learned policy produces optimal actuator forces to reduce feather-displacement error and aerodynamic load variations. The designed controller stabilizes the internally unstable open-loop AFCU, attaining near-zero steady-state error and settling times under 1.6 s for gust magnitudes ranging from 12.5 to 20 m/s. Simulations further illustrate a reduction of up to 50% in gust-induced loads compared to traditional approaches. This integration of bio-inspired design with learning-based active flow control offers a viable avenue for the development of highly adaptive and gust-resilient flapping-wing aerial systems.