Mimic biological flapping motion for a two-dimensional wing by reinforcement learning

Birds, insects, bats and fish demonstrate exceptional locomotion efficiency through adaptive flapping motions, offering a wealth of inspiration for bio-inspired propulsion systems. However, traditional research often relies on simplified motion models with limited degrees of freedom, which may not fully capture the complexity, adaptability, and efficiency of natural movement. In this study, we propose an adaptive motion optimization framework based on reinforcement learning (RL), aiming to address the aforementioned challenges. By integrating high-fidelity numerical simulations with physical models of flapping wings, the framework dynamically adjusts motion patterns in real time, guided by flow field information. Departing from conventional methods that rely on pre-designed motion assumptions, this approach uncovers non-harmonic, quasi-periodic motion patterns through iterative exploration. The system refines behaviors to enhance propulsion performance, adapt to dynamic flow conditions, and reveal biologically relevant features, such as asymmetric oscillations, adaptive rhythmic formations, and progressive fine-tuning of motion strategies. These learned motions not only align with natural flapping characteristics but also surpass traditional optimization methods by expanding the search space to include more complex and effective movement patterns. This framework demonstrates the power of RL to discover sophisticated, bio-inspired motion dynamics, offering transformative potential for understanding natural flapping mechanisms and designing efficient, versatile propulsion systems for real-world applications.