Optimization of Machine Learning-Based Dynamic Torsional Control Strategies for Bionic Flapping-Wing Aircraft - Issue 1 - 2025

Aug 3, 2026

This paper studies a dynamic torsional control strategy for bionic flapping-wing aircraft using machine learning. It first highlights why dynamic torsion control is essential for maintaining optimal aerodynamic performance across flight conditions, then compares energy efficiency between passive and active torsion. The work analyzes limitations of traditional Deep Reinforcement Learning (DRL), including redundant state spaces and unstable training in wing control. To address these challenges, an improved DRL algorithm with an attention mechanism is proposed, with model design, simulation environment, and experimental setup detailed. Experimental discussions confirm the approach improves optimization effectiveness and supports future research directions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Overview of Dynamic Twist Control Strategies for Bio-inspired Flapping Wing Aircraft Based on Machine Learning

The Importance of Dynamic Twist Control in Bio-inspired Flapping Wing Aircraft

The Role of Machine Learning in Flapping Wing Aircraft Control

Global Academic Frontiers, Vol. 3, Issue 1, 2025, pp.1-10 Print ISSN 2995-5688 Online ISSN 2995-570X DOI: [https://doi.org/10.5281/zenodo.15074444](https://doi.org/10.5281/zenodo.15074444) Optimization of Machine Learning-Based Dynamic Torsional Control Strategies for Bionic Flapping-Wing Aircraft JiaJun Gan 1 XiaoGang Liang 1 Ting Yao 1 Yingzhi Peng 1,* 1 School of Science and Technology , Guilin University, Guangxi 541001, China *Corresponding author Email: [2022037@glc.edu.cn](2022037@glc.edu.cn) Received 5 March 2025; Accepted 17 March 2025; Published 24 March 2025 © 2025 The Author(s) . This is an open access article under the CC BY license. Abstract: This paper explores the dynamic torsion control strategy for bionic flapping-wing aircraft based on machine learning. Firstly, it outlines the importance of dynamic torsion control in bionic flapping-wing aircraft and the application of machine learning in this field. Subsequently, a comparative analysis of the energy efficiency of passive torsion and active torsion is conducted, and the challenges faced by traditional Deep Reinforcement Learning (DRL) in flapping-wing control are pointed out. To address these issues, this paper proposes an improved DRL algorithm incorporating an attention mechanism. The design of the new model, the establishment of the simulation environment, and the experimental setup are described in detail. Finally, through the analysis and discussion of the experimental results, the effectiveness of the improved algorithm in optimizing the dynamic torsion control of bionic flapping-wing aircraft is verified, providing insights for future work. Keywords: bionic flapping-wing aircraft; dynamic torsion control; machine learning; deep reinforcement learning; attention mechanism 1. Overview of Dynamic Twist Control Strategies for Bio-inspired Flapping Wing Aircraft Based on Machine Learning 1. 1 The Importance of Dynamic Twist Control in Bio-inspired Flapping Wing Aircraft In the field of bio-inspired flapping wing aircraft research, the optimization of dynamic twist control strategies is of paramount importance. This control strategy plays a crucial role in enhancing flight efficiency by adjusting the twist angle of the wings, enabling the aircraft to maintain optimal aerodynamic performance across various flight conditions. However, achieving dynamic twist control presents numerous challenges, including but not limited to accurately sensing flight states, real-time adjustment of twist angles, and ensuring control stability and robustness. In recent years, with the rapid advancement of machine learning technologies, particularly the widespread application of Deep Reinforcement Learning (DRL) methods, new solutions have emerged for optimizing dynamic twist control strategies in bio-inspired flapping wing aircraft. Nonetheless, traditional DRL methods often encounter issues such as redundant state spaces and unstable training when applied to wing control. These challenges significantly hinder the effectiveness of DRL in the control of bio-inspired flapping wing aircraft. It is noteworthy that recent developments in attention mechanisms within DRL offer new possibilities for addressing these issues. By incorporating attention mechanisms, DRL methods can more effectively focus on key state information relevant to flight control tasks, thereby reducing the redundancy in state spaces. Additionally, attention mechanisms contribute to enhancing training stability, allowing DRL methods to learn effective control strategies in a shorter timeframe. Therefore, the integration of attention mechanisms with DRL holds the potential to bring about groundbreaking advancements in the optimization of dynamic twist control strategies for bio-inspired flapping wing aircraft. 1.2 The Role of Machine Learning in Flapping Wing Aircraft Control In the research on optimizing dynamic twist control strategies for bio-inspired flapping wing aircraft, mach

FAQs

Why is dynamic torsional (twist) control important for bionic flapping-wing aircraft?

It adjusts wing twist angle to maintain near-optimal aerodynamic performance across different flight conditions, improving flight efficiency. The main difficulty is sensing flight states accurately and updating twist in real time while preserving stability and robustness.

What are the key drawbacks of applying traditional DRL to flapping-wing control?

Traditional DRL may suffer from redundant state spaces, which increases computation and harms real-time performance, and from unstable training caused by factors such as environmental noise, model complexity, and reward-function design.

How does the proposed method improve DRL for dynamic torsional control?

The paper introduces an improved DRL algorithm that incorporates an attention mechanism. This helps the agent focus on key state information, reducing state redundancy and enhancing training stability, leading to more effective optimization.

Aug 3, 2026

Upload by River Wang