A highly maneuverable flying squirrel drone with agility-improving foldable wings

TL;DR

The paper tackles drone agility limitations caused by limited thrust by introducing a flying squirrel–inspired drone with foldable silicone wings and a TWCC framework that coordinates thrust with wing-generated drag. It combines a physics-informed recurrent neural network (paRNN) to model wing aerodynamics with a TWCC controller that selects when to deploy wings via WOEG and WIIC, enabling stable, high-performance trajectory tracking in agile scenarios. The main contributions are the paRNN-based aero-learning for silicone wings and the TWCC architecture that leverages learned wing dynamics for real-time control, validated by outdoor experiments showing a 13.1% RMSE improvement over wingless baselines and strong obstacle-avoidance performance. This biomimetic approach can enhance outdoor UAV operation in cluttered environments where rapid, stable maneuvers are essential.

Abstract

Drones, like most airborne aerial vehicles, face inherent disadvantages in achieving agile flight due to their limited thrust capabilities. These physical constraints cannot be fully addressed through advancements in control algorithms alone. Drawing inspiration from the winged flying squirrel, this paper proposes a highly maneuverable drone equipped with agility-enhancing foldable wings. By leveraging collaborative control between the conventional propeller system and the foldable wings-coordinated through the Thrust-Wing Coordination Control (TWCC) framework-the controllable acceleration set is expanded, enabling the generation of abrupt vertical forces that are unachievable with traditional wingless drones. The complex aerodynamics of the foldable wings are modeled using a physics-assisted recurrent neural network (paRNN), which calibrates the angle of attack (AOA) to align with the real aerodynamic behavior of the wings. The additional air resistance generated by appropriately deploying these wings significantly improves the tracking performance of the proposed "flying squirrel" drone. The model is trained on real flight data and incorporates flat-plate aerodynamic principles. Experimental results demonstrate that the proposed flying squirrel drone achieves a 13.1% improvement in tracking performance, as measured by root mean square error (RMSE), compared to a conventional wingless drone. A demonstration video is available on YouTube: https://youtu.be/O8nrip18azY.

Figures (11)

• Figure 1: Overview of the flying squirrel drone. The frames used (Body, World, GNSS) can be found in A, and the actual image of the drone can be seen in B.
• Figure 2: Overview of TWCC. The structural diagram illustrates how the components of TWCC—PC, WOEG, and WIIC—operate. As shown in the figure, details related to TWCC can be found in Section III.B, while information on the Attitude Controller is provided in Section III.A.
• Figure 3: 3D and projected 2D flight trajectories of a drone (a) Trajectory for the proposed neural network(NN) based method (b)(c)(d) Projected 2D trajectories for the wingless pure thrust control method, the model based one, and the proposed NN based one.
• Figure 4: A virtual plane and its corresponding estimated AOA, $\alpha_{est}.$
• Figure 5: RNN-based learning of air resistance from state variables of a drone.