HoLoArm: Deformable Arms for Collision-Tolerant Quadrotor Flight

Technical Summary: HoLoArm: Deformable Arms for Collision-Tolerant Quadrotor Flight

Problem Statement
The increasing deployment of Unmanned Aerial Vehicles (UAVs) in human-centric and cluttered environments necessitates designs that can survive physical collisions and recover rapidly. While traditional approaches rely on sensing and path planning for collision avoidance, collisions remain inevitable due to sensor limitations (e.g., fog, rain, acoustic absorption) and the inherent need for physical interaction in tasks like tactile navigation or landing. Existing mechanical solutions, such as protective cages or rigid frames, often add excessive weight, reduce flight efficiency, or fail to absorb impact energy effectively. Although some recent works have explored soft-bodied robots or compliant joints, challenges remain in balancing collision tolerance with structural simplicity, scalability to larger propeller sizes, and the ability to absorb axial (longitudinal) impact forces. Furthermore, modeling the complex, nonlinear dynamics of hybrid soft-rigid structures for traditional control remains difficult.

Methodology
The authors propose HoLoArm, a quadrotor platform featuring deformable arms inspired by the nodus structure found in dragonfly wings. The design philosophy prioritizes passive mechanical resilience over active morphing mechanisms.

• Hardware Design: The arm structure consists of eight interlinked components. The core innovation is a compliant joint modeled after the insect nodus, which includes a sliding bearing (LM88U) allowing horizontal movement along the arm's longitudinal axis. This is coupled with four elastic ligaments made of Thermoplastic Polyurethane (TPU) that provide restoring forces. The joint is asymmetrically designed to prevent gravity-induced sagging and thrust-induced bending. The main body and arms utilize PLA for stiffness, while the propeller guards use a carbon fiber/PLA composite. The platform supports propellers from 6 to 13 inches; the experimental prototype uses 9-inch propellers, a T-MOTOR Ultralight P2505 motor configuration, and a Pixracer Pro flight controller.
• Control Strategy: Due to the difficulty of accurately modeling the nonlinear interactions between the compliant arms and rigid components, the authors employ a Reinforcement Learning (RL) policy. The policy is trained in a simulator where dynamic parameters are identified online. It takes state estimates (position, orientation, linear/angular velocity) and action history as input, mapping them directly to motor setpoints. This model-free approach allows the controller to adapt to the hardware's complexity without requiring a precise mathematical model.
• Experimental Validation: The system was tested using a Vicon motion capture system (200 Hz) to quantify recovery times and deformation. Experiments included trajectory tracking, payload lifting, mid-flight collision robustness (comparing against a rigid drone), narrow-gap navigation, and high-velocity drop tests.

Key Contributions

1. Bio-inspired Mechanism: A soft-armed quadrotor design integrating a nodus-inspired compliant joint that enables passive deformation and impact absorption in multiple directions, including axial loads.
2. Scalable Modularity: A modular arm structure capable of accommodating propellers from 6 to 13 inches, simplifying fabrication and repair.
3. RL Control Integration: The successful application of a Reinforcement Learning policy to stabilize a quadrotor with significant structural compliance, bypassing the need for complex system identification.
4. Experimental Validation: Comprehensive comparative testing demonstrating superior survivability and recovery performance against rigid counterparts under identical impact conditions.

Results

• Recovery Performance: The HoLoArm can passively deform in any direction and recover to a safe flight state (within 1° of original orientation) within 0.3 to 0.6 seconds, depending on the impact direction and magnitude. Specifically, recovery from lateral/vertical impacts took approximately 0.27–0.72 seconds, while axial compression recovery took roughly 0.75 seconds.
• Collision Survivability: The drone survived collisions at speeds up to 7.6 m/s. In drop tests from 3 meters (impact velocity ~7.7 m/s), the rigid drone suffered catastrophic failure, whereas the HoLoArm remained intact. Impact force measurements showed the HoLoArm generated significantly lower peak forces (e.g., 16.52 N vs. 23.99 N at 3m drop) compared to the rigid frame.
• Payload and Agility: The platform successfully carried a 540 g payload (total weight ~970 g with battery) while maintaining stable flight and tracking lemniscate trajectories with mean errors of 0.23 m (slow) and 0.52 m (normal).
• Narrow-Gap Navigation: Unlike the rigid drone, which became trapped, the HoLoArm successfully deformed upon contact with obstacles to traverse gaps narrower than its nominal width and recovered its shape immediately after.

Significance
The paper claims that HoLoArm contributes to the morphological design of soft aerial robots by demonstrating that high agility and reliable safety can be achieved through passive mechanical compliance rather than complex active actuation. The integration of a bio-inspired nodus structure allows for the absorption of multi-directional impacts, including difficult-to-manage axial forces, while maintaining structural simplicity. The work establishes a foundation for future fully soft drones that combine compliant structures with intelligent control, enabling operation in cluttered and human-shared environments where collision tolerance is critical. The authors note that while the current design uses a hybrid soft-rigid approach, it lays the groundwork for future iterations that may integrate fully compliant structures with safety-oriented control algorithms.