TL;DR
This paper introduces the first "soft drone" prototype, replacing rigid landing gear with a compliant, tendon-actuated soft gripper. This innovation enables dynamic grasping of unknown objects at higher speeds than previously possible, achieving a remarkable 91.3% success rate in real-world tests by leveraging adaptive control and the inherent benefits of soft robotics.
What Problem Does This Paper Address?
Aerial manipulators, particularly small quadrotors, face significant challenges when attempting to grasp objects. Traditional rigid grippers require precise positioning and transmit large contact forces, which can destabilize the drone and necessitate slow, cautious movements. This severely limits the ability of drones to perform "dynamic grasping" – grasping objects while in motion. Biological systems, like birds, overcome this by using compliant structures to absorb impact and accommodate inaccuracies. This paper aims to bridge this gap by developing a drone that incorporates soft robotics principles to achieve more robust and dynamic aerial manipulation.
Methods
The core of this work is the development of a novel "soft drone" platform. This involves:
- Soft Gripper Design: Replacing the quadrotor's rigid landing gear with a soft, tendon-actuated gripper composed of four silicone rubber fingers. This design mimics biological finger structures and is actuated by 16 tendons, controlled by 3D-printed winches.
- Control System: A two-pronged approach is used for control:
- Quadrotor Control: An adaptive geometric controller is employed. This controller is crucial for compensating for the dynamic changes introduced by the soft gripper and the grasped object, as well as unmodeled aerodynamic effects like ground effect.
- Soft Gripper Control: A feedforward controller is used in conjunction with trajectory optimization. The gripper's configuration is optimized offline for "approach" and "grasp" phases, with tendon lengths interpolated between these key points.
- Trajectory Optimization: A minimum-snap polynomial optimization is used to plan the quadrotor's trajectory, ensuring it reaches a feasible grasp pose. This is then coupled with a finite element model (FEM) to compute desired tendon lengths for the soft gripper.
Datasets
Not discussed in the paper. The evaluation relies on physics simulations and real-world experiments.
Key Insights and Ideas
- Morphological Computation in Action: The soft gripper's compliance acts as "morphological computation," passively absorbing errors and forces, reducing the need for explicit, high-precision control. This is a key enabler for dynamic grasping.
- Adaptive Control is Crucial: The adaptive geometric controller is essential for real-world success. It effectively compensates for unmodeled dynamics like the added mass of the grasped object and aerodynamic disturbances (e.g., ground effect), which standard controllers struggle with.
- Softness Enables Aggressive Grasping: The inherent compliance of the soft gripper allows it to tolerate imprecisions in positioning and absorb impact forces, enabling higher grasp velocities and more robust grasps compared to rigid counterparts.
- Simulation vs. Reality: While simulations (Unity-SOFA) provide a valuable testing ground, real-world tests revealed additional challenges like "thrust stealing" (where the object blocks airflow to propellers) and more pronounced ground effects, highlighting the complexity of real-world aerial manipulation.
Key Results
- High Success Rate: The physical prototype achieved a 91.3% success rate in dynamic grasping across 23 real-world trials.
- Dynamic Grasping at Speed: The soft drone successfully performed dynamic grasps at velocities up to 0.2 m/s in real tests, and up to 2 m/s in simulation, significantly outperforming rigid grippers at higher speeds.
- Improved Tracking: In simulation, the adaptive controller showed significantly lower tracking errors, especially after grasping, compared to a standard geometric controller.
- Rigid Gripper Limitations: In simulation, rigid grippers failed at higher velocities (1.5-2 m/s) due to unexpected contact or destabilizing forces, whereas the soft drone maintained stability.
- Consistency: The soft drone's performance remained consistent across multiple trials, even with adverse positioning errors.
Limitations and Future Work
- Speed Limitations: The current prototype's maximum speed is limited by factors like thrust saturation, state estimation delays, and soft gripper latency.
- Simulation Fidelity: The SOFA simulator, while powerful, had limitations with high-velocity contacts, clipping, and did not fully model complex aerodynamic effects like the ground effect.
- Thrust Stealing Effect: This phenomenon, where the grasped object obstructs propeller airflow, was observed in real tests and significantly impacted performance, especially with larger surface area objects or near maximum payload. This is a more complex disturbance than a simple wrench.
- Future Work: The authors suggest pushing for even higher grasping speeds (beyond 2 m/s) and developing systems capable of aggressive grasping on moving platforms, which presents a formidable challenge. Further research into design, algorithms, and fabrication is needed to meet the performance observed in nature.
Summary updated