Tendon-driven Grasper Design for Aerial Robot Perching on Tree Branches

Haichuan Li, Ziang Zhao, Ziniu Wu, Parth Potdar, Long Tran, Ali Tahir Karasahin, Shane Windsor, Stephen G. Burrow, Basaran Bahadir Kocer

TL;DR

Researchers have developed a bio-inspired, tendon-driven gripper that allows aerial drones to perch on tree branches, mimicking the energy-efficient "hanging" behavior of bats. By combining a compliant mechanical design with a deep-learning vision system that identifies stable perching spots, the platform can securely attach to irregular branches (30–80 mm diameter) without consuming power to maintain its grip.

What Problem Does This Paper Address?

Monitoring and restoring forest ecosystems is critical for conservation, but it’s a logistical nightmare for drones. Dense canopies and complex terrains make navigation difficult, and the limited battery life of multi-rotors restricts them to short missions.

While "perching" (staying stationary on a structure) is a known solution to save energy, most existing mechanisms are designed for flat, predictable surfaces like walls or ceilings. Tree branches are irregular, vary wildly in diameter, and are often obscured by leaves. Furthermore, many grippers require constant motor power to stay closed. This paper tackles the challenge of creating a system that can autonomously find a suitable branch and securely hold on using a passive, energy-efficient mechanism.

Methods

The authors propose a holistic system integrating bio-inspired hardware with a specialized software stack:

  • Tendon-Driven Mechanism: Inspired by bat claws, the gripper uses a "passive-closed" design. Stiff thermoplastic polyurethane (TPU) joints act like springs, keeping the gripper naturally closed. A single Dynamixel servo pulls nylon tendons to open the claw; once the drone is in position, the servo releases, and the mechanical tension of the joints secures the branch. This means zero energy is required to stay perched.
  • Modular Friction Pads: The gripper features "fingerprint" modules—soft, textured pads that increase friction and allow the claw to interlock with bark irregularities.
  • Raptor-Inspired Vision: To find a perch, the system uses a DeepLabV3+ semantic segmentation network (with a ResNet-34 backbone). It mimics birds of prey by first identifying a stable tree trunk and then locating the nearest horizontal branch suitable for perching.
  • Trajectory Planning: The robot uses a minimum-snap trajectory optimization (7th-order polynomial) to ensure a smooth, precise approach to the branch, which is critical for avoiding collisions in cluttered environments.

Datasets

The vision system was trained on a specific tree dataset (referenced as [21] in the paper, the "Urban street tree dataset"), which contains 1,385 images across 13 different classes. The authors used this to train the model to distinguish between trunks, branches, and background clutter, achieving a segmentation performance of 72.15% IoU (Intersection over Union).

Key Insights and Ideas

  • Energy Efficiency through Passivity: The most significant "aha" moment is the use of equilibrium states. By designing the gripper to be naturally closed, the drone only spends energy during the brief moments of actuation (opening/closing), not during the hours it might spend perching for data collection.
  • The "Squeezing" vs. "Hanging" Transition: The paper provides a detailed static model showing how the physics of the grip change with branch diameter. On small branches (30-50mm), the claw "hangs" via geometry; on larger branches (up to 110mm), the "squeezing force" from the elastic joints becomes the primary stabilizer.
  • Optimal Sensing Distance: Through simulation, the team discovered a "Goldilocks zone" for vision. At 1 meter, the camera sees too little of the tree; at 6 meters, the branches are too small to resolve accurately. 4 meters was identified as the ideal distance for reliable target selection.

Key Results

  • Branch Versatility: The mechanism successfully perched on branches ranging from 30 mm to 80 mm in diameter in real-world tests.
  • Targeting Accuracy: The vision system successfully separated trunks from branches, allowing the robot to output precise 3D coordinates for the perch.
  • Flight Precision: The quadcopter maintained a mean 3D position tracking error of 0.122 m (approx. 12 cm) during its approach, which was within the physical tolerance of the compliant gripper.
  • Payload Capacity: Static analysis showed the gripper could theoretically support weights significantly higher than the current 1.5 kg platform, though effectiveness drops off as branch diameter exceeds 102 mm.

Limitations and Future Work

While promising, the authors note several areas for improvement:

  • Visual Clutter: The current vision algorithm performs best against relatively "clean" backgrounds. In a truly dense forest, distinguishing a specific branch from a sea of green leaves remains a challenge.
  • Statistical Significance: The real-world tests were successful but limited in number. Future work will involve large-scale testing to quantify success rates across diverse tree species.
  • Material Durability: The rigid claw tips, while effective, are prone to wear or breaking after repeated engagements with rough bark. The authors suggest moving toward tougher, less brittle materials for long-term deployment.

Summary updated

BibTeX
@inproceedings{li2025tendon,
  author = {Haichuan Li and Ziang Zhao and Ziniu Wu and Parth Potdar and Long Tran and Ali Tahir Karasahin and Shane Windsor and Stephen G. Burrow and Basaran Bahadir Kocer},
  title = {{Tendon-driven Grasper Design for Aerial Robot Perching on Tree Branches}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2025},
  doi = {10.1109/iros60139.2025.11247040},
}
Figure