Drones often hover in the air noisy and whining. They can already be used for many tasks, but not when you need some quiet. Their batteries would also last longer if they could take a 'rest' from time-to-time. But when a drone is monitoring a rainforest, its cluttered surroundings and own gripper arm become too tricky for its camera vision. Until now. Researchers at TU Delft (The Netherlands) have developed a drone that does something more like a bird: it feels for a branch before landing. Using soft tactile sensors embedded in a human-like hand, the drone detects contact with a branch, learns where the branch is and how it is oriented, adjusts its position and grips it securely. Time to switch off the motors and enjoy some rest.
Published in Nature npj Robotics, this study introduces a new approach to autonomous perching in which touch becomes a source of information for action.
Search, touch, align, grasp
Traditionally, drones have relied heavily on vision to understand their surroundings. They need a clear view or a precise model of a branch before attempting to land. Drones that have vision become "blind" as they approach a target, because grippers and branches block their view. Basically, all existing studies that use this vision approach not as practical in the real world.
Instead, this tactile drone starts with only a rough estimate of where the target might be and refines that estimate through touch. The drone flies a figure-eight search pattern while opening and closing its three-fingered hand. As the fingers open and close, the first contact reveals the branch's position and orientation. The drone rotates and repositions itself until the sensors on all three fingers confirm stable contact and then switches off its motors. If the grasp fails, the drone retreats to a safe hover and tries again.
Reaching before seeing
The idea was inspired by the way animals interact with the world. Birds use touch to guide the final moments of a landing, sensing a branch through their feet and adjusting their grip in real time. Associate Professor Dr. Salua Hamaza wanted to bring this capability to aerial robots.
"Vision tells you where something is until the moment it matters most — and then your own gripper gets in the way. Animals do not have this problem, because they close the loop through touch. That is what we wanted to give a drone: the ability to reach into a space it cannot see, feel what is there, and correct itself while it is already in contact." — Associate Professor Salua Hamaza, TU Delft.
The result is a lightweight, anthropomorphic hand designed to fly. It has three fingers, each with three phalanges (segments) inspired by the proportions of the human hand. Torsional springs naturally close the fingers, allowing the drone to maintain its grip without using energy once perched. A single tendon per finger opens the fingers, while soft silicone pads provide friction and allow the hand to comply with different surfaces and textures.
Nine touch sensors across the robotic hand
Each phalange carries a tactile sensor: a small copper electrode beneath the silicone surface. When it touches an object, the electrical signal changes, producing a simple yes-or-no contact signal. On its own, this signal gives limited information. But because the drone knows the shape and position of its fingers, it can determine exactly where each touch occurs. A simple touch signal becomes a point in space and a direction to move.
"Each sensor only tells us 'something is here'. But if you know the shape of your own hand, that is enough to reconstruct where the branch is and how it is oriented. The drone builds up an increasingly accurate "picture" of the target purely by bumping into it." Anton Bredenbeck, TU Delft.
Touch as an asset, not a liability
The approach is robust because it uses touch to correct errors in perception. In flight experiments, the researchers found the drone remained reliable even when the target position, orientation, and size were wrongly estimated, while a conventional vision-based strategy quickly failed. The work points toward drones that can monitor environments for extended periods, even where vision alone is unreliable, such as forest canopies, cluttered industrial structures, or places without motion-capture systems.
More broadly, the study suggests a shift in thinking about contact in aerial robotics: not as a risk to avoid; but as a source of information to guide action. Just as in nature, future drones may use touch to navigate the world and find places to land that they cannot yet fully see.
