Researchers at TU Delft have developed a drone that uses tactile sensors to locate and grip branches, allowing it to perch without relying entirely on camera-based perception. Once attached, the drone can switch off its motors, reducing energy consumption during periods when it does not need to remain airborne.
The system combines a quadrotor with a three-fingered robotic hand equipped with nine tactile sensors. Each finger consists of three segments based on the proportions of a human hand. The sensors detect physical contact and allow the drone to estimate the position and orientation of a branch as it interacts with it.
Rather than requiring an accurate visual model of the target before approaching it, the drone starts with an approximate location. It then flies a figure-eight search pattern while repeatedly opening and closing its gripper. When one of the sensors detects contact, the drone uses information about the position and geometry of its fingers to calculate where the contact occurred.
It can subsequently rotate and reposition itself until sensors on all three fingers register contact with the branch. The drone then closes its grip and switches off its motors. If the attempt fails, it returns to a hovering position and can make another attempt.
The approach is intended to address a limitation of vision-based perching systems. Cameras can provide information about a target during an approach, but the branch or the drone’s own gripper can obstruct the camera as the distance decreases. Tactile sensing instead provides information when the drone is already physically interacting with the target.
“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,” said TU Delft researcher Anton Bredenbeck.
The gripper uses torsional springs to close its fingers naturally. A tendon mechanism opens each finger, while silicone pads provide friction and allow the fingers to conform to different surfaces. Because the springs maintain the grip mechanically, the drone does not need to continuously expend energy to remain attached after perching.
During flight experiments, the researchers tested the system when estimates of a target’s position, orientation and size were inaccurate. According to the study, the tactile approach could use contact information to correct these errors, while a conventional strategy based on previously estimated target information was less tolerant of inaccuracies.
The researchers see potential applications in environments where visual sensing is restricted or where drones may need to remain deployed for extended periods, including forest canopies and cluttered industrial structures.
The research by Bredenbeck, Anish Jadoenathmisier and Salua Hamaza was published in npj Robotics. It examines tactile-driven perching as a method for allowing aerial robots to use physical contact as part of their perception and control process.
Photo credit Delft University of Technology – Studio Oostrum
