Flying insects are quite good at maneuvering through tight spots, but they often run into obstacles especially windows. As unique as insects are, there is hardly any space on their bodies to mount fancy computing and sensing systems to help improve their flight. A group of scientists have researched the flying habits of insects particularly how bees are able to perform precise flight maneuvers and quickly fly through tiny holes to access their nests.

Studies show that bees have some sort of minimalist brute force approach to getting through tiny spaces.  They fly up to a small gap or hole, hover around it while collecting visual information on where the edges are, then steer themselves through. It is not a fast or complicated process, but it is reliable for the bees and easy to execute. This reliability and ease of execution is what is drawing scientists working on the next generation of drones.

The Perception and Robotics Group from the University of Maryland have drawn inspiration from the bees and are developing a system that will allow drones to fly through small and completely unknown gaps using just one camera with an onboard processor.  While drones flying through small gaps isn’t really a new concept, this new system is drastically different as the drones have no information about the size or location of the holes or gaps in advance. The drone doesn’t rely on a 3D map or model of the hole which is almost impossible to get in real time using a single camera.

Instead, the drones use an optical flow technique. Just like the bee, the drones hover around the gap while taking photos. Its onboard processors then analyze the pictures. The drones then use this information they processed to pass cleanly through the hole. In this system, the drone takes pictures, moves forward, takes more pictures, and then adjusts the movement to ensure its position is perfect. The system offers something a little different from the pre-planned approach where drones know exactly where and how to fly before taking flight.

The researchers tested the new system using a Bebop 2 drone with an Nvidia TX2 module mounted on it. Holes with varying shapes and sizes were cut in the testing wall. After over 150 trials, the researchers had a success rate of 85%. The maximum speed the drones could achieve while going through the holes was 2.5 meters per second which the researchers believe can be improved after more tests.