Every drone operator has at least had a momentary pang of fear at the prospect of their beloved UAV smashing itself to smithereens after failing to navigate sufficiently well through a space.
In a new project, researchers at the University of Zurich in Switzerland have tried to come up with a way to help counter this — by transforming regular drones into aerial daredevils capable of maneuvering themselves through tight gaps without having to slow down in the process.
To accomplish this death-defying feat, the drones use a forward-facing fisheye camera and some smart planning algorithms. Using these tools it is able to pull off its mission more than 80 percent of the time.
“The goal of this work is to traverse narrow gaps with a quadrotor using only onboard sensing and computation,” researcher Davide Falanga told Digital Trends. “More specifically, we use only a camera, an inertial measurement unit, and an onboard computer to run the perception, planning and control algorithms that let the vehicle execute this agile maneuver.”
As you can imagine, getting quadcopters to fly through narrow, inclined gaps is a very challenging task that — due to the high speeds involved — requires some fast calculations. Thankfully the drone is able to oblige, by coming up with and testing a massive 40,000 different trajectories per second to settle on the right one.
As impressive a showcase as this is, however, Falanga points out that the research — published in a new paper titled “Aggressive Quadrotor Flight through Narrow Gaps with Onboard Sensing and Computing” — is about more than just clever tricks.
“An example where these abilities could be applied in practice is a search and rescue scenario to navigate through collapsed buildings and localize victims,” he explained. “In such scenarios, conventional means of access to buildings may no longer be available, and flying through a narrow gap might in some circumstances be the only way to enter and exit a building.
“One can also imagine a post-disaster area, where it is necessary to provide victims with first-aid kits, but it is not possible to use doors or windows to enter the building. In such a case, a quadrotor can be used to safely provide what is necessary to the victims, taking advantage of possibly available small gaps.”
From here, Falanga said that there are various directions in which he would like to take the research. For example, right now the drone is aware of the size of the rectangle it is flying through and therefore needs to only calculate the right trajectory. This is information it probably wouldn’t have were it to be used in the wild.
Due to the potential search and rescue applications, the team also wants to do work with getting the drone to maneuver while carrying a payload, as well as flying in more complex environments — such as cluttered ones in which it must avoid obstacles like poles or trees.
- The DroNet algorithm teaches drones to navigate city streets like cars
- Controller? What controller? You fly this drone with your facial expressions
- These drones could team up to lift injured soldiers off the battlefield
- Stanford’s laser tech could help self-driving cars see around blind corners
- A sensor-packed exosuit lets you fly a drone by pretending to be one