Watch a Drone Swarm fly through a fake forest without crashing

Soria’s team tested a new perspective against a cutting-edge reactive model he had five drones and eight obstacles in a simulation, and they confirmed their passion. In one scenario, the reagents finished the mission in 34.1 seconds, while the forecast finished in 21.5 seconds.

Then came the real demonstration. Soria’s group gathered small Crazyflie quadcopters used by researchers. Each was tiny enough to fit in the palm of his hand and weighed less than a golf ball, but had an accelerometer, gyroscope, pressure sensor, radio transmitter, and small. motion capture balls, distributed a couple of inches apart and divided between four blades. The readings of the sensors and the camera to capture the movement of the room, which tracked the balls, reached the computer that launched the model of each drone as a ground control station. (Small drones cannot carry the hardware needed to perform predictive control calculations on board.)

Soria placed the drones on the ground in an “early” region near the first tree-like obstacles. When he launched the experiment, five drones were created and quickly moved to random positions in space above the 3D launch. Then the helicopters started moving. They glided through the air, through gentle green obstacles, on top of each other, under, and around, and landed with a gentle bounce toward the finish line. No collisions. The smooth swarming fluctuations allowed by the numerous mathematical computations that are updated in real time.

“The results of the NMPC [nonlinear model predictive control] the patterns are quite promising, “wrote Gábor Vásárhelyi, a robotist at Eötvös Loránd University in Budapest, Hungary, in an email to WIRED. (Vásárhelyi’s team created a reactive model used by Soria, but did not participate in the work.)

However, according to Vásárhelyi, the study does not address any crucial obstacles to the implementation of predictive control: the calculation requires a central computer. Subcontracting long-distance controls can leave the entire swarm under the influence of media delays or errors. Perhaps more decentralized control systems won’t find the best possible flight path, but “they can operate on very small ship devices (such as mosquitoes, beetles, or small drones) and climbing is much better with a swarm size,” he wrote. Drones’ artifices — and natural ones — can’t be high-powered computers.

“It’s a bit of a matter of quality or quantity,” Vásárhelyi continues. “However, nature has both.”

“I say, ‘Yes, I can,'” says Dan Bliss, a system engineer at Arizona State University, who is not involved with Soria’s team, which is leading a Darpa project to make mobile processing more efficient for drones and consumer technology. Small drones are also expected to be computationally more powerful over time. .Vessel tools that map the surrounding world, such as computer vision, require a large amount of processing power.

Lately, Soria’s team has been working to distribute intelligence among drones to accommodate larger swarms and manage dynamic obstacles. They are sets of drones around the forecast, like burrito-delivery drones, for many years. But that’s not it never. Robotics can see them in their future, and probably in their neighbor’s as well.

More great KABEKO stories