Soria’s team targets State-of-the-art reaction model In the simulation of five drones and eight obstacles, their hunches were confirmed. In one case, the reactive group completed their task in 34.1 seconds—the predictive group completed the task in 21.5 seconds.
Next is the real demonstration.Soria’s team gathers Crazyflie quadcopter Researchers use.Each one is small enough to fit in the palm of her hand and weighs less than a golf ball, but with an accelerometer, a gyroscope, a pressure sensor, a radio transmitter and a small motion capture The ball, spaced a few inches between the four blades. The readings from the sensors and the motion capture camera of the trackball in the room flow into a computer that runs each drone model as a ground control station. (Small drones cannot carry the hardware required to run predictive control calculations on board.)
Soria placed the drone on the floor near the “start” area of the first tree-shaped obstacle. When she started the experiment, five drones appeared and quickly moved to random locations in the 3D space above the takeoff area. Then the helicopter began to move. They glided through the air, between soft green obstacles, above, below and around each other, bounced towards the finish line and landed. There is no collision. Through a large number of mathematical calculations updated in real time, a stable bee colony becomes possible.
“Results of NMPC [nonlinear model predictive control] The model is very promising,” wrote Gábor Vásárhelyi, a robotics expert at Eötvös Loránd University in Budapest, Hungary in an email to WIRED. (Vásárhelyi’s team created the reactive model used by Soria, but he was not involved in the work.)
However, Vásárhelyi pointed out that the study did not address the key obstacle to the implementation of predictive control: a central computer is required for calculations. Long-distance outsourcing control may make the entire group vulnerable to communication delays or errors. He wrote that simpler decentralized control systems may not find the best flight trajectory, but “they can operate on very small airborne equipment (such as mosquitoes, ladybugs, or small drones), and increase with the size of the group. The expansion, the effect is better,” he wrote. Artificial and natural drone swarms cannot have bulky onboard computers.
“This is a question of quality or quantity,” Vásárhelyi continued. “However, nature has both.”
“This is where I say’yes, I can’,” said Dan Bliss, a systems engineer at Arizona State University. Bliss was not involved in Soria’s team. He led a Darpa project aimed at improving the efficiency of mobile processing of drones and consumer technology. Over time, even small drones are expected to become more computationally powerful. “I put a computer problem of a few hundred watts on a processor that consumes 1 watt,” he said. Bliss added that creating an autonomous drone swarm is not just a control problem, it is also a perception problem. Airborne tools that map the world around them, such as computer vision, require a lot of processing power.
Recently, Soria’s team has been working on distributing intelligence among drones to adapt to larger groups and deal with dynamic obstacles. The drone swarms with predictive awareness are, Like a burrito delivery drone,after many years.But this is not no wayRobotics can see them in their future—and, most likely, they can also be seen in their neighbors.
More exciting connection stories