New AI-driven drones learn like animals.

"Photo of the “neuromorphic drone” flying over a flower pattern. It illustrates the visual inputs the drone receives from the neuromorphic camera in the corners. Red indicates pixels getting darker, green indicates pixels getting brighter. Credit: Guido de Croon" (ScitechDaily, The Future of Flight: Researchers Develop Neuromorphic Drones That Learn Like Animals)

The animals and their training are inspiring new drones and other robots. The idea is similar to teaching dogs. When a dog makes something as it should, the instructor gives candy to it. That thing motivates the dog, and in the case of a drone or robot, the operator pushes enter and accepts the operation. 

In the traditional learning model, the operator remote-controls the robot when it makes something for the first time. Then those operations are stored in the machine's memory. But the more advanced systems can imitate other machines or humans when they learn something new. 

In those systems, the robot observes other things like similar robots using cameras. And then, it copies those actions into its mass memory. If we want to make man-shaped robots, that imitate humans in the learning process, the robot can see how human moves hands and legs during some actions. Like moving boxes.

Then the robot can copy those actions into itself. Then the robot can use those movement series when it takes order to move similar boxes. The robot sees the box and then connects that image to the databases that involve the movement series. The robot can learn things by watching screens and in that model, the robot can sit in an armchair and look at the TV to learn things. Like how to make food. 

When a computer learns something, it connects a new skill module into itself. If we want to make a drone helicopter that must fly through the labyrinth we can use the film where a remote-controlled drone flies through the route. Then that film will driven to other drones' memories. Things like GPS, inertial navigation, and laser rangefinders can make drones operate independently. 

However, using drone pairs is possible to create similar results using cheap drones. That must not be equipped with complicated sensors. Only cameras are enough to make impressive solutions for independently operating drones. 

In the simplest and the most effective model drone pairs can operate as a team. The drones observe each other and tell if the other drone goes too close to the wall. The drones must only know the speed of the remote-controlled drone.

Then drones must copy the turnings to themselves. If drones are identical, they can use information on how many rotations per minute the remote-controlled drones' engines take. Then the drone needs time what certain part of the labyrinth takes. But if the labyrinth is open, another drone can hover above the second drone and tell where that low-flying drone must go. 

https://scitechdaily.com/the-future-of-flight-researchers-develop-neuromorphic-drones-that-learn-like-animals/