Why, you might ask beyond artistic maneuvers? Seems more an approach to training than anything else. And as they say, addressing the limits of autonomy for tasks and risks.
Algorithm Enables Quadcopters to Perform Acrobatic Maneuvers Autonomously
Daily Mail (U.K.)
Jonathan Chadwick
Researchers at Switzerland's University of Zurich and Intel have created a navigation algorithm that enables autonomous flying drones to perform challenging acrobatic maneuvers autonomously, based on onboard sensor measurements. The heart of the algorithm is an artificial neural network that integrates input from the quadcopter's onboard camera and sensors, and directly translates this data into control commands. The network is trained via simulated acrobatic maneuvers, which makes demonstrations by human pilots unnecessary, and enables training to scale to a large number of maneuvers without posing physical risks to the drone. The algorithm abstracts the sensory input from the simulations and transfers it to the physical world, and drones require just a few hours of training. Zurich's Davide Scaramuzza said, "This navigation is another step towards integrating autonomous drones in our daily lives."
No comments:
Post a Comment