CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

Autonomous Drones Learn Challenging Acrobatics

Researchers at the University of Zurich  have developed a navigation algorithm that enables drones to learn acrobatic maneuvers using nothing more than onboard sensor measurements. Autonomous quadcopters can be trained using simulations to increase their speed, agility, and efficiency — capabilities that have applications in search and rescue operations. The researchers demonstrated maneuvers like a power loop, a barrel roll, and a matty flip. The key to the algorithm is an artificial neural network that combines input from the onboard camera and sensors and translates this information into control commands. The neural network is trained solely through simulated acrobatic maneuvers.

Topics:
Automation Imaging Motion Control Robotics Sensors Software

More From SAE Media Group

    Transcript

    00:00:00 deep drone acrobatics a chromatic flight with quad rotors is extremely challenging maneuvers such as the loop Mattie flip or barrel roll require high thrust and extreme angular accelerations that push the platform to its physical limits human drone pilots require many years of practice to safely master such agile maneuvers yet a tiny mistake could make the platform lose

    00:00:29 control and brutally crash we show for the first time that a vision based autonomous quadrotor that only has access to on-board sensing and computation is capable of flying acrobatic maneuvers with accelerations of up to three G in spite of the perception and control challenges raised by such accelerations our drones can fly all maneuvers in a robust and reliable

    00:00:56 fashion in order to achieve this we train an end-to-end sensory motor controller which directly regresses drone control commands from visual and inertial measurements our controller is trained exclusively in simulation simulation is fast cheap and safe it does not require any human annotation it enables collection of large amounts of data in a limited time but most

    00:01:27 importantly it allows for recovery from a crash by simply pressing a reset button training is done by imitating an optimal controller operating on privileged information and takes four to six hours on a normal desktop machine we ensure seamless transfer from simulation to the real caught rotor by employing appropriate abstractions of the sensory input using such abstractions the policy

    00:01:55 generalizes to different indoor and outdoor environments as well as to different physical drones compared to the traditional approaches based on estimation and control the learnt policy makes the drone fly more consistently and precisely while reducing the odds of a crash by up to 25% equipped with this learned controller our drones are more agile than ever

    00:02:20 before indeed they are able to fly maneuvers which are extremely challenging even for human experts