On the cover page of Science Robotics: Drone dodges thrown objects and wins, thanks to event cameras
We are excited to share that the paper "Dynamic Obstacle Avoidance for Quadrotors with Event Cameras" of the Robotics and Perception Group has been accepted for Science Robotics (which has an Impact Factor of 20 according to Web of Science) and is featured on the cover page of this month' issue. For perception, their drone uses a stereo Insightness camera. All processing happens onboard with an overall latency of 3.5ms. The main research challenge was the segmentation of events generated by the ego motion of the drone and those generated by the thrown objects. For this, they used the Focus Maximization technique (CVPR18) by Guillermo Gallego and explained in the second half of the video below.
PDF (open access): https://robotics.sciencemag.org/content/5/40/eaaz9712
Video (watch our drone play dodgeball and win): https://youtu.be/BzykucxFddI
Press release by UZH, also in English: https://www.media.uzh.ch/en/Press-Releases/2020/Moving_robot.html