Robot Vision: from Frame-based to Event-based Cameras

Davide Scaramuzza

ETH Zürich, CH

Abstract

I will first talk about state estimation (position and velocity) using visual and inertial sensors, with an emphasis on the emerging field of event-based cameras (ultra-fast vision sensors with super-human performance). Then, I will talk about tight coupling of perception and action for autonomous robot navigation in complex environments, using both model-based and model-free approaches. Finally, I will show real world applications to the emerging field of vision-based drones in GPS-denied environments, ranging from search and rescue to autonomous racing at high speeds.