The human fascination to understand ultra-efficient flying beings like bees, birds and flies have led to the study of their sensors and neural architectures. This is in unison with the recent advancements of nano-quadrotors of maximum size and weight of 130 mm and 200 g respectively with on-board computation and sensing due to the inherent advantages of safety, agility and power efficiency. Researchers have started moving away from treating perception as a black box and treating perception and control as a coupled entity - active vision. However, foundations of active vision directed toward nano-quadrotors have to be formulated. This is coupled with the missing hardware and software co-design evaluation metrics.
We propose to introduce a framework for minimalist hardware and software co-design evaluation centered around active vision and use it to develop autonomous nano-quadrotors with onboard sensing and computing. We also propose to lay out foundations for the choice of problems to be solved using deep learning and how to run them on the limited computation available on nano-quadrotors.
Dept. rep: Dr. Dinesh Maocha
Members: Dr. Davide Scaramuzza
Dr. Inderjit Chopra
Dr. Cornelia Fermuller