Echo-based 3D reconstruction of natural environments

Echolocating bats are masters of acoustics, using their bio-sonar system to successfully navigate and hunt in highly dense environments. Our understanding of how they do so, is still limited.

Even after decades of research we do not have an answer to the basic question: can bats reconstruct a 3D ‘image’ of the world using echoes? Alternatively, bats could use the statistics of the echoes in order to navigate and hunt as is suggested by many researchers.

Previous attempts to study this question relied on behavioral or physiological approaches. Here, we plan to harness new learning techniques to address the question from a theoretical point of view, asking whether the echo-based 3D reconstruction problem can be solved.

We will use powerful acoustic simulations to generate the largest ever constructed data-base of naturalistic echoes. We will then train recently developed artificial neural-networks (originally developed for visual data) to reconstruct 3D structure based on acoustic echoes.