Of bats and robots

Abstract

Some bat species feature joint adaptations to their biosonar sensing and flight systems that allow them to navigate and hunt in dense vegetation. Understanding how biosonar sensing and flight are connected in these species poses a challenge: Kinematics recordings of bat-flight can document the system outputs, but due to various limitations, especially on processing, analysis of bat flight has been limited to minimal numbers of flights and hence has had difficulty to capture the natural variability in flight maneuvers. On the input side of the bats' flight-control system, it is necessary to understand the stimulus ensemble that guides flight in dense vegetation. Navigation in these cases must be based on clutter echoes, i.e., signals consisting of many unresolved components. Since the biosonar inputs into a bat are difficult to record without heavy interference with the animals' behaviors. Hence, biomimetic sonar robots can be used to collect stimuli from natural environments and their recreations. With their superior abilities to find patterns, deep-learning offer an opportunity to cut through the complexity of the input and output data of bat flight control and elucidate how clutter echoes are interfaced with the high-dimensional flight kinematics of bats to create the animals' exceptional maneuvering capabilities.