Integration of Soft‐Robotics and Deep Learning to Assess the Coordinated Biosonar Emission and Reception Dynamics in Hipposiderid Bats

Echolocating bat species with sophisticated biosonar systems such as the Old World leaf‐nosed bats (Hipposideridae) and horseshoe bats (Rhinolophidae) actively change the shapes of their ultrasound emission (“noseleaves”) and reception (“pinnae”) baffles during the diffraction of the outgoing and incoming acoustic waves. Our prior work with behaving hipposiderid bats has shown that these dynamic shape changes are tightly synchronized across emission and reception, but a functional significance for this coordination has yet to be demonstrated. As an intermediate step toward this goal, the hypothesis that different synchronizations between the noseleaf and pinna deformations give rise to consistently, i.e., repeatable different acoustic effects as opposed to random fluctuations, has been tested with a soft‐robotic reproduction of the peripheral dynamics in the biosonar system of hipposiderid bats. The functional elements of this robotic bat head were flexible baffles that mimic the animals’ noseleaf and pinna shapes and were deformed with pneumatic actuators. Using this hardware platform, we have recreated eight different synchronization patterns of noseleaf and pinna deformations that include patterns that have been seen in bats along with additional patterns that have not been observed in the animals. A convolutional deep neural network was able to classify the different noseleaf‐pinna deformation combinations with an accuracy of 96%. A transparent artificial intelligence approach (class‐activation mapping) indicated that the different deformation combinations give rise to unique patterns in the time–frequency plane.