Modelling Spatio-Temporal Consistency as a Mechanism for Echo-Segregation in Bat Swarms

Abstract

Active-sensing animals, such as bats, probe their environment by emitting energy and analysing returning signals. Even a solitary bat’s call generates a cascade of echoes from multiple objects, which the auditory system must segregate into one spatial scene. The problem of spatial scene reconstruction is exaggerated in social groups, where an individual must also discriminate its own echoes from the calls and echoes of conspecifics to prevent constant collisions. How, then, do bats navigate dense collectives without colliding even when facing intense jamming? We developed a biologically-parameterised dynamic model of active-sensing agents. We create an agent-based model and explicitly model bat behaviour and sound propagation. Inspired by the idea of optic and echo-acoustic flow, we propose that bats may detect their own echoes in large swarms using spatio-temporal consistency. The generated by obstacles around a bat are temporally locked to the bat’s call emission. Hence, sounds that are consistent in terms of direction of arrival and post-call delay are more likely to be self-generated echoes. We demonstrate a dramatic reduction in group collision rates when individuals integrate information over larger time windows. Collision rates decrease by approximately 29% when bats utilise consistency as compared with agents not using echolocation. Also, agents utilising spatio-temporal consistency avoid jamming cost across an order of magnitude of group sizes (from 5-100 individuals) and show 30% lower wall collisions on average than the negative sensorimotor controllers simulated (non-echolocating agents). The behaviour of agents utilising consistency-based controllers and agents with artificially perfect self-generated echo detection is remarkably similar. The behavioural similarity supports the idea that tracking consistent sounds in both directions of arrival and time delay improves echolocation under masking conditions. The modelling approach lays the groundwork for understanding how multiple echolocators can exhibit collective behaviour.