Evidence for latent regularities in echolocation-guided flight behaviour of bats

Echolocating bats navigate complex environments with remarkable agility, yet it remains unclear whether their flight trajectories are governed by reproducible internal policies shared within a species and across the controlled environments examined. To address this question, we recorded flight paths and pulse emissions of two bat species with contrasting sensing strategies (Rhinolophus nippon and Miniopterus fuliginosus) as they navigated seven obstacle-rich arenas in complete darkness. Using a variational recurrent neural network (VRNN), we suggest that bat flight is governed by consistent latent regularities. Trained on partial flight trajectories, the model captured key features of future paths across environments and individuals, preserving key features such as turning direction, obstacle avoidance, and velocity profiles. Despite substantial differences in sonar systems and flight morphology, the model captured species-specific strategies, supporting the existence of structured internal control in bat navigation. Our results show that bat flight trajectories contain structured, species-specific regularities that can be captured without imposing hand-crafted behavioral rules. Our data-driven framework provides a quantitative approach for testing latent flight policies and for predicting how echolocation-guided trajectories may reorganize under altered environmental conditions.