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

Area of Science:

• Behavioral Ecology
• Neuroscience
• Robotics

Background:

• Echolocating bats navigate complex environments with high agility.
• The underlying control mechanisms and consistency of bat flight trajectories are not fully understood.
• Species-specific differences in echolocation and flight morphology exist.

Purpose of the Study:

• To investigate whether bat flight trajectories are governed by reproducible internal policies.
• To identify species-specific navigation strategies in echolocating bats.
• To develop a data-driven framework for analyzing bat flight control.

Main Methods:

• Recorded flight paths and pulse emissions of two bat species (Rhinolophus nippon and Miniopterus fuliginosus) in seven obstacle-rich arenas.
• Utilized a variational recurrent neural network (VRNN) to model bat flight trajectories.
• Trained the VRNN on partial flight data to predict future paths.

Main Results:

• The VRNN successfully captured key features of future bat flight paths, including turning direction, obstacle avoidance, and velocity profiles.
• The model identified consistent, latent regularities in flight behavior across different environments and individuals.
• Species-specific navigation strategies were discernible despite differences in sensory systems and morphology.

Conclusions:

• Bat flight is governed by structured, species-specific internal policies that can be learned from trajectory data.
• A data-driven approach can reveal latent control mechanisms in animal navigation without imposing behavioral rules.
• This framework enables quantitative analysis of flight policies and prediction of trajectory changes in altered environments.