Fig. 1: Navigation-net can explain visual animal navigation.

The blue line shows the bat’s actual trajectory from its cave (black circle) to its target tree (black star). Red spheres show points where the drone collected 360° images (the drone took off at green points 1–7). Images with green frames show three examples of input-images with the actual (allocentric) direction in which they were taken, i.e., the gaze relative to the target (solid arrow) and the output error in degrees relative to the target. The error is depicted in the image. Gaze angles are given using a 0–360° clockwise notation with 0° pointing toward the fruit tree target. The figure also shows the locations where 360° images were taken a year later to test generalization (yellow points 1–9) with the region where the network performed well shown in green and the error in degrees given for these green sectors. Five example input images taken in the yellow points are shown on the right side (yellow frame) with the gaze and error angles depicted on them too. Note that in point 3 the navigation-net was not able to point toward the direction of the target with an error of less than 24° at all. The satellite image was obtained from Google Earth, 2018.