Echolocating bats accumulate information from acoustic snapshots to predict auditory object motion

Abstract

Unlike other predators that use vision as their primary sensory system, bats compute the three-dimensional (3D) position of flying insects from discrete echo snapshots, which raises questions about the strategies they employ to track and intercept erratically moving prey from interrupted sensory information. Here, we devised an ethologically inspired behavioral paradigm to directly test the hypothesis that echolocating bats build internal prediction models from dynamic acoustic stimuli to anticipate the future location of moving auditory targets. We quantified the direction of the bat's head/sonar beam aim and echolocation call rate as it tracked a target that moved across its sonar field and applied mathematical models to differentiate between nonpredictive and predictive tracking behaviors. We discovered that big brown bats accumulate information across echo sequences to anticipate an auditory target's future position. Further, when a moving target is hidden from view by an occluder during a portion of its trajectory, the bat continues to track its position using an internal model of the target's motion path. Our findings also reveal that the bat increases sonar call rate when its prediction of target trajectory is violated by a sudden change in target velocity. This shows that the bat rapidly adapts its sonar behavior to update internal models of auditory target trajectories, which would enable tracking of evasive prey. Collectively, these results demonstrate that the echolocating big brown bat integrates acoustic snapshots over time to build prediction models of a moving auditory target's trajectory and enable prey capture under conditions of uncertainty.

Keywords: Eptesicus fuscus; auditory localization; biosonar; predictive models; prey tracking.

Fig. 1.

Tracking of simple target motion. Time 0 indicates time at which the target is crossing in front of the bat. (A) Head aim tracking of the target, shown as angle over time for the standard simple motion and catch trials. Target angle relative to the bat’s position is shown on the black dashed line. (B) Tracking angle, defined by the difference between the target angle and the head angle for standard simple motion trials, shows that the bat’s head aim is always ahead of the target. (C) Sonar call rate for standard simple motion and catch trials during the target tracking period of the trial. (D) Head angle offset of data, defined by the difference between the head angle and the modeled head angle for standard simple motion trials (nonpredictive model and predictive model).

Fig. 2.

Models for target tracking. (A) Head angle offset of data as defined by the difference between the head angle for standard simple motion trials and the modeled head angle for the nonpredictive model, fixed head angle model, velocity estimation model, and predictive model; these models integrate sonar call data to estimate head aim. (B) Goodness of fit for the fixed head angle model for different angular shifts; red dot indicates best fit. (C) Goodness of fit for the velocity estimation model for different numbers of echoes for the velocity estimation; red dot indicates best fit. (D) Goodness of fit for the predictive model for different numbers of echoes for the velocity estimation and different angle shifts; the area outlined with the red dashed line is shown in E. (E) Goodness of fit for the predictive model; red dashed line indicates best fit.

Fig. 3.

Internal model of target trajectory enables head aim tracking during occlusion. Time 0 indicates time at which the target is crossing in front of the bat. (A) Tracking angles for standard simple motion in unoccluded trials (green line) and for occluded trials (yellow line) show that the head aim is always ahead of target (dashed line). Time at which target trajectory is behind occluder in occluded trials is marked as a gray box. (B) Head angle offset of data from occluded trials to nonpredictive model and to predictive model. (C) Sonar call rate for standard simple motion for unoccluded trials (green line) and for occluded trials (yellow line) during the target tracking period of the trial. Time at which target trajectory is behind occluder in occluded trials is marked as a gray box. Solid blue line denotes the window where significance between occluded and unoccluded trials was tested.

Fig. 4.

Target tracking in velocity changes and occlusion conditions. Time 0 indicates time at which the target is crossing directly in front of the bat. Red arrows indicate change in target velocity, which separates the two evaluation windows (before and after change in velocity). Time at which target trajectory is behind occluder in occluded trials is marked as gray boxes. (A) Tracking angle in velocity change fast motion for unoccluded trials (unoccluded; green line) and for occluded trials (occluded; yellow line). (B) Head angle offset of data from velocity change fast unoccluded trials to nonpredictive model and to “predictive model.” (C) Head angle offset of data from velocity change fast occluded trials to nonpredictive model and to predictive model. (D) Sonar call rate for velocity change fast for unoccluded trials (green line) and for occluded trials (yellow line) during the target tracking period of the trial for velocity change fast trials. (E) Tracking angle in velocity change slow motion for unoccluded trials (unoccluded; green line) and for occluded trials (occluded; yellow line). (F) Head angle offset of data from velocity change slow unoccluded trials to “nonpredictive model” and to predictive model. (G) Head angle offset of data from velocity change slow occluded trials to nonpredictive model and to predictive model. (H) Sonar call rate for velocity change slow for unoccluded trials (green line) and for occluded trials (yellow line) during the target tracking period of the trial.

Fig. 5.

Experimental setup. The bat sits on a platform and tracks a target (mealworm) that is moved on a tether by a motor from left to right. A microphone was placed opposite to the bat to record the bat’s sonar calls. In some trials, the target trajectory is partially occluded by the “occluder.” Tracking angle is defined by the difference between the target angle (orange) and the head angle (red). Head angle offset to the predictive and nonpredictive models is defined by the difference between the head angle and the modeled head angle.