Active head rolls enhance sonar-based auditory localization performance

Abstract

Animals utilize a variety of active sensing mechanisms to perceive the world around them. Echolocating bats are an excellent model for the study of active auditory localization. The big brown bat (Eptesicus fuscus), for instance, employs active head roll movements during sonar prey tracking. The function of head rolls in sound source localization is not well understood. Here, we propose an echolocation model with multi-axis head rotation to investigate the effect of active head roll movements on sound localization performance. The model autonomously learns to align the bat's head direction towards the target. We show that a model with active head roll movements better localizes targets than a model without head rolls. Furthermore, we demonstrate that active head rolls also reduce the time required for localization in elevation. Finally, our model offers key insights to sound localization cues used by echolocating bats employing active head movements during echolocation.

Fig 1. Statistics of the head roll angle.

(A) Histogram of the roll angle during behavior (blue) and the marginal distribution of the fitted AR(5) model (red). (B) Sample auto-correlation function (blue) and the fitted AR(5) auto-correlation function (red).

Fig 2. System architecture.

(A) The system architecture of our echolocation model. The reflected echo is received at the left (red) and right (blue) ears. The model neural representations are obtained using six GASSOM units (GASSOM₁₋₆), which encode the received auditory signals at each combination of two tune scales (coarse and fine), and three aural combinations (binaural, monaural-left and monaural-right). The coarse scale windows twice as long as fine scale windows, but are downsampled by a factor of two prior to encoding. The action network maps the GASSOM responses to azimuth and elevation head control commands. An additional random head roll movement is introduced so that the bat controls the head orientation around all three axes. (B) The input and output of the fine scale binaural GASSOM. The inputs to the GASSOM are temporal snapshots from sliding windows. The GASSOM₁ contains a dictionary of basis vectors. The response from a few selected basis functions is shown as a function of the location of the sliding window.

Fig 3. Example trajectories.

Each figure shows examples of target direction in head-centered coordinates as the head direction is controlled by our model. Different colored trajectories illustrate eight initial target directions. Perfect localization corresponds to bringing the target direction to (0,0). Each circle marker indicates the target direction at iteration t ∈ {1,…,20}. The target direction converges to a point attractor (solid square). The inset shows a magnified view of the center region. The point attractor moves closer to (0,0) direction as the standard deviation of head roll σ increases.

Fig 4. Summary statistics of generated trajectories.

(A) The normalized time constant of the trajectories in azimuth, elevation direction. (B) The normalized mean squared error of the steady state head direction.

Fig 5. Basis vectors in the perceptual representation.

(A) Binaural and monaural basis vectors. The binaural basis vectors comprise left and right (red and blue) components. The monaural basis vectors corresponding to the left (red) monaural GASSOM are shown here. (B) The distribution of interaural level, time differences of the binaural basis vectors. The mean percentage of the basis vectors are shown.

Fig 6. The reconstruction error surfaces.

The reconstruction error as a function of horizontal and elevation coordinates. The symbols “+” and “o” indicate the (0,0) coordinates and the minima of the error surface respectively. (A) The reconstruction error for the GASSOM (σ = 10) at the head roll angles of -20, 0, and 20 deg. (B) The reconstruction error surface at different head roll standard deviations.