![](/wp-content/uploads/2021/03/abstractheadergraphic2015.gif)
Meeting Abstract
Many animal behaviors involve motion of the body or appendages for sensing the environment to gain information, which we term active sensing. As an example, weakly electric fish will sweep back and forth near a novel object to gain information. Using information theoretic approaches should help explain such movements in active sensing tasks that seem counterproductive. For instance, bats aim their sonar pulses off axis from a target, which decreases signal intensity but maximizes a common measure of information. However, in simulations of a mobile sensor tasked with tracking a target moving along a line, sensor motions that visit portions of the line proportional to the amount of expected information vastly outperform motions that constantly maximize information. The resulting simulated trajectories feature oscillations around the target in order to maintain a low variance estimate of object location. Moreover, as the signal to noise ratio (SNR) increases, these oscillations diminish and the sensor trajectories more closely match target motion. Electric fish (E. virescens) will follow the movements of a swaying shelter to remain hidden but also exhibit extraneous fore-aft and bending motions similar to when they are sensing a new object. This extra movement is more prevalent when visual cues are unavailable and SNR is low. Moreover, by further decreasing the SNR through conductivity increase or by applying a jamming signal (both shown to degrade electrosense), the fish increases the amplitude of the extra movements. These trends are consistent with the simulations of the target tracking task in various levels of noise. Using this information-theoretic approach might elucidate reasoning behind other non-intuitive animal behaviors and guide robotic algorithms for information gathering tasks to be more effective.