Joules for Bits: Energy-Information Trade-Off in Active Sensing Behaviors

Chen, Chen

doi:https://doi.org/10.21985/n2-2myx-n670

Work

Joules for Bits: Energy-Information Trade-Off in Active Sensing Behaviors

Public

Download PDF

Download All Files (.zip)

Movement and sensing fundamentally works in a synergistic manner. Animal's sensory organs --- be they independently movable like eyes or requiring whole body movement as in the case of electroreceptors --- are actively manipulated throughout stimulus-driven active sensing behaviors. Though these sensing-related motions have been individually reported and analyzed across a wide range of animal species from moth to mole, little is known about the their underlying mechanisms in the context of sensory acquisition. This thesis investigates the possible mechanism motivating these sensing-related sensor movements by connecting to behavior to information theoretic strategies of sensory acquisition as an active sampling process, and uses weakly electric fish as a model system to study the interplay between sampling strategy used for active sensing and locomotor control. The first part of this thesis investigates the sense organ control strategy underlying active sensing movement pattern identified from tracking behavior task reported across a wide range of animal species including a weakly electric fish, a cockroach, a hawkmoth, and a mole. Despite the vast difference in their body size, sensory modality, and physical environment, the sensing-related movements under sensory signal degradation among these included animal species share remarkable similarity in their pattern which is summarized as small amplitude wiggle movement. These sensing-related wiggle movements are shown to conform to a information-theoretic sampling strategy where animal gambles on the chance of acquiring more information through motion in a fully implemented algorithm framework termed ergodic information harvesting (EIH). The second part of this thesis investigates the swimming kinematics of a species of weakly electric fish engage in active electrolocation task under different levels sensory noise. Systematic analysis on the swimming kinematics reveals that fish swim in shorter bursts with higher amplitude of estimated thrust force output under weak sensory signal condition, confirming predictions from EIH. Fish also moves its nodal point under weak sensory signal in significantly higher velocity which contradicts with previous predictions from a first-order nodal point control model. Robotics experiment further confirms that the higher velocity nodal point shifting motion indeed lead to higher thrust force output which is believed to support the increased demand for maneuverability in order to execute the sensing-related wiggle motion predicted by EIH. Overall, this thesis investigates the energy-information trade-off in active sensing behavior through findings that shed light on: 1) the sensing-related movement observed across a wide range of animal species might be the result of animal proportionally bet on the chance of acquiring information through motion; 2) information maximization as a sampling strategy alone cannot explain sensing behavior under uncertainty; 3) in weakly electric fish, nodal point control in non-steady state swimming can no longer be explained by a first-order model and a higher-order model is needed; 4) ribbon fin control in weakly electric fish might be evolved partially for addressing the increased demand in maneuverability for sensing-related wiggle movement seen under weak sensory signal as predicted by EIH. More broadly, these work may be useful both for understanding the energy-information trade-off in active sensing behavior and as a control algorithm framework for applications in autonomous systems acting with uncertainty.

Creator