Meeting Abstract

101.9  Thursday, Jan. 7  Linear Dynamical Models for Refuge Tracking Behaviors of the Weakly Electric Knifefish Eigenmannia virescens ROTH, E.S.*; ZHUANG, K.; STAMPER, S.A.; FORTUNE, E.S.; COWAN, N.J.; Johns Hopkins University, Baltimore, MD; Johns Hopkins University, Baltimore, MD; Johns Hopkins University, Baltimore, MD; Johns Hopkins University, Baltimore, MD; Johns Hopkins University, Baltimore, MD eatai@jhu.edu

In numerous behaviors, animals modulate locomotor patterns to stabilize sensory images on receptor arrays. Many behaviors can be explained as a sensorimotor loop: sensing modulates motor action, the mechanical system generates forces on and filters external forcing from the environment, and these actions in turn generate reafferent sensory signals. In this work, we explore the applicability of linear dynamical models to “image stabilization” tasks, a broad class of sensorimotor behaviors in which animals modulate locomotor patterns to stabilize sensory images on receptor arrays. Despite the fact that many of the constituent sensory or mechanical subsystems are highly nonlinear, the closed-loop stabilization makes these behaviors particularly amenable to linear modeling. In this work, we observe the longitudinal tracking response of the weakly electric knifefish Eigenmannia virescens. These fish swim forward and backward using propulsion from an anal ribbon fin in response to motion of a computer-controlled moving refuge. Through an assay of perturbation experiments and control theoretic analyses, we identify that for a restricted yet still rich class of refuge trajectories, a linear dynamical model closely estimates actual tracking performance. We further observe regimes of stimuli which yield characteristically nonlinear responses and explore the implications of and hypotheses furnished by these results. Such data-constrained dynamical models provide parsimonious and quantitative task-level characterization of behavior. In conjunction with these task-level models and locomotor mechanics models (which might be derived from physical principles or fit empirically), system identification methods can be applied to infer the requisite neural processing for a given behavior.