A COMPARISON OF SYSTEM IDENTIFICATION TECHNIQUES FOR REFUGE TRACKING BEHAVIOR IN EIGENMANNIA VIRESCENS

Abstract

Mathematical modeling has commonly been used to represent various animal behaviors. However, for complex system such as sensorimotor process, explaining its dynamics by simple mathematical formulations would become very challenging. Therefore, the data-driven techniques could be used for the identification of animal behavior. In this study, we focus on comparing different data-driven techniques for system identification of refuge tracking response in the weakly electric glass knifefish (Eigenmannia virescens).

In the refuge tracking task, Eigenmannia virescens track a polyvinyl chloride refuge actuated in one degree of freedom via a motor which is controlled by a PC. The PC can give both deterministic signals (such as sum of sines and chirp) and stochastic signals (such as noise) to the system. Our data collection system allows simultaneous recording of movements of the refuge and the fish via a real-time image processing software. Given the input and output data, we estimated frequency response functions (FRFs) of the refuge tracking behavior by using non-parametric system identification techniques. Then, we used these FRFs to estimate parameters of parametric transfer function models for the behavior using parametric system identification techniques.

We investigated how the selection of input signals affect the frequency response function estimations. We then compared different mathematical models for the input--output behavioral response of the fish by using sum-of-sines type stimulus. We conclude the thesis by discussing the next steps of our research.