Investıgatıng The Effects Of Actıve Sensıng Movements On State Estımatıon Performance
Özet
Active sensing movements, characterized by physical movements of sensory organs such as the eyes, ears, fingers, or antennae, is a method employed by animals to gather information about their surroundings. These movements are critical for animals as they enhance the efficiency and accuracy of their sensory feedback systems. This thesis provides a comprehensive framework for investigating active sensing movements in fish. Specifically, the study outlines the fish model essential for this examination, explores the shaping of active sensing movements or signals, and compares two active sensing mechanisms developed through control theory techniques. According to our hypothesis, fish engage in active sensing behaviors in order to improve their ability to sense their environment more accurately. In other words, they attempt to increase their state estimation performance. However, it is not yet clear in the literature how these movements are performed and what information is used to generate them. In our study, we conducted simulations using various forms of active sensing signals on a previously established second-order fish plant model that had been developed through refuge tracking experiments. Initially, we implemented open-loop signals that were specifically designed for active sensing. This was based on the observation that active sensing movements tend to exhibit velocities with a Gaussian distribution. Then, to investigate the theory that active sensing signals are generated under feedback control, which has been studied in the literature, we implemented it on our model. This was done in the form of a closed-loop signal using the state covariance matrix, which contains the state estimation information. In both of our experiments, results were obtained that supported our hypothesis, and the simulation results confirmed that these active sensing behaviors of fish enhance their state estimation performance. Furthermore, we demonstrate through comparison of results obtained from simulations of the open-loop and closed-loop active sensing signals that closed loop performance is superior at low refuge frequencies.
