Biologically Inspired KFLANN Place Fields for Robot Localization

Abstract

This paper presents a hippocampal inspired robot localization model that provides a means for a simple robotic platform with ultrasonic sensors to localize itself. There have been published neurobiological experiments where rats were found to have hippocampal cell activations that positively correlate with the location of the animal [2,3,5]. Such activations found in the hippocampal region are usually called Place fields (PF) or Place cells (PC). The Place Field Model presented in this paper was designed using a unique K-Means Fast Learning Artificial Neural Network (KFLANN) [13, 14, 15] and establishes a series of localization minima points that act as references for navigation. While such evidence of place cells are seen in hippocampal (CA1) and deep layers of the entorhinal cortex (EC) [4], a literature search indicates it is uncertain whether any applications were ever designed using such biological evidence. This paper aims to focus on experimental results relevant for a proof-of-concept of robot localization, rather than illustrating a robustly tested navigation system. As such, basic ultrasonic based experiments will suffice. We use some experimental results, to show that KFLANN is suitable for implementing atomic place field vectors (APFV), a data structure to encapsulate localization information.