Navigating the Unknown: Towards biologically-inspired Simultaneous Localization and Mapping
Date of Award
4-30-2024
Document Type
Thesis
Degree Name
Master of Science in Machine Learning
Department
Machine Learning
First Advisor
Dr. Hava Siegelmann
Second Advisor
Dr. Dezhen Song
Abstract
"In order to successfully perform non-trivial tasks that require spatial navigation, a robot needs to build a map of its environment. Paradoxically however, it needs to build this map while navigating the environment. This problem, known as Simultaneous Localization and Mapping (SLAM), is further complicated by the fact that sensor measurements contain errors which make accurate localization difficult. Animals also need to build maps of their environments to perform tasks required for survival and reproduction. In mammals, the Hippocampal Formation is thought to be the brain area responsible for this task. Various spatially modulated cell types have been discovered in the rodent Hippocampal Formation. While the behavior and function of these cell types have been investigated at depth, how they conspire with each other and with sensory stimuli to create a cognitive map that enables flexible navigation has not. This work introduces a SLAM framework inspired by the workings of the Hippocampal Formation. First, models of Head-Direction cells, Grid Modules and Place cells are integrated to predict the 2D pose from odometry input. Then, visual landmarks observed during odometry are used to calibrate position estimates. The allocentric coordinates of visual cues are encoded in landmark cells and subjected to hebbian plasticity. Grid cells in this model maintained their hexagonal firing fields despite sensor noise. This model also significantly outperformed other existing SLAM algorithms in a simulation environment."
Recommended Citation
K. Belay, "Navigating the Unknown: Towards biologically-inspired Simultaneous Localization and Mapping,", Apr 2024.
Comments
Thesis submitted to the Deanship of Graduate and Postdoctoral Studies
In partial fulfilment of the requirements for the M.Sc degree in Machine Learning
Advisors: Hava Siegelmann, Dezhen Song
with 1 year embargo period