UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Spatial function in animals and robots

Harris, Kenneth D.; (1999) Spatial function in animals and robots. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of Spatial_function_in_animals_an.pdf] Text
Spatial_function_in_animals_an.pdf

Download (6MB)

Abstract

This thesis describes work aimed at discovering computational processes that could underlie the performance of spatial tasks, both in animals and artificial systems. In the mammalian brain, the hippocampus is believed to play an important role in spatial function but is also often said to be a storage place for general memories. The first section of the thesis describes a model of mammalian spatial function in which the neocortex processes sensory information to build an egocentric map of the environment, and the hippocampus acts as an autoassociative memory which stores and recalls snapshots of neocortical activity, providing the animal with the ability to recall a complete egocentric map from a few observable landmarks. The model was implemented on a mobile robot equipped with a sonar sensor. The addition of a modelled hippocampus to previously developed map-making software allowed the robot to perform a task similar to the Morris water maze. The remainder of the thesis describes two novel methods for map construction from sonar sensory data. The first is a neural-network mapping system inspired by models of visual cortex. In this system, the activity of a cell represents the occupancy of a region of space, and lateral connections between cells enforce prior knowledge about probable map configurations. This method produces good results in comparison to other grid-based mapping systems, but is based on heuristics, rather than mathematical principles, and contains many free parameters. The second is a feature-based mapping system which is derived from Bayes' theorem. It requires an accurate probabilistic model of the robot's sensor, which is derived empirically. The sensor model is used to construct a probability function on the space of all possible maps, and the problem of map construction becomes a search for the most probable map. Global optimisation problems of this size are very difficult, but a "mean-field" approximation allows a tractable solution. Tested with identical sonar data, the new method requires much less sensory data to produce a high-quality map than previous heuristic methods. Although derived without reference to biology, the method has similarities with certain previously proposed models of cortical function, which are discussed.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: Spatial function in animals and robots
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Thesis digitised by ProQuest.
Keywords: Applied sciences; Task performance
URI: https://discovery.ucl.ac.uk/id/eprint/10107629
Downloads since deposit
34Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item