The Spatial Semantic Hierarchy approach to robot exploration and mapping has been developed in the context of a simulated robot, NX, and tested on simulated environments with very simple models of sensorimotor errors [Kuipers and Levitt, 1988; Kuipers and Byun, 1988, 1991]. Physical implementations of aspects of the SSH approach have been built by other researchers but they do not provide adequate demonstration of its strengths or adequate analysis of its conditions of applicability.
The proposed research will extend the SSH Mapping theory from its original prototypical version to a version adequate for handling real sensorimotor interaction with a real (office) environment. The extended theory will be implemented on a physical robot to explore a previously unknown environment, and to create a SSH spatial description of the environment. Demonstration and evaluation of the SSH approach will be performed along several of its features.
First, we design and implement a set of reactive control laws which are robust towards sensor and effector inadequacies in real world and useful for defining distinctive places and paths.
Second, we investigate the ability of the SSH approach to build finite topological graph description of a continuous environment, adequate for path-finding, based on the robot's continuous interaction with its environment through its control laws.
Third, we will demonstrate and evaluate the ability of the SSH approach to build a detailed metrical map, and its use in disambiguating topological ambiguities, discriminating alternative routes of different length and identifying shortcuts.
Fourth, we will investigate the utility of the map for model-based control, providing an expectation-driven sensing strategy.
[Interested in my dissertation committee?]
WYL