Carl Kaiser will present his MS thesis entitled "Interaction Space Constructs and Modeling for Application in Robot Design" Tuesday, March 21, at 1 p.m. in Room 115 of the Natural Resources Building.

Abstract

Behavior-based robotics has gained acceptance within the robotics research community during the last fifteen years. The notion of bottom-up design associated with behavior-based control has been shown to be advantageous for relatively simple systems, but more complex systems require more comprehensive design tools. Design tools take functional requirements and drive to an implementation based on abstractions and models. Primitive elements of these abstractions form the basis for a world model and the choice of abstraction affects the nature of the predictions that can be derived from the model. Task, Environment, and Agent (TEA) is an abstraction used to describe robots accomplishing purposeful tasks in the context of an environment.

In this thesis, TEA is taken further to define the concept of interaction spaces. Interaction spaces are sets of interactions between task, environment, and agent, or any two of these elements. Four specific interaction spaces are defined along with subsets that can be used in robot design. System dynamics was originally developed to model complex industrial and economic situations, can be used for quantifying interactions between large numbers of interrelated elements. In this thesis, a system dynamics approach is applied in the form of primitive elements of an abstraction to model interaction spaces. This model begins to provide a form of quantification useful in understanding robot design criteria. A process is suggested for using interaction space models to create a functional design tool.

Three interaction space models are presented to illustrate the technique of interaction space modeling. These models include a model of a wall following robot, a model of multi-agent foraging task, and a model of victim detection for urban search and rescue. Each model is discussed conceptually with a step-by-step explanation of model creation.

Generalization of techniques used for model creation is pursued. Four aspects of the modeling process can be applied generally including classification and decision making, ways to model environments, identification of types of interactions that currently cannot be modeled, and modeling of finite state systems. In summary, a technique of interaction space modeling is developed and presented. Initial work is presented on a generalized approach to this technique and a process is presented to turn interaction space models into a functional design tool.

Future work on this process include validation of the technique with real robots, and further generalization of the interaction space modeling technique.

Committee

Dr. Wade O. Troxell, Advisor, Mechanical Engineering

Dr. David G. Alciatore, Mechanical Engineering

Dr. Louis B. Bjostad, Bioag. Sciences & Pest Management

Category: Faculty Focus
Submitted: Monday, March 20, 2006
Subject: Thesis defense
Contact: Marie Pearl Zimenoff
E-mail: Marie.Zimenoff@ColoState.EDU
Phone: (970) 491- 3872