From nobody@cs.Buffalo.EDU Tue Apr 14 17:37 EDT 1998
From: nobody@cs.Buffalo.EDU
Date: Tue, 14 Apr 1998 17:37:40 -0400 (EDT)
To: techreps@cs.Buffalo.EDU
Subject: techrep: POST request
Content-Type: text
Content-Length: 3138

ContactPerson: hexmoor@cs.und.edu
Remote host: dhcp216-57.cs.und.edu
Remote ident: unknown
### Begin Citation ### Do not delete this line ###
%R 98-04
%U /projects3/hexmoor/Diss/diss.ps
%A Hexmoor, H
%T Representing and Learning Routine Activities
%D December 1995
%I Department of Computer Science, SUNY Buffalo
%K Knowledeg representation, robotics, intelligent agents
%X A routine is a habitually repeated performance of some actions. Agents 
use routines to guide their everyday activities and to enrich their 
abstract concepts about acts.   
This dissertation addresses the question of how an agent who is engaged in 
ordinary, routine activities changes its behavior over time, how the 
agent's internal representations about the world is affected by its 
interactions, and what is a good agent architecture for learning routine 
interactions with the world. In it, I develop a theory that proposes 
several key processes: (1) automaticity, (2) habituation and skill 
refinement, (3) abstraction-by-chunking, and (4) discovery of new knowledge 
chunks.   
The process of automaticity caches the agent's knowledge about actions into 
a flat stimulus-response data structure that eliminates knowledge of action 
consequences. The stimulus-response data structure produces a response to 
environmental stimuli in constant time.    
The process of habituation and skill refinement uses environmental clues as 
rewards. Rewards are used to develop a bias for the agent's actions in 
competing action-situation pairs where the agent did not have prior 
choices.   
The process of abstraction-by-chunking monitors the agent's increased 
reliance on stimulus-response data structures and turns the agent's complex 
actions into primitive acts, thereby eliminating the need for the agent to 
elaborate its plans beyond a certain level.   
The process of discovering knowledge-chunks monitors the agent's use of 
stimulus-action data structures and constructs knowledge about action 
preconditions and consequences and constructs patterns of interactions into 
plans.    
I have implemented several agents that demonstrate parts of my theory using 
an agent architecture I developed called GLAIR. GLAIR models agents that 
function in the world. Beyond my theory about routines, GLAIR is used to 
demonstrate situated actions as well as deliberative actions. Each of 
GLAIR's three levels (Sensori-actuator, Perceptuo-motor, and Knowledge) 
uses different representations and models different components of 
intelligent agency.  GLAIR levels operate semi-autonomously. Whereas 
intra-level learning improves an agent's performance, inter-level learning 
migrates know-how from one level to another. Using the GLAIR architecture, 
I model agents that engage in routines, use these routines to guide their 
behavior, and learn more concepts having to do with acts.   
The significance of my theory, architectures, and the agents are to 
illustrate that computational agents can autonomously learn know-how from 
their own routine interactions in the world as well as learn to improve 
their performance.