@InProceeding{ Haas/1999,
author={Andrew R. Haas},
year={1999},
keywords={Simulated robot,natural language processing,restricted domain
knowledge},
institution={University of Albany},
title={Testing a Simulated Robot that Follows Directions},
journal={JAIR},
month={january},
URL= http://www.cs.albany.edu//~haas/taskbased
annote={
This paper describes a simulated robot that travel through a office
building,following direction given in natural english. He uses a grammar
inspired by HPSG . As the input sentence is parsed the semantic representation
of the sentence is constructed simultaneously. So there is only one phase
in the main routine of the program. The high level control procedures are
such that they can describe almost all sentences of this domain and still
then very easy to be implemented by a real robot using with primitive sensory
and moving capabilities. The author tested this program on natural examples
(example from people who were not present when it was being developed)
and claims that in 66% the robot reaches the right destination.
}
Author implemented a system which
mediated between an reactive mobile robot and domain restricted natural
language. He introduced ROB's (Reactive odometric Plan) for
plan recognition and fault tolerant plan execution and learning to
associate human terms with location in the environment.. A reactive
odometric plan or ROP is a representation of a short range plan to get
the robot between two places. These ROPs provide a way for the robot to
understand a place in terms of a way to get there in the world. The user
can name places ,ask question about the robot's plan and about the spatial
relationships of known places, give the robot long term goals. Author implemented
the system in a physical mobile robot system.