The basic agent is intended for domains where supervisors need mechanisms for natural
language input and output for the system. Since sensing and manipulation are reduced to symbolic acts
in a simulated world, it is unclear how successful the system would be in a
real world situation. In addition, there is little discussion of what types
of learning are possible, which seems important since the requirement for
complete domain knowledge seriously
limits the number of applicable domains.
However, it can function even in environments presenting information in
English and requiring direct natural language output. The language
representations used by the basic agent can easily be extended beyond the
Seaworld domain of Homer, though the authors themselves express
reservations about expressing spatial or visual concepts like "edge" or
"corner". These are limitations of the text interpreter and generator; the planner seems sufficiently general to
handle the extensions.
To return, press HOME.
To go to the next document, press NEXT.