Research and Applications - Artificial Intelligence
by Chaitin, L. J., Duda, R. O., Johanson, P. A., Raphael, B., Rosen, C. A. and Yates, R. A.
Technical Report Institution: Stanford Research Institute
April 1970.
Note: Project 8259 Interim Scientific Report.
From the Nilsson archives – SHAKEY papers
Abstract
The primary objectives of this program are to investigate and
develop techniques in artificial intelligence and apply them to the
control of a mobile automaton, enabling it to carry out tasks, autonomously,
in a realistic laboratory environment.
As part of technique developments, there are reports of progress in
scene analysis, and short-term and long-term problem solving.
Scene analysis is aimed at providing the au tomaton system with its
primary source of information about its environment. Such information
(primarily visual) must be entered into appropriate internal models in
a form useful for the problem-solving and planning systems. Several
parallel approaches—line analysis and region analysis methods—are
discussed, together with how knowledge of the actual environment is
used to help interpret the processed information.
The short-term problem-solving research is primarily aimed at providing
new tools and concepts for coping with tasks of increased complexity
to be implemented in the next year. A major software tool, QA3
(a question-answering system using theorem proving by resolution), will
be revised and upgraded, permitting experimentation with new strategies
and also allowing the theorem prover to be used in planning. Underway
are implementations of a new n-tuple model and set of operators–routines
that, when executed according to some plan, cause the robot to perform
specific actions. Both primitive and more complex operators are being
defined; these are callable in correct sequence by a planner. The planner
organization is the subject of considerable research, which includes the possible use of a GPS-like deductive mechanism, perhaps together
with the revised QA3 system. Finally, executive routines are being
developed to supervise execution of operator sequences and updating of
models, assessing cost-effectiveness of plans, making decisions such as
regarding the need for new sensory information or abandonment of further
planning in favor of execution, etc.
The long-term problem-solving research is devoted to design and
implementation of a general-purpose formal problem-solving system, QA4
based on mechanized theorem proving in higher-order logic. It is being
designed to emphasize the role of semantic information processing and
flexible control strategies. It will provide a rich language permitting
syntax and semantics of any part of the system to be expressed in its
own language, and will include set operations bridging logical and
computational operations. It is expected to provide tools and techniques
suitable for long-range research in such diverse fields as automatic
program writing and robot planning.
Appendices include:
AIC Techical Note 25: “PDP-15 Simulator” by John K. Ellis and Leonard J. Chaitin. April 1970
AIC Technical Note 16: “A LISP-FORTRAN-Macro Interface for the PDP-10 Computer” by John H. Munson. November 1969
AIC Technical Note 28: “Some Remarks on Resolution Strategies” by Robert E. Kling. April 1970
AIC Technical Note 27: “LISP Trace Package for PDP-10” by Robert E. Kling. April 1970
AIC Technical Note 22: “A LISP Implementation of BIP” by Richard E. Fikes. February 1970
AIC Technical Note 29: “A Cost-Effectiveness Bassis for Robot Problem Solving and Execution” by John H. Munson. January 1970
