|
This report describes the results of research during the first
eight months on the project “Application of Intelligent Automata to
Reconnaissance.” The primary goal of this project is to investigate
techniques in artificial intelligence applied to the control of a mobile
automaton in a realistic environment. The main emphasis is in the design
of a hierarchy of computer programs that will accept visual and
other sensory information from the automaton and direct its actions
toward the completion of missions requiring the abilities to plan ahead
and to learn from previous experience.
The automaton itself is a four-wheeled, self-powered vehicle with
two simple arms for grasping objects. It will carry a television camera
an optical range-measuring device, tactile sensors, a transceiver, and
storage and routing logic. The vehicle will be radio-controlled by an
SDS 940 time-shared computer. The computer, with the aid of special
image-processing hardware, will analyze television pictures transmitted
from the vehicle and will calculate a sequence of vehicle actions designed
to accomplish the task(s) given to the automaton. Calculation of the
appropriate action sequence will be performed by problem-solving computer
programs which will use internally-stored “models of the world” generated
and abstracted during the course of automaton explorations.
The environment of the automaton, initially, will be a 20X30 foot
room populated with simple objects. Sample tasks will include transporting
these objects and arranging them in specified ways, under stated constraints. Later it is planned that the automaton attempt more complex
tasks in the larger, richer environment of corridors and offices
outside of its computer room.
At the present time most of the special hardware has been designed
and is scheduled for completion by 1 April 1967. With the SDS 940 computer
now delivered , programming the actual vehicle will begin in four months. In the meantime, simulation experiments concerned with representations
or models of the environment, control strategies, and pattern
recognition are being conducted.
Looking beyond the duration of the present project, toward the goal
of building increasingly sophisticated automata, it will be important
to initiate several independent but contributory research efforts as soon as possible. The most important of these concerns the problems of threedimensional
visual perception and pattern recognition.
Appendicies include:
- “Listing of the ALGOL program SNOOP7” by M. W. Green
- “Some Variations on Discrete Model Strategies” by D.R. Bennion
- “Experiments in Simulating an Automaton Navigating in a Discrete Environment” by J.H. Wensley
- “Heuristics for a Surveying Automaton” by H.D. Crane
- “An Efficient Algorithm for Route Finding” by N.J. Nilsson
- “Route Finding in a Continuous Environment” by W.G. Keckler and R.E. Larson
- “Compilation of Basic Vehicle Operations” by E.B. Shapiro
- “Tests of Vidicon TV Cameras” by G.E. Forsen
- “Components of the Video Preprocessor” by G.E. Forsen
| 
