COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	URIP proposal
C00003 00003	Wish list
C00004 00004	Proposal outline
C00009 ENDMK
C⊗;
URIP proposal

range $50k to $500k cost to DoD

fiscal years 1984 and 1985; no announcement for 1985
  $60M; 1984 $30M and 1985 $30M
single proposal
doing research of high interest to DoD

cost sharing encouraged but not included in evaluation; 
other government funding does not constitute cost sharing;

specialized research computers will be considered
Wish list

video switch
video monitors
hardcopy video device; 35mm
video cameras
video disk
Matrox GXB-1000
GTCO tablets

IRIS
LISP machines:  ARPA will provide (when?)
Star ST-100
Array Processor
SUN configurations
  floating point SKY board
  Marinco
  disks for file servers

GVS-41 
McBain Zscope
electronics equipment for lab
mobile robot equipment
  communications
  sensors
  computation
Proposal outline

Section 2

2.a Abstract 300 words

2.b Budget
*** look at literature and call Bob Collins for quote
Star ST-100 array processor
  specify configuration for quote
    memory
    interfaces: VAX-780; VAX-750; VAX-750
cost sharing  $35,000

Section 3:  Supporting Information on Proposed Research
a.  Describe how proposed instrumentation will enhance research 
currently funded by DoD

This processor would contribute to our research
in the following areas:
1)  architectures for computer vision and intelligent systems;
2)  intelligent stereo mapping systems;
3)  inspection and vision for industrial assembly;
4)  mobile robots.

The Star St-100 is a high performance array processor with a quoted
performance of 100 Mflops.  Although such figures are usually
overestimates, the quoted figure is roughly 300 times the
speed of our current computers for operations which are most important
to our research.  The ST-100 has the advantage that it can be shared
by several host processors.

A major problem in computer vision and Artificial Intelligence is
that existing algorithms require great computation. 
There now exists
a body of algorithms which are useful in research and applications.
Although another major problem is to develop effective algorithms,
it is widely agreed that new algorithms will also require massive
computation.  

The proposed device would speed development of new algorithms by 
cutting the response time of systems by a large factor.  It would
enable the testing of algorithms on many more examples than now
possible, and allow many more cycles of algorithm refinement.

It would enable using algorithms as effective input for subsequent
higher levels of vision and intelligent systems.  Although this can
be done to some extent by using stored files with results from preceding
algorithms, this can only be done for systems which use a simple
straight-thru processing, not for systems which feed back results from
higher levels to control "focus of attention" of lower levels.  
For systems which use foveal attention to obtain high resolution results
in selected areas of an image, on-demand rather than canned results are
important.  Such systems are typical of computer vision.

It would enable using algorithms in effective demonstrations.  In past
research at Stanford, a mobile robot used stereo vision to navigate
over several realistic courses.  It advanced in 1 meter steps.
The robot required about 15 minutes per step, a total of about 6 hours
for a traverse.  A factor of 100 increase in speed would decrease that
time to about

It would allow simulation of 

*** material from ARPA proposal
*** material from ITA