From:: tar@medg.lcs.mit.edu (Thomas A. Russ)
Date: Thu, 29 Nov 90 14:07:08 EST
To: gsl@ai.mit.edu
Subject: A Munching Learning Based...

			 AI REVOLTING SEMINAR
			 --------------------


	     A Munching Learning Based Dietary Branching
	      Algorithm for Automated Sandwich Assembly


			   Evel Darth Vader
		Department of Mastication Engineering



A primary source of difficulty  in automated sandwich  assembly is the
uncertainty in the relative position of the cold cuts being assembled.
This  is further  complicated  by the  presence of  numerous condiment
options goverened by both taste and allergic  reactions.   This thesis
uses a logic branching structure with  ``learned'' branching decisions
to  accomodate this  uncertainty.   Learning the  branching  decisions
eliminates   the  need for  a system  model   and therefore avoids the
associated modeling errors.  By making  the algorithm do all the work,
we can avoid having to understand the domain at  all, thereby creating
additional leisure  time to enjoy the   resulting repast.  Food sensor
information, responses to    recent line movement,  and   results from
previous assemblies are used  to generate the branching decisions (the
correct  direction in  which  to  move   in   order to    assemble the
consitutent parts).    Dietary  branching  using  machine  learning is
compared with other  techniques used for  automated assembly.  Several
heuristic  assembly algorithms  (such as  first-first, strongest-first
and  free-for-all) will be presented  that  were developed and  tested
both in a computer simulation and on a real buffet line.


Friday, November 30th, 12:03.4 PM NE43 8th Floor Playroom.
	        	  Good Food Noon


Hosts: Steve Glim and Alejandro Caro