A random stack of documents is physically sorted into ordered stacks accord
ing to type and condition. This is done by first locating the topmost item
using a machine vision system with algorithms based upon the Hough Transfor
m. The edges of the items are identified from the shadows created by illumi
nating the pile from different directions. Having located the position and
orientation of the topmost item, a robot arm is programmed to pick the item
from the pile using an end-effector with vacuum picking mechanisms. The ro
bot arm offers the picked item "on-the-fly" to devices which ascertain its
physical fitness and classification. Finally, the item is placed on the app
ropriate one of a number of piles of similar category. (C) 2000 Elsevier Sc
ience Ltd. All rights reserved.