A laser range sensor was used to "map" the
pallet at the locations in between the holes. The gains of the servo
controlling the robot joints was modified to make the robot "stiff" while
readings were taken. The optimal loading locations of each set of three
disks was then computed. The range sensor readings were fed directly to
the robot servo motion controller for real time,
on-the-fly sensor guided control. The software control layer has
been written to work on both Ulvac and Leybold sputter
generation. Modifying the code written for 65 mm disks to work for
95 mm disks was accomplished in 1 hour despite significant
differences in the sequencing and the geometry of the pallet and grippers.
This rapid turn around was achieved by developing an automated robot
motion planner which generated a "motion highway map". We thus
eliminated the arduous task of specifying specific motions. Instead we
supplied the motion manager with a map of the highways and a set of
algorithms to compute, depending on the current robot location, the best
route. If the cell is reconfigured, some highways become "closed" and the
system re-computes the new route. The algorithms was first simulated and
then converted to robot program code.
This approach was
adopted to run on both the Ulvac and Leybold Sputter machines in IBM
plants in the US and Germany.