Page 167 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Chapter 11

Flashback…
One of the problems of autonomous robot development is that our creations will quickly
evolve to work well in the development environment, but may be confounded by other
5
environments . The first robot we fielded in a real-world application (circa 1987) had
only a few navigation tricks. It could home in on a beam from its charger to become
oriented (referenced) and it could use wide-beam digital sonar to image wall surfaces and
correct its odometry. In our facility this, of course, worked wonderfully.

When, however, we turned the robot loose in the customer’s facility, it weaved down the
halls like a drunken sailor. The customer was not impressed. This happened on a Friday,
and our customer was scheduled to show the system to his vice president on Monday. It
would be a busy weekend.
We determined the cause of the problem rather quickly. Where the walls in our offices
had been flat, the walls in this new facility were interrupted frequently by recessed picture
windows. There were also large doors that were routinely left slightly ajar. The problem
occurred when the robot imaged these sections. The heading correction implied by these
features was often at a significant angle to the true heading of the hallway. Each time
such a correction was made the robot would change course, only to correct itself a few
meters further down the hall.

The first thing we did was to set the acceptance threshold on the heading correction
6
much tighter . If a fit wasn’t close to our current heading, we would simply reject it. This
eliminated the false angular corrections, but then other places around the building be-
gan to cause trouble.
There were several places where the robot was forced to drive for some distance without
any navigational corrections. When the robot crossed these gaps and began getting wall
readings again, it often rejected them as being at too great an angle.

Over the weekend, we applied the first crude fuzzy logic to our navigation process. The
results were remarkable, and we got through the demonstration that following Monday
with flying colors.

5 The mechanism at work here is clearly parallel to natural selection. This is another example of
how closely robotics parallels nature.
6 We could change this threshold on site because it was an easily accessible blackboard variable as
discussed in Chapter 6.

150