Page 209 - Introduction to AI Robotics

192
5 Designing a Reactive Implementation
The teams with robots properly aligned with gravity went back to programming and
fine tuning their entries.
Cheering at robot competitions.
The chant “Core Dump. Core Dump. Segmentation Fault!” has a nice cadence and is
especially appropriate to yell at competitors using Unix systems.

Subsumption and Soda Cans.
Jon Connell addressed this task in 1989 in his thesis work at MIT, 39 applying sub-
sumption to a robot arm, not a set of flappers. He used a special type of FSM called
an Augmented Finite State Machines (AFSMs), and over 40 behaviors to accomplish
the task.
Being in a Pick Up the Trash competition without a manipulator.
As often happens, robot competitions often pose problems that are a step beyond
the capability of current hardware and software technology. In 1995, arms on mobile
robots were rare; indeed Nomad introduced a forklift arm just in time for the compe-
tition. Participants, such as the Colorado School of Mines with an older robot and no
arm, could have a “virtual manipulator” with a point deduction. The robot would
move to within an agreed tolerance of the object, then either play a sound file or make
a noise. The virtual manipulator—a human team member, either Tyler Devore, Dale
Hawkins, or Jake Sprouse—would then physically either pick up the trash and place
it on the robot or remove the trash. It made for an odd reversal of roles: the robot
appeared to be the master, and the student, the servant!

About grippers maintaining the state of world.
The Pick Up the Trash event mutated in 1996 to picking up tennis balls in an empty
arena, and in 1997 into a variation on sampling Mars. For the 1997 Find Life on Mars
event, the sponsors brought in real rocks, painted black to contrast with the gray
concrete floor and blue, green, red, and yellow “martians” or toy blocks. Because of
weight considerations in shipping the rocks, the rocks were about the size of a couple
of textbooks and not that much bigger than the martians. One team’s purely reactive
robot had trouble distinguishing colors during preliminary rounds. It would mis-
identify a rock as a martian during a random search, navigate to it, grip it, and then
attempt to lift it. Since the rock was heavy, the gripper could not reach the full exten-
sion and trigger the next behavior. It would stay there, clutching the rock. Sometimes,
the robot would grip a rock and the gripper would slip. The robot would then try to
grip the rock two more times. Each time it would slip. The robot would give up,
then resume a random search. Unfortunately, the search seemed to invariably direct
the robot back to the same rock, where the cycle would repeat itself. Eventually the
judges would go over and move the robot to another location.