"In order to win at a RoboMagellan contest (or even successfully finish one, for
that matter), one must solve a number of fairly difficult problems and
integrate all the solutions into one small package. It must be able to navigate
based on map coordinates. It must be able to find its way around obstacles,
which means it has to have ways to sense what's around it. It must be able to
pick out a small object based on shape and color. And it has to find its way
back again. In summary the mission is: 1) Navigate to the general location of a
waypoint, 2) Once there perform a task such as finding a cone, 3) Go to step 1
until finished. "What is it that makes this kind of project interesting? In large part, it's a
challenge that tests ones knowledge, ingenuity and resourcefulness applied to
the development of a sophisticated piece of software and machinery - without
the advantage of a DARPA Challenge budget. One crucial ingredient in winning a
seven-figure prize pot is being able to spend six figures on development. The
creators of RoboMagellan must have had this in mind when they came up with this
contest, because the contest seems cleverly designed to make it extremely
unlikely anyone will throw truly large sums money at it (or extremely foolish,
at least), and some of the restrictions seem specifically designed to head off
the possibility. We feel that the Seattle Robotics Society has done a fine job
of creating the RoboMagellan contest as it is. "Some people have expressed frustration
with RoboMagellan because they view it as having a very high cost of
entry in terms of time and money. I took the approach of building an
experimental platform for experimenting and learning about autonomous
outdoor navigation and the secondary goal was to compete in
RoboMagellan events. Intrepid has more functionality than is the
minimum required for RoboMagellan. The competition however, is
beneficial in staying focused while building and testing. One needs
achievable goals. The time aspect must be seen in light of the
knowledge gained. The cost depends somewhat on your creativity. Start
simple, test it, if it works, go with it. "The JBot, for example, is probably trying to
solve a different problem than what we are facing. It is true that
spending $1500 for an industrial grade IMU is going to improve your
dead reckoning navigation. But it is not required for
RoboMagellan. Sensor fusion using an odometer system with good short
term accuracy and GPS with its long term accuracy is sufficient. And
if you can manage it on half what the JBot's IMU cost, you'll
really have something to brag about. "Neither is GPS a requirement. One must be
robust to loss of GPS signal. Consider: What happens if the given
coordinates of the waypoints or cones have unknown positional error?
Then the most accurate IMU in existence won't do you any good. We find
it is best to test frequently under different conditions and pay
attention to error recovery. "One of the hardest problems in this contest
remains the vision aspect. The CMUcam is probably not the best
choice. There are no relatively inexpensive vision solutions out
there, which means one has to develop this from scratch. That kind of
thing is what makes this project interesting, challenging, frustrating
and rewarding." -- Mark Curry -- | 