🚀 go-pugleaf

In article <9322@aggie.ucdavis.edu> tuck@iris.eecs.ucdavis.edu (Devon Tuck) writes:
>Each year, UC Davis has a "Micro Mouse" contest in which student teams
>compete in a timed maze-run using small robots they have constructed.
>
>A few of us are playing with design ideas, but would be pleased if we
>could bounce ideas off of the membership of this newsgroup.  If this is
>an inappropriate topic, or if there is a better discussion group for this,
>please tell me.

I can't think of a better place to bounce mouse ideas.  I'm currently 
engaged in re-tackling the problem myself after competing at the 
North American Micromouse competition in Toronto (great time.  Lotsa
fun.  Strongly recommend it).  

>Some Design Considerations:  
>	- Awareness of surroundings, memory of where it has been
>	- Ability to maneuver quickly in at least 4 directions, or turn easily
>	- Ability to correct itself if it gets into weird orientations
>	- Well developed algorithm for mapping the maze
>
>In particular our team is concerned with how we can make the mouse move in 4
>directions without having to pause for turns.  We had thought of using 2 DC
>motors (stepper motors won't be fast enough), to run two central wheels; this
>would allow us to turn quickly, and could elminate back-ups since we could
>switch direction.  Another idea is a large central track ball, which could
>virtually eliminate turns.
>
>Are we overlooking other superior ways in which we could accomplish 4-way
>movement?

Superior?  Loaded question.

Well, my mouse uses the guts of four regular Microsoft mice to make
a frame which could independantly raise and lower wheels in either x
or y directions.  The 90 degree roller axes were mounted on a cross frame
and grommets added to the round contact switch sensors (yes, these were old
style mechanical mice, not IR).  A motor on each side of the square mouse
drove two grommet wheels by band drive, and a single motor looked after the 
raise/lower mechanics.  The result was a mouse that was perfectly 
four-way symetric with a change-direction speed of less than a 20th of 
a second.  Four of the eight wheels were in contact with the ground at 
any one time giving maximum stability and traction.

Such a mouse has advantages, not the least being that as it does not have
to turn around, control is simplified to just staying orthogonal to
the maze walls.  This was done by simple speed regulator circuits on
each edge motor, the inputs coming from rolling tactile wall sensors.
As the mouse moved down a corridor, the carefully adjusted sensors thus
forced the mouse to stay dead middle.  They did the same thing every time
the mouse had to make a turn, keeping the mouse from hitting the walls in
a corner.

Fact is, this design took away so many control headaches that my mouse
did not even need a processor, just a simple 3 chip state engine.  A
processor is going on the next one (maybe) but it was nice to see 
what could be done without it.

I tried using the idea of a single ball driving the floor but came
across two problems: One, the drive ball must be under pressure from 
the perpendicular driving shafts just to maintain friction.  As this 
happens in 4 places around the ball, the energy losses of rotating the 
rubber ball surface over two inert areas I think mounts up to 30%.  Two,
because of One, the balls don't last too long and your mouse gets hot.
Conclusion, using a mouse as a position sensor is all it's good for.

Regarding algorithms, the competition I entered did not use a 
connected maze, so the simple right-hand wall following rule would 
not work.  My system used a slight modification to this which was...

	prime = 3
	Right-hand rule, turning left (if possible) every prime.
	every (tee or cross) junction, increment the prime to
		the next prime.
	when goal reached, lock current prime in place.

I haven't tested this algorithm rigorously, but it seems to work.
It was the best non-random algorithm I could implement on my hardware.

It turns out the algorithm you use will change drastically depending on
if you use over-the-wall looking sensors or low sensors which can 
only see the immediate walls around it.  Over-the-wall sensing means
you can fill up your internal maze map with less travel, but the 
weight of the sensors typically makes for a very top-heavy mouse which 
cannot turn quickly.

An interesting discovery by MIT during their recent racing in Singapore
was that a mouse which aggressively seeks the center of the maze without 
checking to see if there might indeed be a faster path eleswhere does
have a better overall chance of winning.  'Mapping' a maze takes time
which will count against you in the final scoring.  Simulations on 
lots of previous competition mazes proves this out.

In terms of mouse designs, the MIT two-wheel-two-castor cart design
is the current world record holder.  The main motors are small and close
to the center of mass, as are the drive wheels.  Balancing castors are placed
at a distance to stabilize the outfit.  It makes for a very low and light 
engine which can cut corners effectively.  There are also close competitors
which use powered outside castors to hasten the mouses turns.  The 
speed is equivalent, but the mouse can be made significantly heavier
if you need it to be so.

My current design ideas include the following: a motorcycle mouse (wide wheels,
high traction, bi-laterally symetric, small), a mouse 'ball' (large sphere
which changes internal centre of balance to move.  Won't be fast but 
will make a good 'demonstration' mouse), a 'jet' mouse (built in ducted 
fan to assist acceleration and also suk the wheels onto the deck for
better traction), a Lamborgini mouse (made from a posi-track Tamaya
model kit.  It'll fit and look good) and a 'mouse' mouse (a mouse built 
out of a single ordinary computer mouse, using the trackball to give 
positional information to a 68C11 uP.  Two motor, two castor drive),
and, of course, improvements to my first mouse. 

One of the interesting things about micromousing is that, like potato
chips, you can't have just one, you've gotta build another.  My preferance
is to go the unorthodox route and build no two mice alike (there's an
awful lot of cart designs around.  It's good but it's peaking in 
effectiveness IMHO.  Don't be afraid to go wierd).  The other reason
to go unorthodox is that in two years the next International Robot Olympics 
will be held, and modifying mice to enter one or more of those events
will be a must for those that can (I just got a brochure of some of the
competitors.  Most look like an explosion in a Meccano factory.  Just 
beautiful).

Of course the whole reason to get into micromousing is not the 
competition so much as the fun of doing it.  Most people who got into
robotics were not inspired by robot assembly lines, but by Star 
Wars and Silent Running and other shows which showed robot
creatures doing neat things.  Robotics needs industrial efficiency
to make it pay, but after hours, micromousing is one way to keep 
the imagination greased.

Anyway, I have all kindsa information (including full on line rules)
if anybody's interested.  Videos may soon be available if I can get my
butt in gear.

Is all.



-- 
Mark Tilden: _-_-_-__--__--_      /(glitch!)  M.F.C.F Hardware Design Lab.
-_-___       |              \  /\/            U of Waterloo. Ont. Can, N2L-3G1
     |__-_-_-|               \/               (519) - 885 - 1211 ext.2454,
"MY OPINIONS, YOU HEAR!? MINE! MINE! MINE! MINE! MINE! AH HAHAHAHAHAHAHAHAHA!!"