First step done!

Can you describe parts?  CD ROM drive?
What are you using for video capture
and processing?

Rich

Hehe, someone noted....

And if you look closely, that red box is a set of drills :-)

In actuality, this picture was taken moments prior to its first (remote
controlled) test drive. It carries a consumer digital camera just to take
onboard footage so we can determine how easy or difficult will be to create
the vision system once the final cameras are in place.
Because the suspension was replaced by a more rigid one, if we didn't put
some dummy weight on top of the rover, it would bounce too much. Therefore
that huge CD-ROM and the drill box are there only to make some weight.

The instruments it will carry are the follow:

1 CMOS camera
1 CCD camera (pan and tilt)
2 GPS units (one at 4Hz and another one weak signal)
Bumper switches
1 digital compass
2 sonars (pan and tilt)
3 gyroscopes (pitch-roll-yaw)
1 wheel encoder
Assorted telemetry (batteries, temperature)
rs485 bus
mini-itx cpu with wi-fi (cardbus) and CF, OS will be installed in a 2GB CF

Sensors are listed in order of importance.

My next step will be to assembly the mini-itx cpu, install the OS and mount
it into the rover with appropriate shock absortion and protection.

It is a regular RC car with two 14.4VDC motors, steering is ackerman style.

Is gas powered not an option?  You can
get more run time for the weight of
a tank of gas.

Rich

I started with gas, but it is not really practical for this application. RC
gas based (or nitro-methane) are adjusted for racing performance, and being
a 2 stroke engine, they require fine tuning of carburators, they don't idle
very well by nature and they are poor performers in low revs. Besides that,
it would be impossible to do demonstrations indoors because of the smoke.

What I'm really hoping for is that we get at least 5 miles of autonomous
navigation on rough terrain, enough to prove our point and get the budget
for a full size 4x4 vehicle to enter DARPA GC.

My opinion is that if you have tons of cash to invest on very sophisticated
sensors, you won't sweat hard enough on efficient vision algorithms, and I
think the answer is in vision and intelligence, not on ladars and sonars...
want a proof?

If I mount one or two cameras on a hummer and let you teleoperate it through
the internet, I bet you can make it go from point A to point B through rough
(not extremely rough, mainly dirt roads) terrain. Consider that besides
vision you also have GPS coordinates.

Most animals in nature don't have ladars or sonars (well, at least not the
ones that have good vision)...

Padu

Is there a time limit for this challenge? Is it at all possible to stop
for periods of time and use a solar cell to "top off" the battery?

-- Gordon

"Gordon McComb"

In the real DARPA Grand Challenge, yes. You must complete the 176 miles (or
something similar) in less than 10 hours. I don't think there's anything
against solar panels, but I do know that you cannot stop to refuel. By
refuel I understand as something we humans pre-position on the way of the
rover so it can get from the environment.

I had thought about solar panels, but their limited amperage made me gave up
the idea, but I've never thought about stopping for a moment until the
batteries are full again... that's a good idea, thanks. Realistically, I
don't have any expectations of completing 176 miles in less than 10 hours
using an RC car, so stopping for a "recovering breath" may be a good option.

How long would it take to recharge my batteries, assuming I have a maximum
of 12x12 inches area for the solar panel and 48 1.2V "sub-C" cells of
3700mAh each? (12 cells in series x 4 = 4 packs of 14.4V, total of 14.8A).

Padu

I understood about the Darpa GC, but I was wondering if this test-case
has to cover those five miles in a specific period of time. That would
allow some recharging time, or a slower pace, and allowing the solar
cells to contribute at least some of the current to the motors.

-- Gordon

"Gordon McComb"

(or

The test-case goal is to prove that we could navigate the 176 miles of the
GC, therefore, I don't believe stopping to recharge would be a problem. If
it proves to be feasible (half an hour instead of half a day), then
implementing solar panels would be an additional attractive, not to say that
it would be able (in theory) to navigate the 176 miles.

Since the test is mainly for the navigation prewess of your vehicle,
would it be allowable to have it simply stop to allow you to replace the
battery pack?

Frankly, if you can get this thing to go even a mile in rugged terrain,
and end up a meter or two from where it's supposed to, you can hang the
Darpa challenge. You'll have something quite marketable right there!

-- Gordon

I don't think solar cells will recharge your batteries quickly enough to
justify the amount of time you will have to "rest". Plus, you are at the
mercy of the weather, in that you hope you have enough sunlight during your
rest period to make the solar cells useful...

Have you considered some type of fuel cell to recharge the batteries, or
even to supplement your power supply? Granted, they are expensive, but they
might be the answer to your problem and there are a few "toy" fuel cell kits
on the market to allow you to experiment with them on your test rc car.

(You can even devise a way to jettison the fuel cell once it is spent, to
help reduce your vehicle's weight - assuming the fuel cell assembly is heavy
enough to justify the effort.)

Just a thought...

~WEC

In six months?!?!  Or you don't expect a winner this year?

Mitch

What is the farthest traveled distance in past
DARPA competitions?

I don't see why someone doesn't get a truck
with automatic trans. (sponsored by ford, GMC)
put in in DRIVE, and throw a brick on the gas
pedal.  Would be a cool commercial.  

Rich

less than 10 miles for sure.

Actually your suggested approach is not too bad. If I recall correctly, one
of the teams that was having problems on the qualifying rounds disabled a
series of sensors, keeping only vision, gps and a couple of basic others.
They got better results with less sensors.

Yeah, nowhere near the full coarse.  I'm pretty sure (90%) the record so
far is that red hummer at 7.somethin miles.  Appearantly it's not an
easy thing :-)

--Andy P

Sandstorm (the red Hummer) did go the furthest last time at seven miles.
   According to Red Whittaker when he came to talk to our company about
the race last year (we're a partner in the project) Sandstorm got hung
up on the last major obstacle.  Most of the rest of the course would
have been relatively straight forward desert driving.  I expect at least
one or two machines to finish this year.

Catman

Andy P wrote:

"Mitch Berkson"

Forget this year, this project started after the deadline for this year's GC
had already passed. Given that last year the longest run was of 9 (or 6,
don't remember) miles out of 170 and something, I don't expect a winner this
year, although looking at some of the top teams' web sites, they look very
promissing.

It's really interesting. I started a similar project with similar goals
(low-cost re-rash of the DARPA GC) and similar components (R/C car, cameras
for vision, PC-based brain). You can take a look at the current status here:
http://andras.tantos.homedns.org/robot.htm .

It's great to see that others think along the same lines as well! I however
have some doubts about your goal of the 10 mile range. My measurements and
experiments show that I won't get more than half an hour of operation out of
two 7.2V 3Ah battery packs. Do you think you can navigate with an average
20mph over rought terrain? Or do you think that you can get significantly
more operating time out of these batteries? Or do you use a different power
source?

Regards,
Andras Tantos

"Andras Tantos"

here:

Very nice work! Any special reason you chose the stampede over the e-maxx?

Regarding the 10 mile range, after initial testings we are doubting that
will be the case, but it will be at least 5 miles. This week I got my
shipment of 64 cells of GP3700 nimh batteries. While some of them are going
to power the computer and the sensors, 48 of them (4 packs of 12 in series)
are reserved for powering the two 540 14.4VDC motors. (14.8Ah or 213W).

For the 10 miles range, we initially though about lithium polymer (lipo)
batteries, but they are still too expensive. We would have spent about 1.5K
only on batteries. Also, they require special care on recharging.

I made some onboard video going over rough terrain this past weekend and I
saw that it's going to be difficult to achieve 20mph over rough terrain. So
we decided to go ahead and work with whatever limitations we find. Hopefully
we will be able to navigate this thing until the battery stops, and if we do
so, we can get another grant to upgrade the platform... a real sized one?
who knows.

I quickread your page, I'll bookmark it and read with more care, as I'll
probably stumble over many of the problems you went through.

Padu

Hi Padu!

Thanks. The main reason for me to go with the Stampade was price. I will
also have to replace the springs just you did due to the additional wheight
but I thought I will not need the extra power of the second motor.

Talking about weight, the 64 nimh cells must be quite heavy. My aproach is
to eventually reduce power consumption of the electronics on board by moving
to something other than a PC and extend the range that way.

I was thinking about using some high-speed accelerometers or gyros to
capture the orientation of the cameras at the time the images were taken.
That way I will be able to compensate for the bouncing around on the
terrain. Kind of like a home-brew image stabilizer.

There are quite a number of problems with high-speed navigation apart from
the mechanical rigidity problems, and so I didn't at least for the beginning
think about doing that.

One obvious thing is that the processing requirements will be quite high. My
initial testings show that the processing of a stereo pair of images on a
1GHz PIII takes about 200ms at least with the algorithm and resolution I
choose. That's not counting the map building, vehicle localization, path
planning, etc. This means that I don't think I can get faster than 4-5
frames per second. If you go 20mph, or around 30km/h, your vehicle travels 5
feet or 1.6m between each frame. It might be OK for a large 4x4 because the
mechanical time constants are quite high anyway, but for a 1:10 model car
such separate samples will make controlling really hard. Though from your
higher number of batteries I would guess that you can get ~1.5 hour
operating time and thus the ~3mph average
speed can get you to the 10mile range. With that speed 5fps sampling might
be adequate.

The other problem is that with traditional HW (USB cameras) it's impossible
to take the frames from the left and right cameras at the same time. If you
move, the time error will amount to a location error, and will make your
processed data less precise. You could for example use a synronized
mechanical shutter to prevent that but than you would have to have some kind
of an iris as well to control the exposure. What are your plans for dealing
with this problem?

The third issue, is that at high-speed travel you will probably have to have
a more precise physical model of your car to prevent slip, roll and in
general keep it in the safe operating region. Again, more processing power.
Because of the long stopping distance you would have to indentify obstacles
sooner and plan ahead longer. For that you would need a more precise model
of your surrounding which probably results in higher resolution requirements
or zoom-lenses. One makes the processing power requirement sky-rocket, the
other is expensive.

One thing that high speed helps in is that since you can't make sharp turns
anyway, you don't need wide angle lens - something that's hard to deal with
because of its high cusion-distortion.

Anyway, all I'm trying to say is your project is really ambitious. I hope
you succeed. Good luck with it and keep us poseted!

Andras