Keep in mind, however, that the contestants will have to travel from Las Angeles to Las Vegas. A simple trip to MapQuest tells us that the minumum distance for driving is approximately 400 land miles. At 15MPG you'd need 27 gallons of gas. Round that off to an even 30 gallons just for inefficiencies.


If I were competing, I'd probabaly do a dune buggy with a 2-stroke diesel engine. The technology has come a long way and fuel efficiencies for some of these new 2-strokes they've created are pretty large. Diesel has much more energy output for its volume, so it would be preferred, even though it weights 1lb more than gasoline per gallon (6.1lbs/gal for gas).

1. I'm not sure what SUV you drive but my 99 4x4 Auto Ford Ranger gets 13.4 MPG and my dad's 96 4x4 Auto Cherokee (that he used to have) got 18 MPG at best.

2. You wouldn't refill. Too difficult. Just carry enough fuel to begin with.

3. Agreed.

Last: Dune buggies are only 2wd though and those air cooled VW engines might not make it in the heat. You could go with a custom powertrain. Wheelbase is long, track is wide. Large tires and a beefy suspension are a must. although not at all difficult to implement. I would even argue to have beadlock wheels and perhaps a HUMMER style CTIS system. Electric motors would work when powered by a diesel generator but why? Much more costly and time consuming to build for hardly any benefit. Diesel engines run more reliably that gas ones however you could use a newer gas powered vehicle with a car computer to monitor systems. For sensing and navigation I think a GPS, stereoscopic cameras and a laser scanning system would get the job done.

I notice how quickly some people think themselves into boxes.

Dune buggies? SUVs?! electric motors?!

this machine doenst need to carry any passengers or cargo - so there is no reason for it to resemble a car or dunebuggy - it doenst need seats, a trunk, windows, headlines...

you could strip a jeep down to its frame and drivetrain, put larger than normal wheels on it, a fuel tank large enough to get you there without refueling

then all you need to do is control the speed and steering.

They will give you the course 2 hours before the event starts. If you WERE doing this in a car or a motorcycle - what is the first thing you would do?

get out the topographical maps and plot your course - see where the natural obsticals are (streams, rivers, cliffs...) and find the best place to cross them

then you would plot out many many GPS waypoints, not just the one or two that you are required to pass through.

I think (after reading the rules) the biggest challenge will be taking advantage of roads when they are present - being able to stay on them and maximize your speed

and, avoiding obsticals in your path, like rocks and logs, and ditches - I dont think that would be too difficult to do. There are many ways to sense objects in your path, and many ways to measure distance to an object.

Sounds like fun. Wish I had enough funding to take a year off from work and do this instead. I think I could pull it off.

I"m very pleased to see all of the ideas put forth so far. It looks as if the coalition will be a strong one. Please continue to contribute your ideas, because soon, we will need to work out a design and put it into a technical paper for submission to DARPA. Thanks

There is a program at my school that does some work with GPS navigation. The teacher told me that some GPS systems can be off by several meters. Other systems can be accurate to even inches. There is something called differential correcting that gives an additional signal from towers on the ground to get the increased accuracy. I don't know much more about it than that.

BTW, this program at my school is called EAST. I know at least a few schools over the US have the program. The reason I bring it up though is that they got a $200,000 dollar grant for it. (which I am still mad about because robotics is way more deserving).

if we can get some of the EAST people to help out, that might be very halpful, especially when it comes time to design the navigation system

GPS use to have a dither added so that commercial users could not get the same accuracy as military users (who have access to a special password)

the dither was taken out a few years ago - so you can now get an inexpensive commercial GPS unit that is accurate to within a foot or two (postition).

They might have turned the dither back on temporarily, due to the war in Iraq - but if they did, it will be turned back off soon, certainly before next march.

due to the fact that they will be putting obsticals in the path, and there will be natural obsticals to avoid (and other vehicles on the course) you can not meet the requirements of this challenge with GPS alone anyway. your machine is going to be able to sense what is around it (somehow).

GPS hasn't been affected by the war, but the resolution is more like 3 meters, not feet.

Eric.

3 meters?!

um, how about 3 centimeters?

http://www.techtv.com/news/business/...327030,00.html

dont confuse the instantainious - once per update accuracy with what is possible when you average position over several seconds - or when you already know your starting point.

there are many clever ways to get incredible accuracy out of GPS.

Reading through the posts inthis thread I have noticed something, Nobody have made any mention of StangPS yet, granted that it's a much bigger task than on a 48 ft field, but I belive if you had satalite and topo maps and some how worked in the whole automous thing they got going on, I have a feeling it just might work, Put aside all the conserns about fuel, and engines and what not, and think abou this for a sec If we could cordinate and make a 3D model of the competition area, and then use modified StangPS software to pick out check points for the robot to go to, I think that it could work. Granted making a 3D model that big would take some time and some major rendering power, but combined with the Software, it just may work.

> 3 meters?!
>
> um, how about 3 centimeters?

Actually, that article states 25-30 cm.

I don't think that's on the publicly available GPS, though. The article talks about NavCom giving a 'correction' to GPS. The article is probably talking about 'differential' GPS...which requires a fixed transmitter in addition to the satellites. This is easy to implement on a farm, which is relatively small, but probably much more difficult on a 250 mile course known two hours before the race.

The following site shows standard GPS accuracy in the 10^1 m range...and also shows accuracies for differential GPS:

Navy GPS error site

Regarding the instantaneous vs. average, I have a hunch that the error in GPS has a lot to do with environmental conditions. In this case you are not talking about random error, but an error that would be hard to average out. There is another page on the navy site that seems to support this.

I'm not saying that a DARPA vehicle using DGPS and achieving accuracy below the 30 cm range is impossible, but I think it will be extremely difficult.

Eric.

Now...about the idea of mapping out the field in advance....the problem with that is they will not give you the data for the route that you must follow until 2 hours prior to the starting time. That data includes checkpoints, route boundaries and other such stuff that must be followed on penalty of disqualification.

We're going to need a system that is flexible enough to actively respond to obstacles, to modify its behavior to accomodate whatever parameters we recieve

You can't map the field in advance, because according to the rules, you only get the course 4 hours before the start of the competition.

I don't know what SUV you're talking about, but my 03' Honda Accord gets about 25 mph City, 29mph highway. And Honda makes some of the most fuel efficient cars out there.

I wouldn't depend solely on GPS. First of all, the update on GPS is fairly slow - a *good* GPS unit can do about one update every second, given clear line of sight to the satellites. The course will contain various obstacles such as underpasses, trees, gulleys, etc - no problem for a commercial pickup truck, but you will loose GPS tracking signal in a hurry. DGPS might not be an option either, there are probably offroad sections where you won't be able to get a differential signal to correct your GPS.

The vehicle will also have to watch for other hazzards on the road - including but not limited to other vehicles. Considering how hard it is just to make a robot U-Turn and go up a ramp...

Good luck, guys!

-=- Terence

Now, I'm east coast, and can't do much as per this competition, but I do have a few ideas. I've seen lots of object detection technologies out there, including radar (think weather radar maps), proximity sensors, even optical recognition (cameras which can sense and analyze features). The only real programming language I know is BASIC and PBasic, so I'm not quite sure how data can be collected and organized/processed in any other languages, but it seems it shouldn't be that hard to compile information like that together.

All one would really have to do is create a model (either 3d mapped, or just in variables/numbers) of objects relative to the vehicle itself, then compare with input data from one (or a few, with the average used) GPS's, such that if an object is detected in the path of it's current motion, software will turn the vehicle until the object is no longer in it's path, analyze similar sensors on both sides of the vehicle until no further hindering objects are detected, then resume the pre-set course (via preset waypoints, etc). Again, since I'm not a real programmer, I have no idea how hard this would be to create, nor how well input data can be manipulated to produce the desired output.

Thinking logically though (and in relation to our own IFI robot controllers), let's say you have 3 GPS's on the vehicle, one in the front left, one in the front right, and one in the rear center. The output data (usually NMEA) can be averaged between the three of them, to produce more accurate positioning than any one of them individually. This average becomes two variables, an X coordinate and a Y coordinate. Any sensor input data would be defined in relation to the GPS coordinates. If a radar detects an object directly in front of the vehicle, say, 9 meters directly ahead, and the vehicle is traveling in the positive direction on the Y axis, the computer can input the 9 meters given from the radar as Y + 9, and create a plot on a preset map (if the general location of the course is known, e.g. a block of space between LA and LV with known coordinates.) The computer should be scanning it's position based on the GPS data, and it's relative location to mapped objects (Detected by scanners). Add in a cushion for the size of the vehicle, and it shouldn't bee too difficult to get it to turn if objects are mapped directly in it's path.

Again, I'm not a programmer and have no idea what would be involved in this, but thinking in terms of PBasic and the IFI equipment, it's really not too hard (on a simpler scale, e.g. optical sensors, yaw rate sensors, etc).