Navigating the Grand Challenge Course

Posted 11 Jun 2004 at 15:42 UTC by steve

CMU's Red Team has released a technical report detailing the mapping, planning, and navigation system used by their robot, Sandstorm, in the first DARPA Grand Challenge. The paper, title High Speed Navigation of Unrehearsed Terrain: Red Team Technology for Grand Challenge 2004 (PDF format), explains the capabilities and limitations of their navigation system. Both the hardware and software are covered. There's also a section about the "lessons learned" from the experience so far.

---

Evolution does not favor their approach, posted 15 Jun 2004 at 19:52 UTC by tim.holt » (Journeyer)

I read over the paper just now at lunch. My first two thoughts are first, they cheated and that evolution will not favor their approach.

Their approach was to say, "OK here is this particular challenge. We will create something which works based on the rules of this challenge. We won't create a generic solution to the general problem (autonomous vehicle), but rather a specific solution to a specific problem (vehicle on roads/trails between two known cities in known terrain in the US)." One could say they solved the problem (rules of the challenge), but not the Problem (making an autonomous vehicle).

Unless they can automate the process of analyzing aerial/sat images and maps to manually create paths, it just seems like a doomed system.

Oh - RE the evolution comment. This navigational system design is like the bird that eats just one kind of bug. That kind of bug dies off, and the bird is screwed if it doesn't adapt.

Route file for 2004 race, posted 16 Jun 2004 at 00:25 UTC by tim.holt » (Journeyer)

Sort of related, kinda. You can see the route file used in the 2004 race at http://www.nacse.org/~holtt/Projects/Autonomous/darpa_grand_challenge_2004_waypoints.txt

red team didn't cheat (IMHO), posted 16 Jun 2004 at 17:06 UTC by while_true » (Apprentice)

The biggest military applications would have path planners and (possibly robotic) aerial mapping. The point is that the path planners needn't be software based, and could just as easily be humans in battle conditions, just like the Predator is controlled by someone stateside.

Also, given that it is a race, there are a number of differences. First, most offroad racers get to test drive the terrain. How do you turn a sharp corner at 30mph without knowing what is on the other side? This is a fundamental property of vehicles, and not just robots, called "lockahead distance". At a given speed, there is a cone aread of the vehicle where you cannot stop in time to avoid an obstacle. If the course is such that you need to maintain a given speed, while the terrain is such that the robot will not be safe at that speed. then those driving the race will fail in time or in collision. Note that this is true for a robot or for a human.

Second, as a race, the goal is not to have the best robotic technology. The goal is to win. Whatever tools are necessary should be used, and aren't cheating. The Red Team certain developed some powerful robotic tools, like the stabilized E-box and the sensor fusion, but that is clearly not enough given race conditions.

Finally, outside the race, the most lucrative commercial & military applications, e.g. border or perimeter security, military supply lines, second assault (after air strikes) forces, etc. ALL of these have pre-planning and pre-mapping as an available tool to solve the problem.

So if to win the race, you need to map well beforehand and in the conditions where this technology could be used, you would also have access to a map, what is the problem?

BUT, if you still want robots that can map on the fly (but go a bit slower) here are some examples of projects (also at CMU) which could solve the pre-mapping problem:

http://www-2.cs.cmu.edu/afs/cs/project/chopper/www/index.html

http://www.rec.ri.cmu.edu/projects/perceptor/

http://www.frc.ri.cmu.edu/~axs/dynamic_plan.html

enjoy...

Think beyond battles and DARPA, posted 16 Jun 2004 at 20:56 UTC by tim.holt » (Journeyer)

Well all I can say is, it's great until something unexpected comes along. You might not have that 1000 man hours to slap on the research phase of things, because you just got told the bridge ahead is out and you need to take an alternate route. Or that there was a landslide, you think the vehicle can get around it, so you're going to just loosen up the max off/cross track distance to 200 meters and let it figure it out.

The real winner will be when autonomous technology gets put in a wheelchair that can make it down the sidewalk on it's own I think. Forget the warfighter.