Stanford and Volkswagen Create Autonomous Vehicle 235
nght2000 writes "Stanford University has created an autonomous driving robot to compete in the 2005 DARPA Grand Challenge Race. The race will be held on October 8, 2005 in the desert Southwest. The team that develops an autonomous ground vehicle that finishes the designated route most quickly within 10 hours will receive $2 million. The route will be no more than 175 miles over desert terrain featuring natural and man-made obstacles. The Stanford Racing Team's vehicle is a Volkswagen R5 turbo diesel Touareg that was donated by Volkswagen of America. The Stanford Team has been working with the Volkswagen Electronics Research Laboratory on the project."
I worked on this project for a few hours (Score:5, Interesting)
Re:Red Team (Score:5, Interesting)
Re:Uhhhh... No... (Score:2, Interesting)
Re:this specifically won't work (Score:5, Interesting)
Google / Stanford team too? (Score:4, Interesting)
Re:But how many humans can do the job? (Score:3, Interesting)
"On the road of life... (Score:5, Interesting)
As someone who went to CMU, I'm of course rooting for the home team, but it is fun to read about the other guys. For the on-road stuff, they had those trucks zipping driver-less, pretty fast, through Schenley Park back in the 90's, so it'll be interesting to see if they can keep on the trail this time for the off-road challenge.
Hi Stanford & CMU (Score:2, Interesting)
-Cornell
Re:I worked on this project for a few hours (Score:1, Interesting)
Reinforcement AI has promise, but it seems to have too much hand-waving and magic black box functions for its own good. Unifying it with algorithmics and logic AI would probably be more useful, but not nearly as sexy as RoboJeep terrorizing the desert fauna. They should award points for the ability to project flame or ripcut undergrowth, now that would be bitchin'.
I Cried (Score:4, Interesting)
Thats simply not the case. DARPA hands out the final destination a day before launch and the teams madly scramble to find a route to send their vehicle down (on nice sat photos). Then they send the vehicle off on its own. What sort of fun is that?
Knowing this, I'm ashamed how poor last years competition was. The winning team was pretty sweet, but I certainly expect a lot more competitive entries this year. Hand most any college worth its salt $25,000 and let the CS & ME's go to. In a year they should build something which could at least contend with the DARPA incumbent.
As it stands the whole thing requires almost no intelligence. The whole point, from a computer engineers' biased persepctive, was to get people building robots aware of their surroundings. The Berkeley city auto-mapper robot is a perfect example; couple that with Sandstorm and then maybe I'm interested. But so many teams can make a robot which FAILS to track a GPS path while staying moderately on the road is just beyond me.
I understand the whole point is that the terrain is supposedly "hostile"... But when you're driving an `86 Hummer, its quite apparent that any area full of enough dangerous terrain to give you a problem will likely be seen on the sat-maps.
Myren
Re:But how many humans can do the job? (Score:4, Interesting)
Parts of that is paved roads, parts unpaved roads and parts "offroad". This means you can do like walking-speed on the offroad-parts and still manage it fine.
Infact I'd take a bet that 9 out of 10 got-drivers-license-yesterday humans would be able to do this in less than half the time allocated to the robots, probably a good driver would do it in a quarter the time the robots get. That'd require him to average 68mph.
My school is in this competition (Score:4, Interesting)
In terms of technology, well, outside of the Turing test, this is sorta like the Super Bowl of AI. My team/part of the project dealt with Machine Vision, which has proven to be quite difficult for a lot of people (including me!). Real time scene analysis is *very* computationally expensive, and you have to make guesses and inferences as optical signal data fluxes around constantly, a lot of the time completely rendering your approach useless.
Even though from life experiences I know that Life Isnt Fair, and the playing field is never level, some of these teams get insane advantages. I wont even go into CMU (ok, I will: they have basically Defense Contractor backing, parts, and consultants, and like 7 million dollars to spend on the project), and here stanford has sponsorship with volkswagen. I was suprised Cal Tech didnt get more major sponsors, but they might have for round 2 of the challenge. No one has near the advantage of CMU though, their main LIDAR cost more than a lot of people's whole car/setup.
Aside from that, for me this project has been a blast. The work, needless to say, is very unique and its almost like a mini-1960's space race, "first one to the finish line!". Its funny how some people try different angles, spend millions of dollars, and then get foiled by a rut in the road that hangs their car up (I'm tellin ya, if the sun shifts even slightly all vision input outside of lidar can basically go to sh!t if you arent careful, and if your lidar doesnt pick it up, well...)
Regardless of whoever makes it to the top 30, it will be interesting to see if anyone finishes this year. Darpa3, maybe?
Re:I Cried (Score:4, Interesting)
For instance, my school's robot was doing well until it hit a chain link fence. As it turned out, the chain-link fence was almost invisible to the car's vision system -- think about it, it's a bunch of air with these little tiny lines which are pretty hard to distinguish from debris in the air and such. You try writing a computer program that can accurately determine the presence of chain link fencing in a photograph and then see what you think.
As for GPS, again it doesn't tell you whether there's a fence, cow, brick wall, etc. standing in the way, and GPS was blocked for large sections of the course. If you're suggesting dead reckoning instead, note that that's really, really hard even under ideal conditions and essentially impossible outdoors. If a car gets one degree off course and travels 60 miles, it'll end up a mile off of the road, perhaps gleefully crashing through houses / oncoming traffic.
Re:The people's car... (Score:4, Interesting)
God, yes.
I live in Atlanta. I work on one side of town, I room on the other. I go 25 miles each way. Average speed of 25mph. That's 2 hours a day I spend in the car.
If I could get into my car, type in a destination, and read, have breakfast, catch the last 10 winks, or write the great American novel while the car did the work, I would jump at it like a shot.
I realize that what I want out of an autonomous car is available, mostly, as public transportation. Unfortunately, public transportation in Atlanta is a joke. To do the 25 miles from my house to the office takes 2 hours, on 2 different systems, with three transfers. That's 4 hours a day in transit, provided nothing breaks down and the buses aren't late. I tried it, and I had just enough time left over in the day to sleep. Not eat, just sleep.
I saw a test car and strip of highway (somewhere in California, IIRC) that worked together as an autopilot. Drivers could enter the freeway, tell the computer what exit they wanted to get off at, and let the car drive itself. Little pips in the tarmac told the car where the lanes were, the on-board did the steering, and the central controller managed congestion by telling the cars what speed was best for the volume of traffic, when to change lanes, and when to wake (pardon me, alert) the driver that the exit was near.
Anyone out there remember this? Is it still under development?
Anyone care to speculate how soon I can get a robot chaffeur or auto-highway?
And does anyone remember...Sally? Asimov fans will know what I mean.
Re:The people's car... (Score:3, Interesting)
Re:But how many humans can do the job? (Score:1, Interesting)
This money can be much better spent developing similar applications whereby urban congestion can be relieved by enforcing "Autonomous Zones". i.e. once you're off the highway, your can goes into Autonomous mode to negotiate the congested city streets.
S.
Re:But how many humans can do the job? (Score:5, Interesting)
Even in relatively benign terrain, a speed of about 15 MPH is actually moving pretty quickly. These aren't $2M one-time-use lightweight 500HP Paris-Dakar desert racers with a navigator, an 8-ton supply-laden chase truck. These are extremely heavy fully autonomous machines. If you read the rules, they're even supposed to refuel themselves without human intervention should it become necessary. It's really, really easy to break stuff at only 15 MPH, particularly when you consider how heavy these robots are.
Also, the paved sections are very short -- I haven't looked at the 2004 course in quite awhile and I'm not sure if the 05 proposed courses are up yet, but it was something like only 10% of the entire route -- and then you're not permitted to exceed the speed limit, which I think was pretty low -- 50 MPH or thereabouts.
It's very, very hard.
Re:Hey, VW, how about making diesel cars (Score:5, Interesting)
Golf, Beetle, Jetta, Passat, Toureg are all available with TDI engines. Try em out, but the waiting list is pretty lengthy because they are selling like freakin' hot cakes.
My wife and I keep our TDI pumped with biodiesel too. Less emissions, less smell, and our gas was living plant material mere years (or months) ago. Staying in the current carbon cycle is better than releasing carbon stored millions of years ago.
DUI (Score:3, Interesting)
In addidition to convenience, autonomous, or even street-directed, vehicles could nearly eliminate the problem with DUI.
With smart streets, traffic control could also be much improved, with, for example, the freeway directing vehicles to shift position slightly for injection of merging vehicles.
Speeds could be significantly increased, and vehicles could be placed on bulk carriers (e.g., trains).
With good enough control and timing (many years after initial introduction), vehicles could be sent through intersections in the holes in cross-traffic . . .