If the computing power to build a 3d model of the world purely from video can be fitted in an Aibo, then why are all these machines using bulky off the shelf modules, strung together with such a small code base?

I don't know what budget and time constraints these teams were under, but seriously I think I could do better with a hand picked team, a year to prepare, and a decent budget.

Actually, the 2nd driver was 3 seconds behind the 1st driver. I don't see how he could have rear ended the lead car by speeding up.

After watching the video a few times, it appears the driver maybe wasn't watching the robot car all that well. Poor accident avoidance from somebody that knew he was driving by a robot is my take on it. Nice find anyway.

Well good for you Matt. So we expect you to have one of these running next year and flogging the competition.

All talk... Get out there and do it instead of putting down those who are actually doing it...

I dont think they just picked some random shlums to build these things. There is so much information those vehicles have to collect, analyze, and interpret. But maybe i will make a better one for my brothers 6th grade science project.

i can see this video but most of the other videos i cannot. this may be a restriction of ie6 or the I.T. guys here at work stopping certain video servers or something but yeah... if you can post them all the same as this one id appreciate it :)

Like the above poster, I can only see this video. I tried at home to verify that it wasn't a work firewall and couldn't view the others there either.

to get out of the way the blue car would have hit the red car in the oncoming lane

Matt Matt Matt

First, cast doubt on the premise:
That Aibo's just walk around creating perfect 3D worlds seamlessly with half a processor power. I can't find this, I suspect it's slower and fairly inaccurate. Cars ... well, they go fast and have a lot of mass. You can't just turn, you must consider possible turns that won't roll it, so physics not just kinematics must be simulated 10x/sec. Should we go on for more details?

Cars also tend to run outside. Sun, reflective strips in the pavement, radio interference from all that DARPA chatter ... all this blows wads into the sensing environment. Aibo, runs inside a room with same lighting. It moves slow, it can bump and fall. Any bump by a car causes major structural damage to both parties. A lot of code is just the radio pause and stop system which must absolutely work.

Um, these cars are actually following missions. A lot of cycles are spent rolling forward physically realizable plans and testing them against sensor intersection. We have GPS, Maps, coordinating the two and fusing with radar and other sensors.

Did I mention that cars move fast? That map has to be accurate and formed in real time, fused with other sensors, fused with world and mission maps. Aibo does not physically simulate, at most its kinematics if that. No physics means smoking heap for a car (did I mention they go fast, have inertia, they bounce on the pavement. 10 cm errors at 25mph are different from 1cm at 0.01mph.

We could go on an on about why this is a hard problem ... I can also win every super bowl -- I'm just better at calling the plays then those quarter backs ... well, until I try to get on the field. Does this answer your question Matt?

I actually got to see this live. The article states that the car could have sped up but, regardless of whether that was even possible, that wouldn't have necessarily prevented the accident. The DARPA traffic vehicle should have swerved like they had for previous bots. However, in this instance there was another vehicle coming in the opposite direction (the red car, like Lan points out) so there was no room to swerve.

As for Matt ... You don't just "hand pick" a team. These people have to be willing to devote a full year of VOLUNTEERING endless hours for this project, in addition to working full time. They're not getting paid. And in regards to the "decent budget" and the "bulky" looks: DARPA doesn't give all the teams money to play with, they have to fund it themselves. The budget is up to what the team can afford. If they have cool sponsors like Volkswagon or General Motors then they'll have a much bigger budget. This team is more practical and is actually using the same vehicle (and a lot of the equipment) that it used in the previous two challenges that were off-road races. You should learn more about the event before you critique it.

Estimated low cost.
Car = $30 000
Cameras x4 = $20 000
Processing hardware = $40 000
Various sensors = $5 000

It's a pretty big investment for an average Joe

That estimate is a little low.
lets explore the value of Team Victor Tango's (www.victortango.com) vehicle "Odin".

Note: this is Virginia Tech so some of these items were received at discount, some donated and some purchased for previous challenges.

Vehicle: $29,000
Inertial Measurements Unit: $35,000
GPS: $15,000
Computers: $6,000x3 = $18,000
SICK LRF's: $5,000x4 = $20,000
Cameras: $1,000x2 = $2,000
IBEO LRF's: $75,000x3 = $225,000
Various Electronic Equipment: $5,000

Second Vehicle equiped with everything except the IBEO's: $124,000

Team travel (including vehicle shipping): $45,000

6 Full time Engineers for 1.5 yrs: $500,000

11 grad students for 1.5 yrs: $300,000

45 undergrads for 1 yr: $0

Total Value not including a @#*% load of engineering time: $1,194,000 (This value is approximate)

For anyone who hasn't driven a Taurus, to accelerate from 10 mph to any mph takes a very long time. My foot was on the floor and the car barely speed up, I may have been going a maximum of 14 mph at the point of impact. Before you judge, know the facts, a video can be very deceiving, the Taurus is slow, and Matt, well lets just say, I'd really like to see you there next year with your "hand picked team" and a "bot" you have no idea what it takes to get there! I hold the utmost respect for all the teams that were involved. Not to mention all the staff members, and my fellow drivers, especially NQE test area A drivers!