Alice Goes Four Wheelin' (Autonomously)

[Kevin Watson]

Well, today was Team Caltech's first attempt at qualifying for the DARPA Grand Challenge. We couldn't finish the course due to a bug in our GPS code, but we'll have another chance tomorrow. If interesed, the team has posted a short quicktime movie showing Alice (Caltech's entry) autonomously four wheelin' over a bunch of hay bales on the course.

Edit: Here's a link to a story about Alice and the team published today.

-Kevin

[jdiwnab]

That looks cool. Were they able to find the glitch and fix it? Good luck!

[phrontist]

Why did it stop?

[Greg Needel]

my guess is that they have problems in their software about reinitializing the system after gps is lost and then picks up again. you would lose gps when you are in the tunnel and when it comes back afterwords your computer needs to be able to handle that.

my team had a similar problem while testing but our vehicle ran into the trees when gps came back and we had to use our ekill

[Kevin Watson]

Originally Posted by phrontist

Why did it stop?

I don't know all of the details yet. When Alice entered the tunnel, the path planner has to rely on our LN-200 Inertial Measurement Unit and some very sophisticated state estimation code to guess where Alice moves from moment to moment. When the GPS unit started providing navigation solutions again, it looks like the arbiter may have freaked, causing Alice to go into safe mode.

-Kevin

[Kevin Watson]

Originally Posted by jdiwnab

Were they able to find the glitch and fix it?

Yes, it'll be fixed by the next attempt.

-Kevin

[scitobor 617]

Four wheelin over hay bails is bad? All in all Alice looks very impressive.

[Tazlikesrobots]

Good luck guys! Wish I could be there!

[dlavery]

With the last news update I had, CMU's H1ghlander had the best performance so far, turning in a 9:22 run with no errors this morning. They had a planned time of 9:26, and the actual performance was able to improve on that. H1ghlander's elapsed time and driving were improvements beyond its Wednesday qualifier performance. Sensors that were skewed on Wednesday were calibrated overnight, resulting in the improvement. But several other teams were still waiting for their second opportunity at the test course, so we will see if that record holds (Alice had not run yet when I got the H1ghlander update).

-dave

I was working at Stanford this summer, and I got a tour of the 'car' in July.
From what little they let me know, it seems their testing was going pretty well. I've tried to hook up with them in August, but they were pretty much out in the field all summer tweaking.

From what I saw, I think they have a viable contender. I tried to get onto the team, but they were in 'launch lockdown.'

[Kevin Watson]

Originally Posted by dlavery

...Alice had not run yet when I got the H1ghlander update...

Alice just successfully finished the course in about twelve minutes thirty seconds.

-Kevin

[phrontist]

Looking at some of these entries, H1ghlander espescially, blows my mind. I simply cannot fathom the work that went into being able to map terrain to that accuracy in real time, and then do something meaningful with that deluge of data.

Vae Victus (1418) was visited a few weeks ago by two DARPA "referees" who refused to speculate on the outcome, but are probably there right now.

[Stu Bloom]

Originally Posted by Kevin Watson

Alice just successfully finished the course in about twelve minutes thirty seconds.

-Kevin

Congrats to you and your team Kevin. I hope to be able to make a similar report about IRV before next Wednesday.

[mechanicalbrain]

Congrats. Ive been watching this since it started and I'm pretty hopeful that someone will win this year (or at least blow last years record out of the water). I got my picks for the winners and I hope you guys do very well today. Whatever is developed here will herald a new bread of autonomous robots and some of the developments made to that goal are truly amazing. In short, DARPA really knows its shtick. By the way is their a live feed we can watch? I looked but only found footage of last year.

[dlavery]

By yesterday morning the Stanford team came through and "Stanley" bettered H1ghlander's time through the course by about 4 seconds. They held the lead for most of the day. But late in the afternoon the other CMU vehicle, Sandstorm, turned in a 8:59 run. I believe that this is the first sub-9-minute performance by any of the vehicles. But any of the vehicles that are able to consistently turn in complete runs over thier set of attempts will likely have a good shot at making it into the race on Saturday.

The fact that about half of the vehicles attempting the course on the first day were able to complete it is a very good sign. As I recall, none of the teams were able to get through the course on the first attempts last year. So this is a significant step forward. The field of finalists next weekend should be quite interesting!

-dave