Digging out and digging in

Foreman's log, 8:21 am. Snowy and cold. Have braved the chill and frost to survey the excavation site.
Foreman's log, 8:23 am. Snowy and cold. Have finished transferring equipment to the site and setting up camp. Workers complain of inadequate compensation.
Foreman's log, 8:24 am. Snowy and cold. Have begun digging.
Foreman's log, 8:25 am. Snowy and cold. Progress is slow. Have breached the outer ice barrier and have begun digging in fresh powder. Excavation depth steady at one foot.
Foreman's log, 8:32 am. Snowy and cold. Have hit bedrock in southwestern portion of the dig site. Workers request dynamite or other suitable explosive to continue.
Foreman's log, 8:35 am. Snowy and cold. Have encountered first foreign object buried in the far reaches of the subterranean excavation. Identity of object unconfirmed.
Foreman's log, 8:39 am. Snowy and cold. Fine brushwork has confirmed color of object in question. Have laid claim to dig site and filed paperwork for confirmation of ownership.
Foreman's log, 8:44 am. Snowy and cold. Having equipment problems. Workers complain of mild convulsions and wet conditions. Returning to base camp for warmth and medical attention.
Foreman's log, 8:46 am. Snowy and cold. Excavation continues east from location of first breach. Workers have confirmed a round structure matching expected parameters. Excavation depth steady at three feet.
Foreman's log, 8:47 am. Snowy and cold. Excavation begins on the north face of the object.
Foreman's log, 9:01 am. Snowy and cold. Equipment problems persist. Have discovered two more round objects on the north face of the dig. Identity of query has been confirmed.
Foreman's log, 9:13 am. Snowy and cold. Have removed several inches of snow and ice from exposed top of the object. Workers complain of faster progress at neighboring excavation sites.
Foreman's log, 9:25 am. Snowy and cold. Attempt to breach object from east face has proven successful. Workers have now excavated to bedrock on the north, south, and west faces of object.
Foreman's log, 9:30 am. Snowy and cold. Total equipment failure. Work continues by hand.
Foreman's log, 9:39 am. Snowy and cold. Workers remove the final inches of snow underneath object. Workers have excavated to bedrock on all four sides of object. Fourth round object discovered on the south face of the dig site.
Foreman's log, 9:45 am. Snowy and cold. Final inches of material removed from object sides and top with fine brushwork. Workers demand better site conditions. Foreman concurs.
Foreman's log, 9:53 am. Snowy and cold. Success! Object finally extricated from icy prison. Excavation site shored with ice for future use.

...Ok, time to go work on DUC.

There will be no autonomous driving today

Being up north, Ithaca knows snow. So, when the mayor starts declaring all the emergency snow routes in effect several hours before the snow even starts, you know it's probably going to get interesting. And, true to her word, mother nature didn't disappoint. There definitely will be no autonomous driving today.

I almost think we need autonomous cars just so they can say "beep--boop--don't drive today, moron--beep--boop." At any rate, today's definitely a simulation day.

Hang on, let me shift to park

I usually work at home on the weekends. I have a number of reasons for doing so, not least being the improved concentration, quiet, and the drastic reduction in the number of times I hear the words "meatball," "linear systems," "architecture," and every curse word known to mankind. This weekend was by no means an exception to the rule. I had an important architecture decision to ponder, and I really needed to think about all the ramifications of learning the road structure in a factorized form vs. learning it all simultaneously (incidentally, this is part of what I call scene estimation, for those who are curious).

Anyway, the weekend was almost business as usual. I showered, jumped into my pajamas, and sat down to eat as I watched the NBC lineup from Thursday. I was halfway through an apple when I got the call to go testing. I got dressed, splashed some lemon juice on the half-eaten apple, and ran out the door.

I arrived at the parking lot to find the Tahoe driving itself(!) Many congratulations go to Noah for successfully disassembling the brand new Tahoe, building the steering actuator, and putting the thing "almost" all the way back together. Kudos also go to Brian for writing RemoteView2, which has several important features over the 2005 version:

1. When the car comes to the end of a path, it does not do donuts.
2. Screen glitches involving large sky-blue triangles no longer appear.

Also, for those who remember, "long path" and "long path 2" made a reappearance. Here are some pictures of the inside of the Tahoe at present:

Poor thing, that used to be a brand new car. You can see the inverter and the power reserve in the foreground, where the middle seat used to be. Here is a picture of the steering actuator:

It's kinda tough to explain what's going on in the picture, but at least I can point you to the black breadbox-sized motor just under the steering column.

Long story short, a lot of people did a lot of good things this weekend. To close, I'll leave you with a quote I heard yesterday evening:

"You know, I feel like we're getting better at this... the whole building autonomous cars thing."

And in case you were curious, I returned home to a perfectly unoxidized apple.

Where pose goes, nobody knows

We had a bunch of people come to Cornell to visit yesterday and provide feedback on our system design and test plan (a bit like a critical design review). Anyway, one of these fellows had the nifty suggestion of evaluating the fused GPS / INS system (the thing that tells our car where it is) by plotting it against one of the new fancy-schmancy mapping programs. I spent a few hours doing that this morning, and here is a sample of the results:

I made this using the filtered output of the pose estimator (currently rate gyros + accelerometers + GPS) dumped into Microsoft's Virtual Earth. This is what a nice tightly-coupled pose estimator can do, when GPS is not allowed to run rampant by itself. The result is a very smooth, very accurate, non bridge-seeking system.

Also, our steering actuator setup has apparently been nicknamed "two bolts and a dude." I don't know what that means exactly, but at least the guys who are building it do.

Ok, ok, I know

Yeah yeah yeah. Look, I know what you're going to say. Yeah, it's been a long time since I've written anything of substance up here. I mean, sure, the Discovery Channel thing was interesting, but it had nothing to do with the combination of Cornell + cars + city that draws everybody's interest (and by everybody, I really just mean people that read this, all... both of you).

Anyway, the last time I really wrote anything interesting was before we went on break, and before you ask, no, I did not take a month and a half break. It was more like... well actually, it was more of a day and a half break. See that's the problem with zealous perfectionists, we just can't slow down. It's not that we've been on break, it's more that I've been too busy to think about writing.

So what has happened since December? Well, mum's the word on a lot of things, at least until I can figure out how to get video up to the server. Until then, I'll just give you a few of the big highlights.

First off, the car still hasn't driven itself (at least not outside of simulation). Yeah, that's about a month behind schedule. Nobody's really pleased about that (especially our poor mechanical guys that had to put up with a few choice shipping and procurement snafus in the actuator department), but the team is so parallelized that it's not really fazing anybody. (no warm fuzzies yet, but hey, we're patient) That being said, the new alternator and inverter kit have been installed, and our car quite literally has enough power to run a small house.

Second, pose is just about finished. I know just about everybody else would just buy an off-the-shelf unit and be done with it, but I'd like to think that we have more information about exactly what happens to GPS in different urban situations. It gives us a lot of control and better performance, and nobody can argue at the 90% discount over the commercial version.

Third, our mapping and fusion algorithm is getting ported over to the vehicle as I write. We seem to be able to track cars reliably well past 100 meters, and I think the effort we put into it over the past 6 months will prove to be a boon later. I have some nifty videos for you guys, as soon as I can make them manageable without drowning us in bandwidth bills.

Lastly, our planner is..., well, it is. There are of course the rules of the road and the general routing algorithm, but I don't think anybody would argue for them as the tough part of the Urban Challenge. Granted, last week's slug-fest between the sensing team and the planning team was borderline hilarious, but the real problems start to surface when there are other cars on the road. There is no "optimal" when it comes to unpredictable drivers- even the magic 8-ball would say, "ask again later." I find it humorous that DARPA emailed out a surveyed list of all the vehicles' bumper heights, and I find it utterly disturbing that the smallest competing car can nearly pass unimpeded under the largest competing car. It just goes to show you that even self-driving cars need a hand once in awhile.