DISCLAIMER: the opinionated political satire expressed here is in no way affiliated or aligned with the views of Cornell University.
DISCLAIMER: the technical investigations of this team should in no way be perceived as anything but a dedicated and rigorous research effort. Please do not reproduce or otherwise make available any portion of the material found within unless permission is obtained from the author first.
I didn't think we'd have to do it, but it turns out we do... we have to estimate the ground. As if there weren't enough estimators running on the car- estimating time, estimating system parameters, estimating position, estimating the road, estimating cars, estimating blockages, estimating existence, estimating stoplines, estimating behaviors, estimating tracks, estimating obstacles, and my personal favorite: estimating the errors in the estimators (next up will surely be estimating the errors in the error estimates, and other epistemological nightmares). It turns out that's not enough- not enough to tell whether the stuff picked up by the sensors is stuff we care about or stuff we don't... at least not in all situations. I guess that's the good part about having an overgrown army depot to ourselves... all the algorithms that look great on normal roads break eight ways from Tuesday on our teeny tiny roads. Well, it's good in the sense that we discover problems... bad in the sense that we lose more sleep writing solutions to said problems.
Naturally, in order to estimate the ground you have to look at the ground, meaning that the fifty billion sensors currently on the car need to be updated to fifty billion and one. One way to do that is to wave the magic wand of estimation. Another way is to strap a seventy-pound gimbal onto the roof of the Tahoe with magnets. Maybe I'm just the pot calling the kettle black, but I'm not exactly a fan of instant death:
Don't worry, don't worry, it's not permanent. We actually threw this on a week ago, tried it out temporarily, and took it right off again. After all, there's no point sweating an aesthetic solution until the ugly one works, right? Well, we've got three things to try, so hopefully one of them will work.
In other news, the guys ran the car 27 miles yesterday before getting bored. I'm told it looked really good, except for the one turn that the planner always decides is impossible to drive. There's a lot left to be done (heck, nothing's finalized yet), but I like the fact that nothing's been hacked too badly yet. Testing like that is weird for me (in the lab programming), because I hear the radio crackle all the time with comments, then I hear the car approach, "mop mop mop mop mop mop mop mop MOP .. MOP .. MOP .. .. .. MOP .. MOP .. MOP .. .. .." you'll know what I mean when you hear our siren. Of course then you'd hear someone in the lab, "frick, why's my program barfing? ... Oh, the car's in range." It's always fun getting rogue sensor data when you're trying to work from log files. Incidentally, we need to check to make sure we have gas available before running the cars so long. I think we may actually have to drive the Tahoe and the Suburban to the gas station to fill up this time. That will be sketchy... just picture the neo-techno vehicles passing poor confused Amish people in horses and buggies on public roads. Can anyone say anachronism?
