Just as we experienced last year in Acapulco, the American Association for Artificial Intelligence (AAAI) conference is a time for very little sleep, very good food, and more coding in a 5-day period than should be humanly possible. My body clock feels like I’m running on Mars time or something. Speaking of which, there’s someone here selling Mars time alarm clocks. Wacky. At any rate, I’m stuck somewhere in between exhaustion, insanity, and enthusiasm. Good times.
At this point, we’ve finished the entire preliminary match, and one of our three final rounds. We weren’t supposed to start until 11 today, but we were ready and another team wasn’t, so we’ve been moved up in the schedule. Given our state, this is a good thing be several metrics: we get to watch other runs (you can’t watch if you haven’t gone yet for that arena configuration), we get done earlier in the day, and I’ve always been a fan of getting things out of the way. It’s the best way to avoid stage fright or butterflies or whatever.
Our first three rounds went quite well; we’re scoring consistently. We found three victims in the first round (two with tags, which is worth a bonus) for a total of 5.25 points, after a penalty for moving an arena wall when trying to squeeze through a tight gap. We found four victims in the second round, but no tags. We got 8.44 points because two of the victims were in the orange arena, which is generally more difficult, and therefore worth more points than finding someone in yellow. The last round we found three victims in yellow, but we didn’t get any tags, so we only scored 4.88 points. We did have better mapping in the last round (read: not hand-drawn by me in pencil).
The NIST (National Institute of Standards and Technology) guys who prepare the arena for “earthquakes” were particularly devious with the placement of tags. They were putting tags behind blankets on the floor next to the victim, and that sort of thing. Short of running the victim over, I think the only way you could read the numbers on the tags would have been to have some sort of multi-jointed arm to move the camera into position. That, or have a much taller robot, which wouldn’t be able to fit through much of the rest of the arena. I think Bruce complained about this setup decision. While it’s good to encourage mobility by awarding big points to teams who can find and read tags, there’s not much point in doing it when pretty much no one has that capability.
One thing that hurt us relative to the other teams was the best-2-out-of-3 aspect of the preliminary match. We scored in every one of our rounds, but the team behind us had a zero-point round (due to penalties), so we lost more points in terms of the final round score. The guys from PARC have a pair of really neat reconfigurable snake robots, but they tend to run over important parts of the victim’s anatomy.
Our first final round, which was this morning, didn’t go nearly as well. I only found one victim in yellow, although I was able to read its tag. I had actually found another victim entirely by sound, but since I couldn’t see it well enough in the dark, I didn’t call it as an identification. I think I was worried too much about getting a penalty for a false positive. As it turns out, there was definitely a point where I looked right at it, but since the baby was under a bed and in the dark, there wasn’t much to see. I’m still fairly happy with my performance in the round, since I was able to cover most of the yellow area of the arena, and since I used both robots to take different paths.