Older blog entries for Nic (starting at number 162)

June 6th, 2005

Rejected by DARPA. Again.

5 Jun 2005 (updated 13 Aug 2009 at 18:45 UTC) »

June 5th, 2005

I made some adjustments to some software and tested. While pushing the Xpeditor down the street it successfully detected and avoided several cars parked along side the road and made it half way around the neighborhood. At that point, accompanied by a flash of lighting and an enormous clap of thunder, it started hailing extremely hard so I had to abort my test. By the time I got back to my garage the street was under water. That's what I get for living on Washington. The neural networks did output some false obstacles and didn't do the best job detecting the real ones, but with only 30 seconds or so of video to train on they did pretty well.

5 Jun 2005 (updated 13 Aug 2009 at 18:46 UTC) »

June 4th, 2005

I trained a neural network to avoid cars parked along the side of my road so the Xpeditor will be able to autonomously drive all the way around the 1/3 mile loop. I'll test it tomorrow.

June 3rd, 2005

I worked a bit more on the heading-from-video algorithm, did more testing, and worked on the wiring a little.

5 Jun 2005 (updated 13 Aug 2009 at 18:46 UTC) »

June 2nd, 2005 On the DARPA Grand Challenge Discussion Forum user "mdrumbeat" made a somewhat researched guess on teams that will make it to the NQE. He posts:

I've tallied the following so far: A) About 14 teams with flawless runs (see seperate thread on site visit results) B) About 6 teams with ok runs but with a few flaws like hitting cans or going out of bounds once or twice C) About 15 other teams reporting with failed runs D) About 5 other teams reporting very bad visits or dropped out

By my best estimate, I am guessing that maybe 5 other teams had flawless runs (bringing that total to 19) and maybe 5 other teams had ok runs (bringing that total to 11). If we assume that DARPA will invite those 30 teams with flawless or ok runs, that leaves roughly 10 teams that had failed visits that could be invited. I would venture to guess that it is the last 10 or so teams that require some time to review and trade off based on technology promise and media worthiness.

This is of course a total guess but that would make the break down something like this:

1) 20 teams invited who had successful visits 2) 10 teams invited who had ok visits but with a few flaws 3) 10 teams invited who had bad visits but have promise in terms of technology or media worthiness

That's my crack at a guess.

If that turns out to be the case I'd say The Prodigies (http://www.craterfish.org/teamprodigies) have a pretty good chance of making the cut - our site visit wasn't a complete failure and we certainly show promise in technology as well as media worthiness.

Today is my brother William's 11th birthday. I got my money back for the accelerometer, also.

5 Jun 2005 (updated 13 Aug 2009 at 18:47 UTC) »

June 1st, 2005

I tested some more by pushing the Xpeditor down the street, and made some modifications to the software. It's getting there. DARPA will announce semifinalists on Monday (June 6th) instead of today. Today is also my sister Kathleen's 13th birthday. Another thing I did was add some news from Starwars.com about the Discovery show to the news page.

30 May 2005 (updated 13 Aug 2009 at 18:47 UTC) »

May 30th, 2005

I found a flaw in the heading-from-video derivation method, and whereas it worked very well before, it now works extremely well. I uploaded the software on the Xpeditor, and it drove well. If only I had spent my week before the site visit working on heading from video.

One of the guys from Evergreen Films suggested a way to keep mud and dust from blocking the camera image, which was to spin a piece of plexiglass in front of the camera. He even gave me the plexiglass. After grinding the plexiglass down into a circle I attatched it to a motor, built a "frame", and added the camera. With the plexiglass disc spinning at several thousand RPM, any dust or dirt that comes in contact with it will be deflected. I posted a picture under progress at http://www.craterfish.org/teamprodigies.

May 29th, 2005

I incorperated my video-heading method into the Xpeditor's software, and it seemed to work beautifully. I took a break from testing, until it was dark out however, thinking I could continue by street light, but it's not bright enough to get heading changes from the camera image, so I don't have the full results of the experiment. No school tomorrow, so I can continue testing first thing in the morning.

May 24th, 2005

After school I wrote a program that keeps track of a camera's heading by tracking twenty-five points of high brightness or contrast on the camera image. It worked well, but only at slow speeds. I thought of a better method that I will try out. This would only be used as a back-up heading determiner, but it seems it would be able to keep track of heading precisely around several corners while still remaining fairly accurate.

UPDATE: I wrote the same program using a different method in about two minutes. The new method works much better and at high speeds.

May 23rd, 2005

I'm looking at the MIDG II unit in addition to the Trimble GPS. Neither company has contacted me in response to messages I have left.

I experimented with heading determination based on light patterns, and my very basic experiment worked pretty well, but in order for it to be reliable it would have to be much more complex. Maybe I can work something out using the camera image.

153 older entries...

X
Share this page