Snuffy III came in 8th in Expert class for 2003
Hey, Eighth place doesn't sound so bad! Actually, I could say 8th out of 29 entries....Not so bad. Or, I could say 8th out of 9 robots that ran. Hmmm, not so hot. Or, I could just admit that Snuffy III didn't blow out a single candle.
This year, I had been distracted working with a team on the DARPA Grand Challenge race (www.darpa.mil/grandchallenge) and my software was "finished" just a couple days ahead of the contest and had many bugs left in it as well as situations which were not yet covered (but hopefully wouldn't happen!). However, during Saturday practice it was working fairly reliably with the occasional screw-up due to incomplete software or outright bug. Sunday; on the first run of the competition, the robot just stopped before reaching the candle. I thought "RATS! another software bug", and left things the same for the second trial. When the second trial made a similar stop, I got suspicious that something had gone wrong. Some testing revealed that one of the D12 IR range sensors, which scans the area in front of the robot, had failed to a relatively low reading which indicated there was ALWAYS something in front of the robot...hence the stops, it had nowhere to go. Unfortunately, I usually test the sensor in this short range and it had looked ok.
So, I replaced the sensor and it seemed to work fine. On the third and final trial, as I powered up the robot, I noted that the camera power did not come on. There was no time to fix it, so I ran anyway; but it couldn't find the candle without the camera. The camera (and some other sensors) were mounted on a servo which rotated +/- 90 degrees. While all the wires SEEMED to have plenty of flex in them, the camera power wire was apparently bending around a sharp point on the connector and had weakened. It probably finally broke when we made the change to install the new range sensor.
Lessons learned:
Get the software done plenty early so it can be well evaluated
under actual conditions (Duh!)
(and so you don't mistake hardware problems
for software bugs!)
Test sensors at multiple points to ensure they work over their full range.
If you have moving wires, make sure to clamp them on each end so that the motion does not occur at sharp points (like a connector).
Ah, well....NEXT YEAR!