Robot enters stateless "state" on official FMS

[Thomas DeSilva]

The last two finals that we've been in this season, our robot has had control issues in one of the matches. If you watched the PNW District Championship in Portland last weekend (shoutout to 360 and the event staff for allowing that replay of the second match), you probably saw the symptoms: The robot started the match off fine, but a few seconds into auto, it stopped moving and didn't start again until close to the end of the match. Looking at the driver station logs, I see that the robot was in auto as it should have been, but then it just wasn't. Not disabled, just stateless. Our driver noted that according to the driver station, we still had comms and the code was still running and enabled. After he rebooted the RoboRIO, everything was fine and we played the last few seconds of the match.

We ran into a similar problem in the Philomath finals. The robot made it through autonomous, but a bit into teleop we lost code and the robot drove itself into the low bar (we suspect that the Talons were just running using their last input). Just like in Portland, the robot code had somehow lost its state.

We've been trying to figure out what the problem could be, but we're at a loss. Typically, when the robot code crashes, its handler will restart it fairly quickly; we don't see that happening here, so we don't think it's a run-of-the-mill crash. The logs--from the driver station and our own logging--don't indicate that the program crashed either. We know we didn't get an errant disable signal from the field because we would have seen that state change. Does anyone have any ideas as to what could be happening here? The logs in question are attached.
A few notes:

• The RoboRIO continues running after the code stops and our round trip times stay fairly low.
• The errors in the Portland log that occur near the crash are coincidental. Similar errors show up in most of our logs, and in particular they don't show up before the crash in Philomath.
• Some of the messages in the Portland log got a bit polluted by a stray "cout" that made it into the program.