CrossRoads:
Post a schematic - I didn't see any mention of diodes across the motors.
Thanks for the suggestion! It's an H-bridge going both ways -- I can't really use diodes there.
The N-channel MOSFETs (high and low -- driven by IRS2301 drivers) have built-in body diodes that go the way they can go.
The controller code grounds out both terminals for 10 microseconds when switching power, to help out. I've verified this on a scope. (Well, a Saleae USB logic analyzer)
As I said, the AVR is on the other end of a LM350 regulator and a bunch of capacitance. This should protect against spikes AFAICT. And the AVR itself is actually fine -- it's the connected programmers that get screwed up, and I'm looking for ways to make that not happen. It seems like a programmer dev tool should be able to be robustified enough to take all kinds of abuse...
Is there a RuggeduISP? 