[mjcoss]

I think that one of the biggest stumbling block for development of autonomous code is time. Programmers are usually the last folks to get their hands on the robot. While we can code things up, we must wait for the robot build process to get to a sufficiently stable point that we can grab the robot to run tests.

If the autonomous code is just a hand crafted set of instructions, then it must be tested, and refined many times to get it working properly. If the autonomous code is more of a state machine with transitions triggered by sensors, then this to must be laboriously tested. If it's an even more elaborate learning system, then the testing is even more intensive.

Bottom line is that there typically just isn't the time to do the needed testing. So some times just code up something simple, others never get around to putting anything into autonomous mode.

As far as the original idea of a fully autonomous robot, while I applaud the ambition, I suspect that it will be unrealistic for most, if not all, teams to be able to do that, given the empirical evidence of autonomous performance to date for most teams.