Awesome. I was a little confused by what you said when they got stuck. The algorithm knows they're stuck, but just doesn't do anything about it yet? Is that what you mean?
Yeah. Part of the way it moves the entity knows exactly how far away "where they are supposed to be" is. So those little guys that were stuck, every time they tried to move they already knew they were many blocks away from where they should have been if they'd followed the ideal path.
A real AI would use this to decide which actions to plan next. Simply finding another path may not be the right answer. For example, if they were trying to catch some target but then got stuck and that other target is now in a place they can't get to easily then they might pick a different strategy.
It would have been pretty simple for me to just "repath" them when they got stuck for the demo but I wanted to emphasize that I hadn't added any intelligence to the adaptive path finding. It's using a fairly simple and extremely dumb approach that just happens to work pretty well.