Unfortuantly I cannot help you with the code, but my observation is that AI orders seem to updated, or generated, everytime it reaches a tile. This seems to be the reason behind the AI running in circles and wasting all APs for moving for and back between two adjecting tiles.

You can see this quite frequently when an enemy soldiers is standing behind a row of windows. He will often switch for and back between two windows until having spent all APs. Another very typical situation is an enemy who already has been shooting (means he can see his foe) but then suddenly advances a few tiles only to then run back beyond the tile he had started his movement from.

My guess would be that these orders are not executed (or generated?) en bloc, in the way of "move to tile 114 112 and then take cover", but step by step: "Take cover by moving to tile 114 112" and when the AI has reached this tile a new order is generated, which can include to move back to the tile it just came from.