The rules are relatively simple most of the time but have very complicated edge cases. The key thing is the AI likes to target enemies who are near to them, breaking ties towards lowest initiative. And it will never trigger a trap if there’s another route to where it wants to be (however long and even if it means it can’t move this turn).
The basis of exploiting the AI involves limiting the number of options it has so it is ineffective (usually by failing to get within range for its attack). Traps are most useful to prevent movement, rather than to actually be stepped on. This is one reason why ranged attackers are much more dangerous: it’s really hard to convince them not to attack you.