The problem is the fact that beyond rudimentary linkages, first principles is pretty difficult to model without additional complexities or time being spent on p

The problem is the fact that beyond rudimentary linkages, first principles is pretty difficult to model without additional complexities or time being spent on perfecting the state equation. That's where RL or whatever else is new on the block come in to save time & needing reduced expertise. By this rational you could also throw PID at many problems & claim its faster for a basic task.