I implemented my first TD off-policy method for discrete states and discrete actions (increase or decrease to last executed action) on a patient in real time. Now I am extending my problem from one goal state to two goal states. First goal state is always achievable but I want that if my agent is in goal state 1, it should try to reach goal state 2 if it is available and can be reachable. I am confused that which will be terminal state.
I just want your recommendation that how to tackle these two goal states. In my problem, goal state 1 means 90 percent of the job done while goal state 2 means more than 90 percent. I just want some sort of optimization at goal state 1 to reach goal state 2 if reachable. Any input about the above problem would be much appreciated. Thanks