Is there an RL method which in it the next state depends on the "current action" and "current state" AND an "External Input"?
3
-
$\begingroup$ Why the external input cannot be part of the current state? Could you give an example? $\endgroup$– ConstantinosAug 23, 2021 at 0:49
-
$\begingroup$ because the number of external inputs are too many and we need high resolution of them, so if we consider external inputs in our current state we will have too many states. For example consider external inputs to be several time sequential input which vary between -10:10 with resolution 1e-6. $\endgroup$– Hamed MolaeiAug 23, 2021 at 13:29
-
$\begingroup$ Your problem formulation then is not a MDP. If the external input is considered in the decision making process then it has to be included in the state. If you give a concrete example of your problem it will be easier for someone to help. The question is very vague and would benefit from an example that clearly indicates what is to be optimized, states, actions etc and what actually you want to do. $\endgroup$– ConstantinosAug 26, 2021 at 4:10
Add a comment
|