Action result analysis

For each action we build a Markov model, describing state transition probabilities. From such model we can tell:
- if given action is deterministinc in result (only one posible outcome for every state)
- if there are any strong abstraction calsses definable over outcome states (ex. some of state attributes changes deterministically)

Such knowledge can be used in replanning or local-area planning where an agent is in close range to goal and more precise action selection is required.