"Classical AI planning is concerned with the generation of plans to achieve a set of pre-defined goals in situations where most relevant conditions in the outside world are known, and where the plan's success is not affected by changes in the outside world" Qiang Yang.
In Classical Planning the initial situation is known; all the effects of the actions are known; and the world does not change or the changes are negligible. A goal is a condition on final states. This meansthat, Classical Planning ca be applied only to domains which are not dynamic, unpredictable and in real-time.
Classical Planning is concerned with generation of plans. The planning algorithms can be categorized in t
otal-order and partial-order depending on the ordering relation among plan actions, and can use two different inference method that can be described as backward chaining ( working from the goal to the initial state) and forward chaining (working from the initial to the goal state).
Plan generation can be based on different search strategy:
- state space search
- partial plan space search
- theorem proving
- case memory search
Each ofthem use search algorithms,and they canbe classified in:
- systematic search: starts from an empty solution and systematically explores the space of solutions.
- and local search algorithms: starts from a randomly generated solution and proceeds by "repairing" the conflicts and errors of the soluion itself.
Fromclassicalplanning,new approacheshave emerged:
- non-STRIPS style planning
- reactive planning
- deferred planning
- conditional planning
- case-base planning
- multi-agent planning
- Model checking based planning
