Laura Sebastia and Eva Onindia

STeLLa v2.0 is a forward search planner developed at the Technical University of Valencia. This version uses a new problem-solving approach that consists of computing the set of subgoals to be achieved at each time step. Solving a planning problem can be stated as successively decomposing the problem into a set of intermediate goals. STeLLa's goal is then to find the (parallel) actions to reach those intermediate goals. That is, STeLLa obtains parallel plans and tries to minimize the overall number of actions in the plan.

In each iteration, STeLLa uses the landmarks graph (LG) corresponding to the current state and the top level goals in order to build the following intermediate goal set. A landmark is defined as a literal that must be true in every solution plan. The LG is obtained by extracting a set of landmarks and ordering them under the concept of "reasonable order". Once the LG has been built, STeLLa computes the following set of literals to be solved. This set is called fringe and its literals must be reachable after applying a single action in the current state. If this condition is not satisfied, a regression step is performed in order to approach the current fringe to the current state.

Under this new approach, efforts are simply concentrated in obtaining the literals in the next intermediate goal set and then finding the set of actions to reach those literals from the previous state, easing the resolution of the planning problem.