SimPlanner

Oscar Sapena and Eva Onaindía

SimPlanner is an integrated system for on-line planning and replanning in dynamic environments. The planner obtains a sequential plan by computing the next executable action at each step. SimPlanner uses a hill-climbing search variation, focused to obtain very fast solutions at the expense of plan quality.

SimPlanner uses a heuristic function to compute the next action to be executed in the plan. This heuristic, based on the information of the current state and the top-level goals, works in three stages:

  1. It computes a relaxed plan to achieve each top-level goal separately.
  2. The relaxed plans are grouped together in branches according to their initial actions.
  3. Finally, an ordering procedure is applied to branches to determine the first branch that would be executed. The heuristic function returns the first action of the first branch.

The main objective of SimPlanner is to be able to work in dynamic environments with incomplete information. It works jointly with a replanning module in order to quickly repair the plan being executed when an unexpected event is received in the world. The replanner adapts the plan to the new situation by retaining as much as possible of the original plan. This integrated system enables to incorporate additional features like, for instance, modifying the objectives while the plan is being executed. Additionally, during plan execution an optimal planner can be used to refine the remaining plan to be executed.

This first version of SimPlanner only generates sequential plans for simple Strips problems.