The path planning system described above has the goal of reaching a state in which the moving object is in ``free space''. It assumes that there will only be one layer of obstacles between the current location of the moving object, and free space on the other side of the obstacles. In a more complex scene, where the obstacles compose a ``maze'' which the moving object must traverse, a high level strategy is necessary to oversee the planning of paths between layers of obstacles.
The basic methods used by the present path planning system could be retained in this case; once it is possible to plan a path between two objects, it is possible in general to plan a path through many, taking them two at a time (access to a via point between two obstacles might, however, be restricted by a third obstacle).
Complex path planning of this type could be carried out with the aid of a supervisory system, which would be informed of all possible via points between obstacles, and would recommend the order in which to approach them. The supervisor would make use of some search strategy to organise these decisions. A ``hill climbing'' search would be sufficient, if a satisfactory function was provided for evaluating how close the moving object was to leaving the maze. This function would be used to replace the ``free space'' goal of the current system. If the supervisory search strategy involved backtracking, it would be necessary to extend the spatial reasoning system, providing facilities for retracting state changes in the scene representation that had resulted from previous envisionments.