Improving search efficiency using possible subgoals

A primary function of most advanced traveler information systems involves the ability to plan optimal routes. Although the route planning ability of ATI systems can be facilitated using centralized computing resources, most ATI systems currently under development use in-vehicle computational resources. The primary advantage of this type of approach is fault tolerance; vehicles can continue to plan routes even in the absence of a centralized computing center. However, there are significant challenges associated with planning routes using in-vehicle electronics. These challenges result from trying to keep the cost of the in-vehicle electronics to a reasonable level. Thus, the route planning algorithm must be both space and time efficient to compensate for the limited amount of memory and computational resources available. This paper describes a new heuristic search algorithm that can be used for in-vehicle route planning. Certain types of heuristic search problems may contain an identifiable subgoal. This subgoal can be used to break a search into two parts, thus reducing search complexity. However, it is often the case that instead of one subgoal, there will be many possible subgoals, not all of which lie on an optimal path to the eventual goal. For this case, the search cannot be simply broken into two parts. However, the possibility of reducing the search complexity still exists. Previously, Chakrabarti et al. [1] developed a search algorithm called Algorithm I which exploits islands to improve search efficiency. An island, as defined by Chakrabarti et al. [1], is a possible subgoal. Previously, this algorithm had only been analyzed theoretically. In this paper, some experimental results comparing Algorithm I to A^* are presented. Algorithm I has also been generalized to cases where more than one possible subgoal can appear on an optimal path. Two new heuristic island search algorithms have been created from this generalization and are shown to provide even further improvement over Algorithms I and A ^*. The use of possible subgoals can make any type of search more efficient, not just A ^*. Modifications are discussed which describe how to incorporate possible subgoal knowledge into IDA ^* search.