Some Experimental Results of Applying Heuristic Search to Route Finding

Search is often used to find shortest routes within real road networks. Standard search algorithms are memory intensive and inefficient for stand alone in-car systems. In this paper, we evaluate a number of heuristic search strategies in terms of space and time complexity for the route finding problem. Route Finding and Search Much research in recent years has focussed on developing technologies to enable improvements in the efficiency of road transport systems. Programs geared toward this goal include the TravTek (Rillings & Krage 1992) and Advance (Catling 1992) programs in United States and SOCRATES (Catling & Harris 1995) in Europe. With each of these programs emphasis is placed on providing traveler information including route finding capabilities. With current technology and infrastructure, many current and proposed systems are of the decentralized autonomous variety, with processing and route map information stored on a local CD (Collier & Weiland 1994). Thus with high hardware costs, it is advantageous to keep the memory, computation, and storage device access to a minimum while keeping the accuracy of computed routes high (Dillenburg & Nelson 1995). Heuristic search offers the potential to reduce time and memory use in route finding systems by incorporating knowledge of the properties of the problem space. We evaluate heuristics for route finding and a number of strategies with potential to increase efficiency over standard heuristic search. Uniform Cost Heuristic Search Standard uniform cost search approaches, such as Dijkstra’s method (1959), suffer the fate of combinato*Data and computing resources kindly provided by the Spatial Analysis Facility, University of Auckland. tSupported by the University of Auckland Research Committee under grant A18/XXXXX/62090/F3414065. 394 Pearson rial explosion; the time and space complexity for such methods are of the order O(bd), where b is the branching factor and d is the depth of the search tree (Korf 1996). The A* algorithm (Hart & Nilsson 1968) corporates heuristic search into uniform cost search, finding optimal solutions yet decreasing the effective branching factor. A number of heuristics which exploit the geometric nature of roading networks are applicable to the route finding problem and can be incorporated into A*. We evaluated the air distance and Manhattan distance heuristics (the L2 and L1 metrics, respectively) in the Auckland road network and compared the decrease in node expansions over an uninformed search. Figure 1 depicts the node expansions created by the three heuristics for a shortest path between a location represented by the large dot on the left to a destination given by the large dot on the right.