Route Planning, State-Space Search, and Case-Based Learning

Route planning is the process of determining a route, or path, from some location to another. The route may be described by a line on a map, a sequence of movements (e.g., “Turn right, walk forward, turn left at the next intersection”), or a sequence of way-points (e.g., “Go to the corner of Snelling and Grand”, “Next go to the corner of Grand and Pascal”, “Last go to 1430 Grand Avenue”). Humans are generally very good at figuring out routes from one place to another, whether within a building, around a campus, across a city, or accros the country. We can take into account any number of intangible or qualitative factors in determining a route. In fact, you’ve probably seen or been a part of an argument about what the “best” route is to some goal location. Computer systems that generate route plans are used for providing directions to humans, arranging complicated travel plans (like connections between transit options), or providing routes to control the movement of robots, whether simulated or real. Most of us are familiar with web sites such as Google Maps that can generate driving directions. You may also have used a transit system route planner; I once planned a train trip in Europe, changing trains included, from the comfort of my office! Relevant links: Google Maps http://maps.google.com Metrotransit Trip Planner, Twin Cities http://tips.metc.state.mn.us/mntest/cgi-bin/itin_page_ie.pl Italy’s train travel web site http://www.trenitalia.com/en/index.html This project will explore some AI techniques for generating route plans, focusing on plans made up of way-points. We will start with an informed, heuristic-driven graph search algorithm, A*, for finding a route. Generating a route from scratch can be computationally expensive and wasteful, especially if the same plan will be needed repeatedly. We will, therefore, examine a machine learning approach, case-based learning, which tries to avoid the effort involved in generating routes from scratch.