An Algorithm for Planning `Sensible' Routes

The problem of nding an optimal route between two points in a space strewn with obstacles has been extensively studied in robotics. The standard approach to such problems is rst to reduce the original geometrical speciication to a nite graph of possible routes, and then to use graph-searching techniques together with the metrical information available in the original geometrical speciication to select the best of these routes. The algorithm presented here, by contrast, employs a ``rst-past-the-post' approach to route planning. Speciically, we show that the algorithm is guaranteed to produce a route satisfying a certain criterion of`sensible-ness', because routes failing this criterion are always slower to get constructed. The approach taken here diiers in three respects from most current work on route planning: (i) it nds sensible, but not necessarily optimal, routes; (ii) it is uninnuenced by perspective (though it is limited by occlusion problems); (iii) it produces its output in a qualitative form, which is nevertheless well-suited to guide the appropriate route-following behaviour. 1 Background The problem of nding an optimal route between two points in a space strewn with obstacles has been extensively studied in robotics. As the problem is normally posed, the set of obstacles cluttering space, the dimensions and degrees of freedom of the moving robot, and the start and goal points, are all speciied in some numerical coordinate system. In the special case of planning routes for a point object moving in a plane with polygonal obstacles, this speciication usually amounts to a Cartesian coordinate description of the scene in plan view, with the output being given in terms of the coordinates of line segments deening the optimal route. The standard approach to such problems is rst to reduce the original geometrical speciication to a nite graph of possible routes, and then to use graph-searching techniques together with the metrical information