A robust, qualitative method for robot spatial learning

摘要: We present a qualitative method for mobile robot to explore an unknown environment and learn map, which can be robust in the face of various possible errors real world. Procedural knowledge movement, topological model structure environment, metrical information geometrical accuracy are separately represented our method, whereas traditional methods describe mainly by information. The consists distinctive places local travel edges linking nearby places. A place is defined as maximum some measure distinctiveness appropriate its immediate neighborhood, found hill-climbing search. Local terms control strategies required travel. How find follow procedural learns dynamically during exploration stage guides navigation stage. An accurate created edges, allows accumulated with reduced vulnerability errors. working simulation robot, NX, range sensors explores variety 2-D environments we give successful results under varying levels random sensor error.