This paper describes the work for a map-guided robot (AGV) maneuvering in dynamic factory floor environment. The robot is equipped with a narrow-beam sonar to detect obstacles. In factory applications, the location and orientation of the robot will be determined by external sensors. A path from start to target location will be digitized into a sequence of waypoints and the robot navigates locally by dead-reckoning between two adjacent waypoints. Autonomous navigation, in this setup, is viewed in terms of three components: automatic path plannihg for the given floor layout, the start and target locations of the robot, automatic path replanning after detecting obstacles, and local navigation which should always lead the robot out of a trap (cul-de-sac) if one occurs. For any given floor layout, we model the passage ways between obstacles (processing machines) as a connected graph knowns as the Voronoi graph. From the robot's point of view, the graph is the map to plan paths and navigate. Path planning is based on the retraction method. After detecting an obstacle, path replanning will be invoked. Algorithms for path planning, path replanning, and local navigation are given.