For the use of mobile robot in a workspace of obstacles, a collision-free path planning is necessary. The global path planner constructs roadmap including all possible collision-free paths from a top view of the environment, then select the optimal path for the robot. In previous works, Roadmap construction algorithm can operate in a polygonal obstacle environments. To build a road feature map with non-polygonal obstacles, we propose the algorithm using modified Kohonen's self-organizing feature map(SOFM) in neural network theory. The key to this technique is a simple modification of learning algorithm for obtaining the Voronoi-like diagram. By this algorithm, fast roadmap construction can be achieved and then fast path planning is possible. In addition, path tracking controller is designed to track the generated path by Lyapunov Techniques. Capabilities of this method are investigated through simulation performed on the unicycle-type mobile robot.