I. Introduction
The motion planning problem consists of finding a valid (collision-free) path from a start state to a goal state. One solution to this problem is to define a roadmap that captures the topology of the collision-free portion of configuration space. However, the complexity of the workspace and robot can make this process challenging. Probabilistic Roadmap Methods (PRMs) have addressed this challenge by constructing a roadmap of randomly sampled robot configurations and testing each configuration for collision [20]. Connections are made between two samples when a collision-free transition can be made. These samples (vertices) and connections (edges) define a roadmap that the robot can safely traverse. Recently, PRMs have been extended to be adaptable [26] [4]. These new methods can deform paths [18], update roadmaps due to moving obstacles [15] [27], map both collision and collision-free states [9], and deal with uncertainty in the motion model [6] [2] [1] [25]. However, despite all these advances, PRMs require that the model of the problem must be accurate, e.g., there must be a clear delineation between collision and collision-free states. Error-prone collision detection can lead to erroneous roadmaps which produce feasible paths in the modeled environment but lead to collisions in the actual world.
Whole arm manipulator (WAM) feeling an obstacle boundary.