Medial Axis Transform assists path planning in configuration spaces with narrow passages

Abstract

Medial Axis Transform (MAT) is a representation that encodes an object with symmetric (medial) axes in the object interior. MAT has been employed in a variety of applications such as pattern recognition, biological shape analysis and robotic motion planning. This work considers the problem of planning robot paths using a Probabilistic Roadmap (PRM) approach when obstacles create "narrow passages" in the robot collision-free space (the free-space). Past studies suggest that the performance of PRM planners degrades considerably in the presence of narrow passages. However, experiments also show that the dilation of the robot's free-space may dramatically alleviate the narrow-passage problem. Here, a novel MAT-based method is presented to "skinnify" the robot and/or the obstacles, hence to dilate the robot's free-space. A new PRM strategy exploits the dilated free-space to efficiently construct paths through narrow passages. Simulation results validate the use of the proposed MAT-based skinnification technique.