A simple path optimization method for motion planning

Most algorithms in probabilistic sampling-based path planning compute collision-free paths made of straight line segments lying in the configuration space. Due to the randomness of sampling, the paths make detours that need to be optimized. The contribution of this paper is to propose a gradient-based algorithm that transforms a polygonal collision-free path into a shorter one, while both: requiring mainly collision checking, and few time-consuming obstacle distance computation, constraining only part of the configuration variables that may cause a collision, and not the entire configurations, and reducing parasite motions that are not useful for the problem resolution. The algorithm is simple and requires few parameter tuning. Experimental results include navigation and manipulation tasks, e.g. an robotic arm manipulating throw a window and a PR2 robot working in a kitchen environment, and comparisons with a random shortcut optimizer