Path Planning for Robotic Manipulators via Bubbles of Free Configuration Space: Evolutionary Approach

In this paper, a planning algorithm for computing collision-free paths for robotic manipulators is described. The algorithm is based on incremental growth of tree-like structures that are formed by so-called bubbles, which represent volumes of free configuration space easily computed using the distance information in the workspace. The growth of the trees is guided by a criterion function that reflects several aspects including exploration of the configuration space, distance from the goal, size of the bubble, etc. The problem of learning, i.e., assigning the weights to specific criteria is tackled using evolutionary algorithm. Learning itself is aimed at minimizing the number of distance computations that represent the most of the computational burden. The performance of the resulting planer is validated within several scenarios that include 3-dof and 6-dof robotic manipulators.