In order to manipulate objects, robot arms need to approach objects, and therefore need to move to certain positions in a certain orientation. The question solved by path planning is how to approach the goals, i.e., how to move the manipulator's joint angles without colliding with any of the surrounding objects. This article gives an overview of manipulator path planning methods, and shows the benefit of combining neural networks with graph-based techniques. This combination results in an adaptive modeling of free space, together with the connectivity of free space regions being captured in a graph.