Learning Manipulation Tasks from Human Demonstration and 3D Shape Segmentation