Grasping Partially Occluded Objects Using Autoencoder-Based Point Cloud Inpainting