Continual Learning for Multimodal Data Fusion of a Soft Gripper