This paper proposes a Continual Learning (CL) algorithm capable of continuously and incrementally acquiring new knowledge from the environment while maintaining previously learned information. Specifically, it is designed to incrementally learn from various data modalities (e.g., tactile and visual data) and leverages both class incremental learning and domain incremental learning scenarios in artificial environments with insufficient labels but abundant Independent and Identical Distribution (NIID) unlabeled data. The algorithm enhances efficiency by storing only prototypes for each class. We evaluate its performance using a custom multimodal dataset consisting of tactile data from a soft pneumatic gripper and visual data of objects extracted from still images, as well as the Core50 dataset. Furthermore, we validate the robustness of the algorithm through real-time object classification experiments using the ROS framework.
