Controlling anthropomorphic robot hands with multiple joints is challenging due to their high degrees of freedom, which makes it difficult to provide effective demonstrations for dexterous manipulation. Moreover, existing approaches often require large training datasets or suffer from computational delays due to intensive calculations. This paper presents a study on imitation learning for the control of a prosthetic robot hand with 21 degrees of freedom, including wrist joints. By adopting a nearest-neighbor-based algorithm, the proposed control system achieves efficient learning with a small amount of data and provides an intuitive demonstration interface. Experimental results for three different tasks using the robot hand are also presented.
Sun et al. (2026) studied this question.