Bioinspired grasp primitives for a dexterous robotic hand to catch and lift a cylinder

摘要

Joint motion profiles from nine human test subjects were recorded while catching and lifting a cylinder. Significant differences between the two tasks presented themselves consistently across all test subjects. In general, initial rapid hyperextensions were observed while catching the cylinder versus simply picking it up. Slightly larger ranges of abduction were also seen for the catch, but, as expected for the cylinder, these were approximately symmetric across the four fingers. From the recorded motions, cubic polynomials were fit to the joint angle data to form two distinct grasp primitives. The joint motions of the test subjects were simulated in Simulink with a skeletal structure of the human hand as well as a model of the Shadow Hand. These primitives were subsequently used with the physical Shadow Hand by a human operator who was able to successfully pick up and catch a cylinder. An electromyogram signal measured from the forearm of the test subject was used as the input to the Shadow Hand controller. After a brief training session with the EMG-controlled hand, the cylinder was successfully caught in seven of ten attempts and successfully grasped and lifted in five of five attempts.