Human-machine shared control: New avenue to dexterous prosthetic hand manipulation

Abstract

At present, albeit the dexterous hand prostheses of multiple degrees of freedom (DOFs) have become prosperous on the market, the user’s demand on intuitively operating these devices have not been well addressed so that their acceptance rate is relatively low. The unintuitive control method and inadequate sensory feedback are frequently cited as the two barriers to the successful application of these dexterous products. Recently, driven by the wave of artificial intelligence (AI), a series of shared control methods have emerged, in which “bodily function” (myoelectric control) and “artificial intelligence” (local autonomy, computer vision, etc.) are tightly integrated, and provided a new conceptual solution for the intuitive operation of dexterous prostheses. In this paper, the background and development trends of this type of methods are described in detail, and the potential development directions and the key technologies that need breakthroughs are indicated. In practice, we instantiate this shared control strategy by proposing a new method combining simultaneous myoelectric control, multi-finger grasp autonomy, and augmented reality (AR) feedback together. This method “divides” the human sophisticated reach-and-grasp task into several subtasks, and then “conquers” them by using different strategies from either human or machine perspective. It is highly expected that the shared control methods with hybrid human-machine intelligence could address the control problem of dexterous prostheses.