Development of an intent interpretation system for a minimal anthropomorphic prosthetic hand

Abstract

Upper-limb amputation brings about its challenges, including loss of identity as well as independence. Several prosthetic hands have been developed to aid amputees with regaining normal function following their traumatic experience, however, acceptance of such foreign body extensions is challenging. This is especially true when the chosen artificial limb poses a number of operational limitations whilst feeling unnatural. This results in a reduction of the overall experience whilst posing the risk of device rejection. In order to satisfy the end user with an innovative intent interpretation system to be used with an anthropomorphic prosthetic hand, the ultimate objective of this research project was to carry out a systematic exercise to seek a practical solution for the framework which best addresses a trade-off problem between simplicity, dexterity and usability. This problem was mainly addressed by investigating surface electromyography signal acquisition methods in relation to isometric and anisometric contractions which led to the choice of using transient signals for motion identification and steady-state signals for force estimation. The intent interpretation framework was also developed with the ability of reliably detecting movement phases from no movement phases, only triggering a classifier to make a motion prediction with true movement detection. A classifier was designed to distinguish between four specifically chosen motions from a rest or neutral state, providing the amputee with the capability of performing a large percentage of activities of daily living. Most importantly, the system was set to recognise human instructions in the most simple and intuitive manner to humans. The intent interpretation framework was designed to be user-specific and five normal limbed subjects and two transradial amputees performed offline, episodic experiments which resulted in successful statistical results on the movement detection and identification performance of the system. Successful inter-session performance, as well as force estimation performance results were also obtained through further episodic, continuous and force estimation tests performed by a smaller subject pool. A successful maximised trade-off between the three important attributes was achieved through this framework, which is capable of exploiting the natural human sense of intuition, whilst still providing the end-user with the capability of performing a vast majority of daily tasks in a reliable and practical way.