Johnson Space Center is sponsoring continuing research on the use of myoelectric signals to control dexterous robotic and prosthetic hands. This research is expected to advance the state of the art beyond that of today's most common, commercially available, myoelectrically controlled prosthetics - clawlike devices, little altered since the 1970s, that are restricted to motion in one degree of freedom and are basically motorized hooks. Thus far, the research has shown that by use of a suitable combination of electronic hardware and software, it is possible to recognize the myoelectric signatures of at least two different grips in real time, with an accuracy of at least 90 percent. While this level of performance is below the nearly 100-percent accuracy required for highly precise teleoperation (e.g., for controlling remote manipulators on the Space Station), it is high enough to offer potential benefits to the prosthetics community, and hence would be well worth further investigation. The fruits of this research might eventually include comfortable, lightweight, and relatively inexpensive prosthetic hands that could act, in a nearly lifelike manner and in real time, in response to the myoelectric signatures of as many as six different grips.
Experts in prosthetics generally agree that the following five types of grips are vital to a person's daily activities:
- The three-jaw chuck or pincher grip, used to hold small objects;
- The lateral or key grip, used to hold and actuate a key in a lock;
- The hook grip, used to carry an item like a book or a briefcase;
- The spherical grip, in which the thumb and fingers are wrapped around a sphere; and
- The cylindrical grip, in which the thumb and fingers are wrapped around a cylinder.
The designs of majority of commercially available electrical hand prostheses generally address the need for only a couple of these types of grips, and do not provide for independent control of fingers and thumbs. Most are capable of chuck grips only. Designs of the most advanced research models have provided for chuck and key grips, with options for spherical and cylindrical grips provided through passive compliance of fingers. All of these devices, at whatever level of sophistication, present control and interface problems.
One of the problems is caused by the fact that available myoelectrically controlled hands are limited to one-degree-of-freedom performance, although some offer proportional gripping forces as well. Worse yet, the most complex artificial hands are significantly larger and heavier than are human hands, and therefore amputees experience difficulty in controlling them. Indeed, the human hand is a complex mechanism that has 27 degrees of freedom and is capable of a wide range of powerful and yet delicate movements with quick, quiet, and smooth operation while exhibiting both high performance and efficiency. Therefore, it has long been thought that, given constraints on size and weight, it is well nigh impossible to design an artificial hand that performs at the level of the human hand. However, as has now been shown by the research reported here, it may be possible to construct a prosthetic hand that has multiple degrees of freedom and that performs the key motions of a human hand under control by processed myoelectric signals.
In this research, attempts were made to recognize, in real time, the myoelectric signatures of as many as six different grips by use of frequency (and, alternatively, time-frequency) techniques that "split" the signatures and by use of artificial neural networks to recognize the signatures. The development of an ability to distinguish among different grips would be a critical extension of the state of the art, inasmuch as commercially available myoelectrically controlled prostheses recognize only two usual states, i.e., "grip" and "open." In the research, various degrees of accuracy of recognition were obtained for various combinations of grips. The best results were obtained (accuracy >90 percent) in distinguishing between the chuck and key grips. Inasmuch as it was heretofore generally believed by experts in the field that it was not possible to extract grip-signature information from myoelectric signals, this is a significant achievement.
This work was done by I. D. Walker of Rice University forJohnson Space Center.