As part of a DARPA program (Revolutionizing Prosthetics), the Center for Intelligent Mechatronics at Vanderbilt University developed an arm prosthesis for transhumeral (or above elbow) amputees. The prosthesis itself has nine actuators an twenty-seven degrees of freedom. The actuators power the arm's elbow and three degree-of-freedom wrist. The other five actuators of the arm drive its hand through the use of tendon actuation. The hand of the prosthesis contains several springs so that, along with the tendon actuators, the hand can easily interact with objects. In other words, the springs allow the hand to transition between force control (while interacting with an object) to position control (while acting in free space), without the need for complicated control algorithms.
My portion of the development of this prosthesis was spread across three main areas. First, because neural interfacing (i.e. determining an amputee user's intent by measuring some biological signals) was not part of the Vanderbilt contingent's work, some way of commanding the arm by a healthy individual was needed. In other words, it was desired to have a human actor carry out some demonstrations and for the arm to ape the motions of the actor. To accomplish this, I constructed an exoskeleton with integrated position sensors that acted to measure the wearer's elbow angular position, as well as the three angular positions of the wrist. The components for the exoskeleton were built up using a stereolithography (or rapid prototyping) machine so that pieces could be quickly and cheaply manufactured and iterated until the human form was matched.
Taking measurements from the hand was somewhat more complicated. A first prototype for this problem was a cloth glove with integrated bend sensors (similar to a CyberGlove, or the Nintendo Power Glove). The measured overall bend of the finger was approximated as the actual prosthesis's pneumatic actuator excursion. Because the way to do this was somewhat ambiguous, the second portion of my work in this project was to develop a three-dimensional visualization of a virtual prosthetic arm. The glove input device was interfaced to this virtual prosthesis to evaluate its characteristics.
Because a major feature of the prosthetic hand is its natural ability to interact with objects, and because this aspect was not intuitive for a glove wearer to demonstrate, a second hand input device was constructed. This second device was a joystick, attached to the distal end of the exoskeleton, that is pressure sensitive. In other words, the joystick incorporates five load cells that sense how firmly the user is grasping the joystick with each finger. By taking the load signals from this device and displaying the resulting motion (or applied force, as the case may be) in the virtual prosthesis, it was found that the joystick device was much more intuitive for someone to use when demonstrating the arm's ability to interact with its environment.
The third major portion of my work with the prosthetic arm was to make the observed action of the virtual prosthetic hand match the action of the actual pneumatic hand. This included development of the low-level tendon actuator force control and a human interfaces aspect that allowed the exoskeleton wearer to demonstrate the full range of the arm's abilities. That is, because the springs in the prosthetic hand are relatively weak, the range of applied pressures that the user must apply to cause full deflection of the prosthetic hand's tendons is very small. This small range meant that the joystick input device was very imprecise. On the other hand, the pressures needed to apply high forces on an object with the prosthetic hand were very large.
To flatten these ranges and make the input device easier to use, a nonlinear mapping function was inserted between the input device and the prosthetic arm. This increases the pressure range that causes full deflection of the prosthetic fingers (while operating in free space) while decreasing the force that must be applied by the user on the joystick to achieve maximum force in the prosthesis.
The operation of the input device was quite successful, with the arm able to demonstrate very fine motions as well as demonstrations with props that require a great deal of strength.
