Integration of hardware and optimization control for robotic and prosthetic systems

With medical advances through the past half century, survival rates following trauma have risen. Along with this rise has come an increase in the number of survivors with amputated limbs. Many of these survivors are Soldiers, Airmen, and Marines, who are relatively young and could benefit from sophisticated prostheses to replace the lost function.;These prostheses would be very maneuverable and able to better mimic the natural human motions. Such devices would likely be high degree of freedom with many actuators. Control of prostheses with intelligent algorithms may provide improved performance for the user.;To this end, a novel experimental transmission for driving the several joints of such a device has been developed and tested. Also, it has been used in the design, production, and testing of a 3 DOF digit actuator for use in a prosthetic hand.;Embedding the control hardware would make such a prosthesis more compact and portable. Using custom printed circuit boards, Microchip PIC microcontrollers have been used to control the digit actuator. Taking advantage of surface mount packages, control boards have been developed which integrate motor drivers with microcontrollers, and fit into a space comparable to that of the aforementioned prosthesis. Furthermore, the networking capability of these controllers has been demonstrated, presenting an extensible framework for addition of processing power as technology develops.;Given the non-linear nature of the several joints in the system, intelligent controls have been explored. Model reference adaptive control (MRAC) was used in simulation of digit models. Also, coupling MRAC with artificial neural networks yields ANN-MRAC (artificial neural network model reference adaptive control). Training these ANN control systems using ALOPEX yields good tracking performance across the non-linear range of the system. Such control logic may prove effective in a time varying non-linear system such as a hand prosthesis.;Human machine interfacing is key in the use of prostheses. Since a minimal amount of training is most desirable for the user, adaptive and intelligent methods may provide a control interface framework that reduces learning time for the user. To accomplish this, an algorithm for optimization of large dimensionality sensor grids was developed. This algorithm uses several template matrices to optimize the gain of each sensor in the grid. This both identifies a region of activity, and reduces the signal-to-noise ration of the sensor grid output by reducing gain on channels not containing information. The desired region is identified through enhancement of the signal gain on the sensors above the region. This would allow the placement of sensors on the body in an inexact fashion and instead let the computer optimize the sensor network gains for regions of activity associated with a given motion. Such an adaptive system reduces learning time for the user, thus reducing human error and easing use.