NSF Award
8960206
A six legged parallel link manipulator (called a Stewart Platform) has the ability to perform six degree of freedom manipulation, and is mechanically much simpler than conventional serial mechanisms with similar manipulative capability. Despite their mechanical simplicity, such mechanisms have not been exploited except for low accuracy, low speed applications because of the difficulty in performing real-time control of the highly non-linear, highly coupled mechanism. The control problem is exacerbated by the fact that there is no known closed from solution to the forward kinematics of such platforms. This project will develop two cooperating neural nets which together help control a six legged parallel link manipulator in the high speed, high accuracy domain. Each of the two neural nets is itself composed of tow cascaded neural nets. The proposed system is interesting because it solves a problem for which there is no known acceptable solution, and also because there are many important applications which could be cost effectively implemented with a stewart Platform if the control problem could be overcome. The proposed system is also significant because is cascaded organization of neural nets should greatly reduce the training time required for the system to achieve good results.
