Book Description

This report describes studies of the feasibility of designing adaptive systems that have adaptive ability distributed among their component elements. Each element is capable of adjusting its own behavior and each accepts a small portion of the adaptive responsibility. Adaptive systems built from these elements would display the flexibility, reliability, and damage recovery capacity found in living creatures. Analytical techniques using optimal control theory and digital and analog computer simulations were used to further the study of distributed adaptation. Trainable networks of threshold logic units, which had been proposed as modules of an adaptive controller, were analyzed and simulated on a digital computer. A trainable universal Boolean function generator was found which has guaranteed rapid convergence to any desired Boolean function. An adaptive optimal autopilot for a roll-yaw coupled, high-performance aircraft was simulated on the analog computer. An evaluation of test results provided detailed information about the convergence of the adaptation process and the effect that adaptation has on system performance. The optimal adaptive controller adapted rapidly enough to prevent violent divergence of the aircraft attitude. The results suggest that these concepts are suitable for further development. Techniques developed to instrument the adaptation process would be useful for the construction of large-scale adaptive systems. (Author).