| This dissertation deals with the formulation, analysis and implementation of control systems for intelligent mechanical machines. These machines must operate safely under uncertain conditions without external supervision, and must determine and achieve, through adaptation and learning, the task that optimizes a prescribed performance criterion. The primary application described in this dissertation is an intelligent exercise machine which is safe to operate and determines the optimal exercise routine based on the a-priori unknown strength characteristics of its user.; In the self-optimizing control problem a performance criterion, which depends on the machine behavior as well as on other unknown parameters, is to be optimized. Thus, unlike standard adaptive control applications where the desired behavior is specified a-priori, it is necessary to explicitly determine the optimal behavior as part of the adaptation process, and to control the machine so that it behaves optimally. The proposed solution consists of an adaptive controller and a reference generator in tandem. The adaptive controller is capable of tracking arbitrary behaviors and the reference generator commands the control system to alternately follow either a training behavior or the estimated optimal behavior. The reference generator switches between these two types of behaviors by monitoring an internally generated signal. It is shown that, after a finite number of switchings, the optimal behavior is executed arbitrarily closely.; The autonomous behavior of a mechanical system is encoded by means of a velocity field in this dissertation. A dynamic feedback controller, which tracks the prescribed velocity field and maintains a passivity relationship between the controlled machine and its physical environment, is derived. The proposed dynamic controller mimics a flywheel in that it stores and releases energy, but does not generate it. The unforced response of the closed loop system converges to a scaled multiple of the prescribed velocity field. The robustness of the feedback system to environment forces is analyzed. Several application examples are given, including a solution to the contour following problem.; The intelligent exercise machine application is developed based on the self-optimizing control and passive velocity field control results. Experimental results verify that the exercise machine indeed optimizes the user's workout. |
