Automatic induction of an expert system for a finite state model of locomotion

Date of Award




Degree Name

Doctor of Philosophy (Ph.D.)

First Committee Member

Dejan Popovic - Committee Chair


The dissertation's contribution is in developing and testing a new method of non-parametric identification of multi-input-multi-output (MIMO) systems. In this study an automatic method for obtaining production rules from a set of examples is described. The method is based on minimization of entropy. The rule base was automatically induced using external or afferent sensory signals (input) and EMG patterns (output). A production rule is able to estimate muscle activity patterns using sensory information. In other words, the problem addressed in this dissertation is how to construct a rule base that will map sensory space into muscle timing.Data recorded from six able-bodied individuals during ground level walking, with and without ankle-foot orthoses, was used for rule base design. Rule bases performance was also evaluated during stair climbing, and walking at different speeds.Since the initial rule estimates only muscle timing, in this dissertation two possible extensions are suggested to allow estimation of the muscle activation level. Dividing muscle activity into several predetermined levels and attempting to estimate quantified levels using rule based or decision tree methods can lead to complicated rules and large decision trees. The two methods suggested in the dissertation are capable of estimating the level of muscle activity without a significant increase in the base's complexity. The first method uses a rule base with discrete muscle activity levels. The second method uses a combination of a rule base and a linear estimator.


Engineering, Biomedical; Engineering, Electronics and Electrical

Link to Full Text


Link to Full Text