Finite State Model of Walking Determined by Adaptive Logic Networks

Abstract

We developed a method for determining a finite state model of locomotion that is applicable to real-time control of walking in individuals with paralyzed legs. The finite state model represents walking as a set of If-Then rules. An If-Then rule uses coded sensory information as inputs (If) and levels of electrical activities of muscles as outputs (Then). The model incorporates temporal and spatial synergies between muscle groups based on sensory information. The sensory input includes accelerations of leg and body segments, and ground reaction forces at toe and heel zones of the sole. The output of the rules is generated by detecting the onset of muscle activity from the amplified and rectified recordings of EMG signals from the prime movers of the leg. The coding uses a local threshold technique. Adaptive logic networks (ALNs) were used for estimation of If-Then rules. The training consisted of various samples of walking recorded in healthy individuals. The application of ALNs was optimized for low misclassification error and fast training. The overall performance of ALN (correct responses that would lead to correct stepping) when applied on test data, not used for the training, was >82%. We assumed that 80% is the margin for correct stepping for the walking in hemiplegic individuals