A powered ankle-foot orthosis (AFO) can be very useful for people with neuromuscular injury. Control of powered AFOs will be more efficient to provide assistance to individuals with lower limb muscle impairments if we can identify different gait events during walking. A walking or gait cycle can be divided into multiple phases and sub-phases by proper gait event detection, and these phases/subphases are associated with one of the three main functional tasks during the gait cycle:​ loading response, forward propulsion, and limb advancement. The gait cycle of one limb can also be characterized by examining the limb’s behavior over one stride, which can be quantified as 0% to 100% of a gait cycle (GC). One easy approach to identify gait events is by checking whether sensor signals go above/below a predetermined threshold. By estimation of a walker’s instantaneous state, as represented by a specific percentage of the gait cycle (from states 0 to 100, which correlate with 0% to 100% GC), we can efficiently detect the various gait events more accurately. Our Human Dynamics and Controls Laboratory previously developed the portable pneumatically powered ankle-foot orthosis (PPAFO), which was capable of providing torque in both plantarflexion and dorsiflexion directions at the ankle. There were three types of sensor attached with the PPAFO (two force sensitive resistors and an angle sensor). In this dissertation, three aspects of effective control strategies for the PPAFO have been proposed. In the first study, two improved and reliable state estimators (Modified Fractional Time (MFT) and Artificial Neural Network (ANN)) were proposed for identifying when the limb with the PPAFO was at a certain percentage of the gait cycle. A correct estimation of percentage of gait cycle will assist with detecting specific gait events more accurately. The performance of new estimators was compared to a previously developed Fractional Time state estimation technique. To control a powered AFO using these estimators, however, detection of proper actuation timing is necessary. In the second study, a supervised learning algorithm to classify the appropriate start timing for plantarflexor actuation was proposed. Proper actuation timing has only been addressed in the literature in terms of functional efficiency or metabolic cost during walking. In this study, we will explore identifying the plantarflexor actuation timing in terms of biomechanics outcomes of human walking using a machine learning based algorithm. The third study investigated the recognition of different gait modes encountered during walking. The actuation scheme plays a significant role in walking on level ground, stair descent or stair ascent modes. The wrong actuation scheme for a given mode can cause falls or trips. A gait mode recognition technique was developed for detecting these different modes by attaching an inertial measurement unit and using a classifier based on artificial neural networks. This new algorithm improves upon the current one step delay limitation found as a drawback of a previously developed technique. Overall, this dissertation focused on addressing some important issues related to control of powered AFO that ultimately will help to assist people wearing the device in daily life situations during walking. The proposed approaches and algorithms introduced in this dissertation showed very promising results that proved that these methods can successfully improve the control system of powered AFOs.