Accurate gait phase detection is a fundamental technology for wearable rehabilitation robots, enabling them to deliver effective, safe, and personalized assistance that closely aligns with the therapeutic needs and safety requirements of users. Unlike force sensors or inertial measurement units, surface electromyography (sEMG) signals establish a direct link to the wearer’s motion intentions, making them ideally suited for real-time applications. This study investigates the utilization of sEMG signals to identify gait phases using three types of machine learning models: k-nearest neighbors (KNN), support vector machines (SVM), and artificial neural networks (ANN). The models are trained on a publicly available dataset (i.e., SIAT-LLMD), which consists of data from 40 healthy individuals and utilizes three training methods: hold-out, 10-fold cross-validation, and leave-one-subject-out cross-validation. ANN achieved the best results, demonstrating an overall accuracy of \(92.6\% \pm 2.6\%\) for detecting stance and swing phases, and \(94.4\% \pm 4.2\%\) for detecting five gait sub-phases (i.e., stance flexion phase, stance extension phase, preswing phase, swing flexion phase, and swing extension phase).