Artificial intelligence has become a central driver of next-generation prosthetic mobility, particularly through EMG-based movement and intent recognition. This review examines 16 recent studies covering a wide spectrum of AI techniques applied to neuromuscular signal interpretation for prosthetic and rehabilitation applications. A structured taxonomy is introduced to classify models into five main algorithm families, revealing that traditional machine learning methods constitute 56.25% of all reported approaches, while deep learning and hybrid architectures each represent 12.5%, highlighting a slow but emerging shift toward more advanced temporal–spatial modeling. The analysis identifies a substantial imbalance in the literature, with most systems developed for upper-limb control and very limited translation to lower-limb prosthetics, despite the unique demands of gait stability, terrain adaptation, and real-time user safety. The review further discusses methodological trends, dataset limitations, and computational considerations, ultimately outlining a focused research agenda for developing robust, lower-limb-specific AI models capable of delivering intuitive and adaptive human–machine teaming in mobility contexts.
Al-Zubaidi et al. (Thu,) studied this question.