Abstract For a long time, dance education in Chinese universities has relied on teachers watching students and students practicing over and over again. This method often makes it hard to give objective feedback, correct mistakes quickly, and give personalized feedback, especially in big or diverse classes. In these circumstances, it is challenging to detect and rectify subtle biomechanical and rhythmic deviations using traditional teaching methods. Recent developments in artificial intelligence (AI) and wearable sensor technologies provide an alternative by facilitating continuous motion capture, quantitative movement analysis, and data-driven instructional support. This project creates an AI-based Dance Movement Teaching Support System (DM-TSS) that aims to improve the accuracy of motion acquisition, the reliability of feedback, and the effectiveness of instruction in higher education dance training. The system combines wearable inertial sensors with deep learning and reinforcement learning models to analyze dance movements that involve more than one joint in real time and give personalized feedback. We present a new framework called Namib Beetle Optimization–Twin-Stage Hierarchical Deep Reinforcement Learning (NBO–TSH-DRL) that helps with adaptive feature selection and hierarchical decision-making for classifying and evaluating dance movements. We used sensor data from traditional Chinese dance training sessions to test the experiment. The proposed framework outperformed baseline methods, such as GRU, 3D-CNN, and PSO-optimized models, achieving an accuracy of 97.9%, an F1-score of 0.98, and an AUC of 0.99. The system not only makes things work better, but it also lets teachers give real-time feedback and lets students see their movements through AI-assisted dashboards. These results show that combining AI with wearable sensing technologies can make dance education more objective, personalized, and consistent. The suggested method shows how intelligent teaching systems could help modernize dance training in Chinese universities while also helping to preserve cultural heritage and making teaching methods more flexible. Future efforts will concentrate on augmenting cross-style datasets, enhancing model generalization, and integrating multimodal feedback to facilitate wider educational implementation.
Lu et al. (2026) studied this question.