Optimization of Wushu Movement Learning and Training Methods Based on Deep Reinforcement Learning

Abstract

The present research introduces WushuDRL-Net, an intelligent framework based on deep reinforcement learning (DRL) to enhance Wushu movement learning and training. The system employs the Proximal Policy Optimization with Motion Alignment Feedback (PPO-MAF) technique, which combines expert imitation, rhythm alignment, and scoring accuracy, to give real-time motion evaluation and correction. Using competition-grade datasets, WushuDRL-Net aligned expert performances with over 90% accuracy and increased execution consistency by 17%. The technology offers dynamic feedback, rhythm synchronization, and adaptive skill enhancement for routines such as Changquan, Nanquan, and Taolu. The experimental results reveal significant increases (p < 0.01) in rhythm control and form precision, demonstrating the scalability and efficiency of WushuDRL-Net. By bridging the gap between traditional martial arts instruction and state-of-the-art AI-driven training, this study encourages intelligent, data-driven Wushu practice.