Innovative martial arts teaching methods based on deep reinforcement learning and biomechanics of movement

To train martial arts effectively, it is necessary to make precise movements, have consistency in their biomechanics and receive feedback in a timely fashion. The teaching of traditional martial arts is founded on subjective evaluation and delayed feedback, which restrict the individualization of training and the accuracy of movements. Intelligent systems that deliver adaptive, real-time guidance are made possible by deep reinforcement learning (DRL) and biomechanical Examination. The aim of this research is to create an intelligent method for martial arts pedagogy which integrates the DRL and movement analysis system to provide students with personalized, constant and adaptive training feedback. The data is from 4000 rows of martial artists recording basic skills such as chopping, kicking and grappling while using IoT sensors. Data preprocessing involves data imputation and min–max normalization. The Independent Component Analysis (ICA) is used to excerpt self-governing movement features from varied biomechanical signals. Pose approximation and sensor fusion are used to derive biomechanical limits such as joint angles, angular velocity, stability of the center of mass, and force delivery. A new Intelligent Cuckoo Optimization fused Dueling Deep Q-Network (ICO-2DQN) is developed, in which the 2DQN determines optimal training strategies and the ICO algorithm to optimize hyperparameters, reward weighting, and learning stability. The proposed ICO-2DQN model was applied in Python and attained an accuracy of 98.7%, overtaking the baseline models. To improve martial arts teaching. The ICO-2DQN method advances martial arts education, resulting in precise, adapted, and efficient sports teaching requests.