Research on Virtual Human Motion Generation Based on Biological Skeletal Models

Haochuan Wu
Guangzhou Vocational University of Science and Technology, Guangzhou 510550, Guangdong, China

Abstract:

With the growing demand for highly realistic digital characters in virtual reality, video games, film production, and other fields, virtual human motion generation has become a hot topic at the intersection of computer graphics and artificial intelligence. Although traditional motion capture methods can produce high-precision motion data, their high cost and limited flexibility make them unsuitable for large-scale virtual character motion generation. Therefore, this study focuses on virtual human motion generation based on a biomechanical skeletal model, aiming to construct a motion generation system that balances realism and controllability. Starting from human anatomy and biomechanics, a skeletal structure model conforming to human motion principles is established, clarifying the connections between bones and their degrees of freedom. Based on this model, motion control parameters and physical constraints are introduced to design a virtual human motion driving algorithm. A hybrid strategy combining keyframe interpolation, dynamic simulation, and machine learning is adopted to generate natural, continuous, and responsive motion sequences. An action library is also built to enhance the system’s generalization ability across different application scenarios. In the experimental validation phase, a virtual human motion simulation environment was developed on the Unity 3D platform, and typical actions such as walking, running, dancing, and jumping jacks were tested. Experimental results show that the proposed PFNN (Phase-Functioned Neural Network) method outperforms traditional interpolation, template matching, and LSTM methods in terms of motion naturalness and coordination. This research integrates dynamic characteristics with intelligent algorithms to realize an end-to-end motion generation process from structure modeling to control parameter design and motion synthesis, laying the foundation for large-scale virtual human behavior modeling and simulation.