Demonstrating the effectiveness of combining heuristic and data-driven methods to achieve scalable and adaptive motion styles

Motion style transfer has significantly progressed through data- driven methods, offering high-quality results but demanding extensive datasets, making them resource-intensive. In contrast, heuristic methods operate with minimal motion data but lack adaptability. This paper introduces a hybrid framework integrating specific heuristic techniques, including Radial Basis Functions (RBF) and phase features, with data-driven autoencoder models to balance efficiency and precision in motion style transfer. By leveraging the efficiency of heuristic methods and the adaptability of autoencoders, this proof-of-concept demonstrates the potential for scalable and adaptive motion styles with reduced dataset dependency. While this work focuses on combining these specific methods, the framework can be generalized and extended to incorporate more advanced techniques in future developments. Our experiments on XIA, 100Style, and SHS(Share Space) datasets validate the framework's ability to produce high-quality, adaptive motion styles, paving the way for scalable motion adaptation in applications such as gaming, VR, and robotics.