动能模型引导的动态虚拟人控制

Physically based virtual human control is a typical dynamic problem, which is of great significance to the fields of game development, film special effects, etc. However, traditional dynamic controller model is complex and unstable. To solve this problem, a novel physically-based control framework for dynamic virtual human using kinetic energy model is proposed. Firstly, preprocess reference motion in Riemann geometric space and establish the thermal distribution diagram of kinetic energy. Secondly, obtain the control parameters by analyzing the thermal diagram. Finally, compute restorative torques to rebalance the virtual human and improve the accuracy of posture based on the estimated parameters. In addition, a time alignment algorithm is also presented to integrate multiple reference motion. Simulate the full-body bipedal motion in several situations, including complex terrain, motion transition and external force, to obtain walking and running data of various speeds and directions for evaluation. The results show that, compared with DeepLoco, the proposed framework provides smaller fluctuation coefficients of the change of the center of mass velocity under external force, which demonstrates the robustness. In addition, it achieves 2X performance increase compared to DeepLoco, demonstrating its efficiency.