Research On User-Oriented Motion Planning For Virtual Human

Motion planning for virtual human is a hot topic in the field of virtual reality which is wildly used in movie, game, military simulation, etc. User-oriented interactive motion planning is a significant aspect of virtual human techniques. The virtual system will be more intelligent and interactive by interact with user using natural language as input. It is meaningful in improving the operation mode and fitting users' visual perception. This paper takes research on the following four tasks.First, we describe the realization plan of motion generation using data-driven approach. Virtual human are kinematically modeled by hierarchical skeleton. We analyze the data structure and the motion updating method of hierarchical model. Quarternion is used for rotation interpolation to avoid gimbal lock. We also use motion blending to create more natural, realistic motion by existing motion dataThen, we create an invisible model for each character controller which is called shadow controller. Each frame, the coordinator receives the output poses from each shadow controller and performs a weighted multi-step blend to produce a final poses without conflict.Next, path planning is realized based on navigation mesh. We introduce the generation of constrained Delaunay triangulations. Local clearances are used to determine if an agent can pass a passage. A high-quality and collision-free locally shortest paths can be computed efficiently by A* algorithms and tangent-funnel algorithms.Finally, the output of character controller and path planning are leveraged and organized into behavior. We use a language-BT method which can turn a high-level graphical description paragraph composed of simple phrases into several events. Then a behavior tree is built using these events according to logic of the time. The behavior tree can rapidly generate a multi-character interaction scene descriped by the paragraph.