物理仿真虚拟人运动交互控制技术研究

Despite the fact that physically simulated characters can overcome the inherent drawback of data-driven methods which are restricted by the scale and contents of the available motion samples, the use of physics-based simulation in virtual characters animation is still limited because of the challenges such as activeness, redundancy and under-actuation faced in the process of human movements modeling. The controllability of physically simulated characters becomes the bottleneck problem which needs to be solved for the furthe development of the virtual characters animation technology..In order to make physically simulated characters response adaptively and life-likely to different unpredicted interactive control inputs from users or dynamic changes in external environments to the largest extent, the interactive control difficulties in physically simulated characters are solved coordinately under the high and low level from different aspects. For the low-level control aspects, firstly, the determination of preview control variables and their corresponding on-line planning based on the simplified dynamic model are researched to increase the robustness and responsiveness of physical controllers. Secondly, the method to satisfy kinematic constraints is provided to increase the control precision. For the high-level control aspects, the task-centric algorithm to abstract and integrate different skill-specific controllers in sequence along the timing axis is researched to complete more complex tasks and increase the cover scope of the whole task space..The purpose of the research is to establish the technology framework which can increase the controllability of physically simulated characters comprehensively. The results of this research would have a significant theoretic and applicable role on the further development and improvement of the virtual characters motion control technology.

物理仿真方法产生的虚拟人运动可以克服数据驱动方法受限于动作样本库的缺陷，成为数据驱动方法的必然补充和深化。但由于人体运动系统的主动性、冗余性和欠驱动特性，物理仿真虚拟人的可控性成为亟待解决的瓶颈问题。.项目具体研究内容包括：在底层，研究动力学仿真计算过程中前瞻控制量的确定及其在线规划控制算法，增强控制器的鲁棒性和反应性；研究运动学约束的精确满足方法，提高物理仿真方法的控制精度；在高层，以任务为中心对底层控制器进行抽象，研究多控制器在时间轴上的衔接集成算法，提高对任务空间的覆盖能力。主要创新和特色在于整个思路框架从不同侧面在高低两个不同层次上协调一致的对物理仿真虚拟人交互控制技术进行研究，拟使物理仿真虚拟人能最大限度地适应各种难以预料的交互控制输入和虚拟场景的动态变化。.项目力图建立一套全面增强物理仿真虚拟人运动交互可控性的技术框架，对虚拟人交互控制技术的发展和完善具有重要的理论与实用价值

按照研究计划，本项目重点从增强虚拟人运动的可控性和对外部环境的适应性角度出发，对虚拟人多层次运动规划集成方法进行了研究，将虚拟人运动控制过程划分为三个主要层次：一是全局路径规划，二是局部实际运动路径生成，三是实际动作姿态的生成。在解耦降维上，采用任务空间描述和虚拟人运动状态空间描述相分离的思路进行，增强了虚拟人的交互控制能力。在在线规划控制方面，提出了具有速度和加速度约束的时间最优轨迹规划算法，可规划出一条从初始位置至目标位置的具有时间最优的运动路径，同时保证虚拟人满足最大速度和最大加速度的约束；在运动学及各类高层约束满足方面，利用模糊规则、模糊推理建立情感、社会学角色和个性参数对虚拟人行为的影响关系；在控制器集成方面，重点对障碍环境下的组行为生成方法进行了研究。总体来看，项目研究基本按照原计划进行，研究任务全部完成。