Motion Skills for Simulated Humans

描绘人体运动是计算机图形和动画的一个重要方面，其应用包括训练模拟，计算机游戏和电影视觉效果。与计算机图形学中的其他现象一样，例如皮肤的逼真渲染或流体或头发的逼真动画，从第一原理和对给定现象的深入理解中开发的模型为实现高质量和可推广的结果提供了最佳前景。基于原则的物理模型的发展从根本上改变了计算机图形渲染和被动现象动画中使用的工具集。然而，在基于物理的人体运动模型方面取得类似的进展仍然具有巨大的挑战性。它需要解决许多相同的控制问题，这些问题在我们的日常运动中似乎可以毫不费力地由人体解决。我们的研究针对的是目前主要基于运动剪辑和“布娃娃”模拟的技术水平与人类相当大的运动技能之间存在的相当大的差距。我们的重点是运动，如步行，攀爬，室内和室外环境中与运动相关的任务，以及物体抓取和操纵。 我们工作的首要目标是为动作技能的设计和表现开发一个强大而富有表现力的“词汇表”。具体目标包括：（a）用于创作和整合技能的语言;（B）用于技能渐进学习的发展策略;（c）解决作为推理问题的运动计划;以及（d）开发特定于任务的运动模型，例如，开发人们在特定工作场所或任务相关场景中如何移动的特定模型。 我们的研究有助于为下一代人体运动创作系统奠定基础，如训练模拟，游戏和电影所需。特别是，下一代游戏机和电影视觉效果越来越多地暴露出人体运动动画的质量作为内容制作的限制因素，因为它进一步福尔斯于渲染质量的不断进步。虽然计算机动画仍然是我们的主要目标应用，但我们的工作也为机器人技术提供了见解和解决方案，更具体地说，为敏捷的腿部运动和灵巧的机器人操作提供了见解和解决方案。最后，我们的工作为人类运动控制和生物力学提供了见解和理解，因为我们有能力为许多运动技能创建高性能的合成模型。

Depicting human movement is an essential aspect of computer graphics and animation, with applications that include training simulations, computer games, and visual effects for film. As has been the case for other phenomena in computer graphics, such as the realistic rendering of skin or the realistic animation of fluids or hair, models that are developed from first principles and from a deep understanding of a given phenomenon offer the best prospects for achieving high-quality and generalizable results. The development of principled physics-based models has radically transformed the tool-sets used in computer graphics rendering and for the animation of passive phenomena. However, achieving similar progress for physics-based models of human motion remains hugely challenging. It requires solving many of the same control problems that are seemingly solved so effortlessly by the human body during our daily movements. Our research targets the considerable gap that remains between the current state of the art, which is largely based on motion clips and "rag doll" simulations, and the considerable motor skills of humans. Our focus is on motions such as walking, climbing, movement-related tasks in indoor and outdoor environments, and object grasping and manipulation. An overarching goal of our work is to develop a powerful and expressive "vocabulary" for the design and represention of motion skills. Specific objectives include (a) languages for authoring and integrating skills; (b) developmental strategies for the progressive learning of skills; (c) solving for motion plans as an inference problem; and (d) developing task-specific models of motion, e.g., developing specific models of how people move in specific work-place or task-related scenarios. Our research helps lay the groundwork for a next generation of authoring systems for human motions, as required by training simulations, games, and film. In particular, the next generation of game consoles and visual effects for film is increasingly exposing the quality of human motion animation as a limiting factor in content production, as it falls further behind the relentless advance of rendering quality. While computer animation remains our primary target application, our work also offer insights and solutions for robotics, and, more particularly, for agile legged locomotion and dexterous robotic manipulation. Lastly, our work  offer insights and understanding for human motor control and biomechanics, given the ability to create highly capable synthetic models of many movement skills.