| Many interactive (real-time) 3D visual simulation applications need large numbers of animated human figures, which incurs a significant computational load. To maintain the frame rate requirements of interactive visual simulation and enable scalability, the load must be kept relatively constant over a range of human figures, actions, and camera placement. Further, for productivity reasons, animation tasks may be written in low-performance interpreted procedural animation languages.; We have created a procedural animation system for human figures in the high-level scripting language Python. Each animation procedure provides a discrete multi-resolution tradeoff between computation cost and result detail. We prioritize actions by their figure's size, importance, and visibility. We use a reactive fixed frame-rate level-of-detail (LOD) control to maintain a nearly constant animation load. Traditional behavioral culling alone can eliminate non-visible actions to lower overall cost. Our 'behavior prioritization' further attempts to hold costs of visible actions relatively constant over a range of figures and actions, while allocating motion task computation to mirror author-defined priorities and camera placement. It is a load-shedding algorithm which can reduce the output detail to save time. Our work is done in the context of an idealized motion generator, a human arm reaching action and a walking action.; We analyze several attributes of the resulting controlled animation system which affect its performance such as the number of levels-of-detail within an animation procedure, the use of randomization in switching delays, a priority factor, and benefit and value functions.; The coupling of time-critical control on top of animation systems written in lower-performance, but more general-purpose, interpreted languages should lead to wider use of those languages for serious realtime animation work. Our control performs a type of run-time optimization on the interpreted code in the context of animation tasks. Data models of human figures bound tightly to multi-resolution procedural animation code form 'encapsulated models' enabling composition of scalable, real-time, animated virtual worlds. |
