Perception Based Real-Time Dynamic Adaptation of Human Motions

This paper presents a new real-time method for dynamics-based animation of virtual characters. It is based on rough physical approximations that lead to natural-looking and physically realistic human motions. The first part of this work consists in evaluating the relevant parameters of natural motions performed by people subject to various external perturbations. According to this pilot study, we have defined a method that is able to adapt in real-time the motion of a virtual character in order to satisfy kinematic and dynamic constraints, such as pushing, pulling and carrying objects with more or less mass. This method relies on laws provided by experimental studies that enable us to avoid using complex mechanical models and thus save computation time. One of the most important assumption consists in decoupling the pose of character and the timing of the motion. Thanks to this method, it is possible to animate up to 15 characters at 60Hz while dealing with complex kinematic and dynamic constraints.