Natural posture prediction using fast iterative algorithms

Inverse kinematics is now widely used to predict poses of human-like characters (Wang 1999). As the human skeleton is highly redundant, infinity of postures verify the imposed kinematic constraints. However, only a few of those postures are natural. Energy and Jerk minimization can help choosing a pose during bipedal walking (Nicolas et al. 2007). Additional secondary tasks can be used to drive the solver to a posture that verifies a wide set of constraints. Inverse kinetics solvers (Boulic et al. 1996) are based on such a technique in order to control the position of the center of mass (denoted COM). Whereas the previous methods were based on inverting a non-linear function, cyclic coordinate descent (denoted CCD) enables to decrease computation time (Wang and Chen 1991). As those techniques generally lead to unrealistic postures, a set of biomechanical constraints were added (Kulpa and Multon 2005). However, this method did not offer the range of capabilities that classical inverse kinematics does. An improvement consists in controlling the position of the COM (Kulpa et al. 2005). This paper aims at validating this last method by comparing simulated poses with captured ones.