Motion Tracking with Dynamic Simulation

This paper presents a physics-based real-time animation system for human-like articulated figures. We introduce a novel method for tracking motion data using dynamic simulation. By tracing a desired motion that is kinematically specified by a user using dynamic simulation, our system produces a motion that dynamically and realistically responds to a changing environment ensuring both controllability and physical realism. A tracking controller uses a human strength model as primary constraints, and controls joint angular acceleration within the available range of torque using inverse dynamics. As secondary constraints, the spatial accelerations of the center of mass and end-effectors are controlled. Unlike existing dynamic controllers that control joint torque for each degree-of-freedom (DOF) separately, our dynamic controller controls joint angular acceleration considering the influence of all DOFs using a pseudo-inverse matrix technique. In addition, this paper proposes two extensions of the Newton-Euler inverse dynamics method. One is a proximate solution for handling the closed loop problem. The other is for computing a minimum-moment point between the supporting segment of a figure and the ground for simulating falling motions. We demonstrate the efficacy of our approach by applying our method to a simple lifting task and generating various motions in response to the weight of the lifted load.