A Knowledge-Based Approach to the Synthesis of Human Motion

Abstract

In several current realizations of human figure animation systems, motion control is achieved through physical models of human movement dynamics of various degrees of complexity. Although realistic motions can be produced in this way, the computational time required to solve the equations characterizing such models is considerably high and reduces the possibility of real-time interaction between the animator and the motion control system. Recently, several proposals on the way to build models that require management of complex physical systems have appeared in the field of qualitative modeling. The main assumption of such proposals is that it is possible to obtain convincing simulations of physical phenomena through specification and use of common sense knowledge of the world.

In this paper we present a knowledge-based approach to the synthesis of human motion in computer animation. The challenge of the proposal is to verify the possibility of controlling the movement of the human figure by developing a qualitative model of human motor behavior that avoids the significant calculations necessary if complex mathematical models are adopted. The most relevant characteristic of the system is that knowledge about the human figure and its motor behavior is composed dynamically to fit the task at hand. Task execution can be varied to account for the flexibility and adaptation shown by humans through the control of modeling assumptions expressing common sense knowledge about human motor behavior. Three examples of motor skills implemented according to such assumptions are presented in detail: sitting, stepping upwards, and level walking.