Homepage for the Virtual Human(oid) Project

MPEG movies of human(oid) running

Project Participants

• Robert Bridson

• Ron Fedkiw

• Eran Guendelman

• Sergey Koltakov

• Neil Molino

• Igor Neverov

• Joey Teran

Overview

In the virtual human project we are interested in developing a dynamic deformable model of a human being that is capable of recreating complex biomechanically realistic motions.  The final model will impact research areas from medicine to entertainment.  A virtual human could be used as a test subject in designing medical devices when human testing is dangerous, for example to test the functionality of an exoskeleton that would help paraplegics and quadriplegics to walk.   

This realistic virtual model will require simulation of a number of diverse materials including bones, muscles, fat, organs, skin, hair, eyes, etc. as well as other peripheral materials such as clothing and an environment (air, water, smoke, fire, etc.).  Algorithms are needed to model both the dynamics of human motion and the interactions of the various components with each other and their environments, for example collision detection and response or solid/fluid interaction. 

The modeling process consists of a number of coupled layers.  At the lowest level, the model is simply a skeleton of articulated rigid bones.  However, though seemingly simple, simulation of articulated rigid body dynamics is a complex area of open research.  Issues like collision detection and resolution as well as enforcement of complex constraints in six dimensional joint spaces between rigid bodies complicate the simulation environment considerably.  The next layer of the model is muscle.  Our muscle models are volumetric and are capable of deforming dynamically.  The dynamics are simulated using the finite element method and are at the state of the art in biomechanical muscle simulation.  The active force generating properties are based on tension producing fibers as in nature and the passive muscle properties are generated from an anisotropic constitutive model with a nonlinear stress/strain relation (i.e. material nonlinearity).  At the next layer of the model are fat and skin which envelop the musculoskeletal system.  Fat is also modeled using the finite element method but with a simple linear viscoelastic constitutive model.  Skin is a two dimensional manifold which we model using mass/spring networks as we have done with cloth.  The last layer of the model involves modeling hair as well as rendering.  Collision detection is a very difficult problem for thin materials like hair and merits the development of new algorithms.  Also, because of our familiarity with human outward appearance, rendering a realistic human requires state of the art rendering techniques like sub-surface scattering.