Colloquium Speaker

Speaker:Yan Zhuang
University of California, Berkeley
Topic:Real-time Simulation of Physically Realistic Global Deformations
Date:Thursday, April 6, 2000
Time:11:00 AM
Place:Gould-Simpson, Room 701

Refreshments will be served in the 7th-floor lobby of Gould-Simpson at 10:45 AM


Physically realistic modeling and manipulation of deformable objects has been the bottleneck of many applications, such as human tissue modeling, character animation, surgical simulation, etc. Among the potential applications, a virtual surgical training system is the most demanding for the real-time performance because of the requirement of real-time interaction with virtual human tissue.

So far real time simulation and animation of deformation has only been achieved in two special cases: 2D problems such as cloth simulation, and small or local deformations for 3D objects.

In this talk, we address the bottleneck problem of real-time simulation of physically realistic large global deformations of 3D objects. In particular we apply the finite element method (FEM) to model such deformation. By global deformation, we mean deformations, such as large twisting or bending of an object, which involve the entire body, in contrast to poking and squeezing, which involve a relatively small region of the deformable object.

Real-time simulation and animation of global deformation of 3D objects, using finite element method (FEM), is difficult due to the following 4 fundamental problems: (1) The linear elastic model is inappropriate for simulating large motion and large deformations (unacceptable distortion will occur); (2) Inverting large sparse matrices is computationally expensive; (3) The time step for dynamic integration has to be drastically reduced to simulate collisions, if penalty method is applied; (4) The size of the problem (the number of elements in the FEM mesh) is one order of magnitude larger than that of a 2D problem.

We counter those 4 difficulties by: (1) using quadratic strain instead of the popular linear strain to simulate arbitrarily large motions and global deformations of a 3D object; (2) diagonalizing the system and restricting the time steps to enable preprocessing; (3) applying an implicit simplified impulse to a decoupled system, which makes an integration step for collision as cheap as a regular dynamic integration step; (4) using a graded mesh instead of a uniform mesh, which reduces the asymptotic complexity of a 3D problem to that of a 2D problem.