Precomputing Interactive Dynamic Deformable Scenes Doug L.Jams and Kayvon Fatahailian 报报报 报 :宋
Jan 28, 2016
Precomputing Interactive Dynamic Deformable Scenes
Doug L.Jams and Kayvon Fatahailian
报告人:宋超
Physically Based Modeling and Interactive Simulation
Approach
a. Analysis Method: to get analysis solution of the physics equation
b. numerical method : FEA,caculus of differences,etc
c. data driven
▪ Challenge: a. the difficulty of getting the analysis solution
b. no-linear question widely exiting
c. how to acquire the data?
d. how to use the data?
About data-driven An important strategy How to identify and control complex
systems Former works ▪ Nelles 2000----Nural Network
▪ Reissell and Pai 2001 ----ARX models
▪ Atkeson et.al.1997---Locally weighted Learning (Lazy learningl)
▪ D.Jams and K. Fatahalian
Impulse response functions (IRFs)
Precomputing Interactive Dynamic Deformable Scenes
Contribution ▪ black box offline simulators
▪ Dimensional Model Reduction
Excellence ▪ Robust
▪ Real-time
▪ Handle nonlinear deformation
▪ Illumination
▪ can be synthesized on programmable graphics hardware
Using Scope ▪ particular system ▪ very particular interaction conditions
Precomputing Interactive Dynamic Deformable Scenes
Procedure
▪ Fore treatment (including mesh ,creating
the mechanics model)
▪ Dimensional Model Reduction
▪ Analyze the interaction condition
▪ Pre –calculate and create IBFs
▪ Implement
About Fore-treatment and acquire deterministic iteractor
Get geometric mesh Determine the system DOF Determine the pre-computing
• According the interaction based on probability.
Dimensional Model Reduction(1) Deterministic static space model Dynamics: Appearance
State nodes: Time step edge: Orbits: a temporal sequence of nodes,connected
by time step edge Discrete phase portrait(P): the collection of all
pre-computed orbits
),( )()()1( ttt xfx
)()()1( ,( ttt xfx )()()1( ,( ttt xfx
)( )()( tt xgy
)(tx),,( )()()1( ttt xxe
Dimensional Model Reduction(2)
Model Reduction Detail
N state nodes,v vertices
N displacement field
that is (each u has 3 vector components)
vikiu ,...1
Nvvv
N
Nu
uuu
uuu
uuuA
21
122
11
21 ],,,[
Dimensional Model Reduction(3)
Model Reduction Detail
▲ a small number of vibration modes can be sufficient to approximate observed dynamics.
(SVD)
Re-parameterization of the phase portrait
the state vector:
Dimensional Model Reduction(3)
Reduced state vector coordination ▲displacement
▲velocity
Precomputation Process Data-driven modeling complication
insufficient data;high-demensional state space; divergence of nearby orbits;self-collisions.etc
Impulse Response Function
IRFs Index: IRFs:
FIx ,,
Impulse Response Functions(2)
An important special case : FI
Impulse Palettes
Impulse palette based on IRFs:
Impulsively sampling the phase portrait
▲sample time
▲no redundancy
▲orbits terminate
Simulate Implement
Blending Impulse Responses
Approximate the IRF at
That is
x
Example(1)
Dinosaur on moving car dashboard
Plant in moving pot
Cloth on moving door
Example(2) The pre-computing time
Example
Example(3)
Thank you!