Graphics Reference
In-Depth Information
10.3 PRT and Dynamic Simulation
Soon after the original paper was published in 2002, precomputed transfer found
its way into more general applications. Over time, practical experiments to im-
prove the method were attempted, with varying degrees of success. From this
work, PRT itself became widely diversified for specific purposes. Animation was
one area that immediately benefited from PRT. The paper “Precomputing Inter-
active Dynamic Deformable Scenes” by Doug L. James and Kayvon Fatahalian
presented a pioneering method for applying PRT in dynamic simulation [James
and Fatahalian 03]. Doug James had been actively researching real-time dynamic
simulation from the late 1990s, and was particularly interested in connecting dy-
namic simulation with other human senses such as hearing and touch. The 2003
paper introduces a unique way of precomputing physically based deformation that
incorporates PRT for rendering.
Deformation is a key aspect of realism in the animation of virtual creatures
as well as in general virtual environments, yet physically based animation of de-
formable models is much more complicated and expensive than it is for rigid
models. Physically based deformation requires the simulation of how force prop-
agates through an object. A displacement at a single point on a deformable object
can affect the entire object rather than just the neighborhood of the point. In
this aspect, deformation is akin to global illumination: the addition of a bright
light beam on a small highly reflective object can produce a substantial change of
lighting throughout the environment. However, deformation and global illumi-
nation have been treated separately in computer graphics despite the conceptual
similarity. Precise physically based deformation is apt to be considered of sec-
ondary importance in real-time interactive dynamic simulations.
The 2003 paper by James and Fatahalian proposed a new approach for the pre-
computation of interactive physically based deformable scenes from the perspec-
tives of both deformation dynamics and global illumination. The paper defines
a data-driven state space, which combines the dynamic deformation and globally
illuminated appearance. It is modeled as a collection of “orbits” that store pre-
computed values describing the displacement and appearance of each state. The
collection of all precomputed orbits comprises a discrete portrait, and it is a sub-
set of the full portrait that describes all possible dynamics required for the scene
representation with which a user interacts. The challenge is how to sample and
parameterize a representative collection of precomputed orbits so that they can
approximate the full dynamics for a particular range of interaction at run-time,
as well as supporting the global illumination in the scene for real-time rendering.
In both deformable simulation and animation of global illumination effects, data
reduction is necessary to limit storage requirements and increase the efficiency of
