Adam Herbst

Period 1

Project Proposal: Solid-State Physical Simulation in Voxel Space


The purpose of the project is to develop efficient yet accurate methods for 
simulating interactions between voxelated object models. This technique is 
meant to retain the simulation's realism while allowing it to run in real time 
on a standard personal computer, for use in physics-intensive games or other 
applications.

There are different ways to implement individual voxels. The simplest, least 
memory-intensive, and most efficient way is to treat each voxel as a boolean 
entity, a small cubic region of space whose only attribute is whether or not it 
contains matter. However, this method compromises the accuracy of the description
of space for efficiency. More accurate simulations might include information on
the density and color of matter in each voxel, but calculations on these quantities
would reduce program efficiency.

The method of determining voxel interactions also affects accuracy and speed. A
quick and easy way is to calculate a given voxel's interactions with only its 26
immediate neighbors, or those within a given radius; this assumes the effect of
more remote voxels is negligible. Also, what attributes of a voxel should be 
affected in its reactions to external influences position only, or internal 
characteristics as well. Again, a middle ground must be found that optimizes 
program speed and accuracy. The intended scope of my research is the implementation
of the simulation in as many of these different ways as possible, so that I can 
find the optimal simulation.

There are already many programs consisting of countless algorithms for physical
simulation in continuous space. The models in these simulations are usually 
constructed from polygonal faces that share vertices, thus creating a continuous
surface. I propose to create a model in which space is divided into infinite 
identical cubes (voxels, or volume elements), each of which can either contain 
matter or be empty. Although I have found research on graphical rendering using
voxel space, attempts to use it in physical modeling are rare. This project 
would combine the concept of voxel space and current methods and knowledge 
relating to physical modeling in the hope of improving upon simulation accuracy.

I plan to use C++ for all code in my project, employing OpenGL for graphical 
display of the simulation. I will also need to find (online) or experimentally 
determine reasonable data for the voxel attributes of all models I build. Laws 
governing voxel interactions will depend on the size of each voxel and the 
formation of matter I assume to exist within it; optimizing these laws for 
accuracy and efficiency will be the core experimental section of the project.

I will present the results by comparing the program efficiency to its physical 
accuracy (possibly a subjective measure, since a truly accurate simulation of 
this magnitude is intractable on a PC and therefore cannot be measured against 
mine) for each configuration of the physical system that I use. A working program
should be possible within a few months, but I can spend the entire year developing
new versions of it. These results will be helpful to future developers of applications 
that use simulation of macroscopic bodies in a realistic physical environment.