Applications and Benefits of Voxel Constraints in Parallel XPBD Physics Simulation

The fidelity and efficiency of physics simulations in computer graphics, particularly those involving deformable bodies, hinge significantly on the computational methods employed. This thesis investigates the optimization of Extended Position-Based Dynamics (XPBD), a variant of Position-Based Dynamics known for its stability and real-time performance, through the integration of voxel constraints in a parallel computing framework. The primary focus is on three aspects: efficient computation of XPBD models from large meshes, optimization of constraint partitioning to enhance simulation performance, and the automatic generation of long-range constraints to increase stiffness.

Our approach leverages voxelization to transform complex 3D models into a manageable, discrete representation, facilitating the use of Levels of Detail (LoD) to handle varying mesh resolutions effectively. This technique allows for the straightforward application of XPBD simulations by reducing the computational complexity associated with traditional tetrahedral meshes. Additionally, the thesis explores the use of compute shaders to manage parallel read/write operations efficiently, thereby addressing the challenges commonly associated with real-time physics simulations.

The research demonstrates that voxel-based constraints, particularly when combined with LoD strategies, not only improve the performance but also enhance the stiffness and stability of XPBD simulation. This allows for more detailed and complex simulations.