Fast and accurate simulation of deformable material is a long standing goal across a variety of disciplines ranging from engineering to feature film industry. Despite great strides, physical simulation is still impeded by long design cycles requiring both intuition and technical expertise. Furthermore, as we cannot account for every material property, every interaction, and every particular environment, the scalability of existing techniques poses great challenges.
This research seeks to overcome these obstacles with novel methods that combine acquisition techniques, differential geometry, and mechanics to develop novel simulation models that accurately reproduce realistic deformation behavior. Traditionally, physical assumptions are synthesized into the modeling process by projecting the constitutive equations and energy balance relations onto a priori chosen set of basis functions (e.g. polynomials in the finite element method). This choice of basis can have deep implications on the simulation. Indeed, major challenges such as spurious energy, locking, and hourglass effect can be attributed to inadequacy of the basis. The scientific objective of this project is to develop new deformation models, where basis functions and material behavior are adaptively learned from acquisition, and thus have inherently a clear physical meaning. In this way, the simulation goes on par with the real deformation behavior and the above mentioned problems are avoided in the first place.
The PhysiGrafix project consists of three major tasks
i. Develop efficient techniques for non-intrusive (e.g. optical) tracking and capturing of relevant geometric and physical properties of deformable material and reconstruction of a coarse representation of the deformation sequence;
ii. Design new discrete deformation models where deformation behavior is fully learned from acquisition; with special focus on problem reduction by encoding the physics in relevant deformation modes and elimination of irrelevant parameters (e.g. rigid body modes);
iii. Adapt the simulation to refined reconstruction as well as to further acquisition data.