When all’s said and done a wide range of core geometry, simulation and design tasks, ranging from parameterization to physical modeling, boil down to a variational problem: minimizing some interesting energy subject to satisfying a set of important constraints.
While numerical optimization is now an advanced field, its off-the-shelf solutions can be slow, and regularly fail for important graphics and mechanics applications. Particularly they can generate low-quality solutions, fail to converge, or even produce spectacular garbage by blowing up altogether. This is largely because geometry and physics problems have unique challenges that are often not covered by standard optimization toolkits and, likewise, there are important, problem-specific structures that are not taken advantage of.
In this talk I’ll discuss our recent work towards developing optimization tools customized for geometry, physics and animation tasks that ``just work’’ by paying attention to these ``details’’. The goal is to enable methods that are automatic, robust and efficient. As examples I’ll cover some recent optimization algorithms we’ve developed for distortion minimization, mesh parameterization, shape interpolation, inverse-kinematics, and deformable-body simulation. If time, I may also briefly cover contact mechanics and fabrication design. For a preview of these projects see dannykaufman.io.
BIO: Danny Kaufman is a Senior Research Scientist at Adobe Corp, based in Seattle Danny’s research focuses on developing computational models and tools for animation, design, fabrication, and physical simulation. His postdoctoral work was with Eitan Grinspun (Columbia Computer Graphics Group) while his PhD was with Dinesh Pai (SSL, University of British Columbia).