Enriching Coarse Interactive Elastic Objects with High-Resolution Data-Driven Deformations

Abstract

Efficient approximate deformation models allow to interactively simulate elastic objects. However, these approaches usually cannot reproduce the complex deformation behavior governed by geometric and material nonlinearities. In addition, objects having slender shapes require dense simulation meshes, which necessitates additional computational effort. We propose an approach where a dynamic interactive coarse simulation is enriched with details stemming from a more accurate quasi-static simulation in a data-driven way. While the coarse simulation is based on a low-resolution (low-res) mesh and a fast linear deformation model the accurate simulation employs a quasi-static non-linear deformation model at a higher mesh resolution (high-res). We pre-compute pairs of low-res mesh deformations and corresponding high-res details by applying a series of training interactions on both the coarse and the accurate model. At run-time, we only run the coarse simulation and correlate the current state to the training states. Subsequently, we blend detail data in order to obtain a spatio-temporally smooth displacement field that we super-impose on the surface skin, resulting in a plausible display of the non-linearly deformed object at real-time rates. We present examples from both computer animation and medical simulation.