Research on GPU-accelerated High Quality Cloth Simulation Algorithm (Supported by Natural Science Foundation of Zhejiang Province, No. LZ16F020003)

The project aims to research on the accuracy and efficiency problems of the high-resolution simulation of cloth, by designing parallel streaming algorithm optimized for GPU architectures, and by mapping the entire process of cloth simulation to GPU kernels which are executed fully in parallel. To ensure accurate computation, the explicit and implicit integration, collision detection and collision response are carefully accelerated. The algorithm can handling either collisions between the fabric and the environment objects, or self-collisions. The research focuses on: GPU architecture based cloth simulation streaming mapping algorithm for implicit integration, a compact representation of sparse matrix, streaming collision detection and response algorithms, a GPU-accelerated strain limiting algorithm, and an adaptive re-meshing algorithm which is optimized for the GPUs. Expected result is a GPU-accelerated high-resolution cloth simulation system. Its efficiency on a NVIDIA Tesla K40c can be two orders of magnitude faster comparing to a single-threaded CPU implementation. This algorithm makes high-quality cloth simulation with the resolution of 2M triangles satisfying real-time requirements (processing time per frame is less than 30ms). The project will open a new avenue for developing high fidelity physically-based cloth simulation system, provide strong technical support for the high-precision real-time physical simulation, and present a useful reference for solving physics simulation related issues. Therefore, to carry out the research is with both scientific significance and application prospectives.