Hardware-ased parallel computing tools for realtime haptic and deformation rendering of soft objects

点击翻译按钮获取中文摘要

Interactive virtual reality simulators are becoming increasingly realistic by incorporating sensing modalities such as vision, kinesthesia and force feedback, also known as haptics. Interactions with non-rigid deformable objects often occur in medical applications involving biological soft-tissue. While a great deal of progress has been made in physics-based mathematical modeling of these interactions, practical applications of such models in real-time simulations have remained limited mainly due to their computational complexity. In particular, applications with haptic feedback would require very high simulation update rates to maintain the system stability, imposing strict timing constraints beyond the capabilities of existing single-processor computers. Building upon our previous work in haptics, this research explores a new paradigm in parallel computing for real-time high-fidelity simulation of soft-object interaction involving visual and haptic feedback. Customized computing tools will be developed that concurrently employ thousands of processing units to solve a large sparse system of equations arising from the finite element models of soft-object deformation. The solution will be obtained at sufficiently high update rates for simulation stability and fidelity. These computing tools will be essentially parallel implementations of recursive equation solvers on synthesized hardware architectures using the Field-Programmable Gate Array (FPGA) technology. Massive parallelization of the computations on multiple interconnected FPGA chips will provide enormous computing power in an inexpensive and compact package enabling the simulation of high-fidelity soft-tissue deformation models. The results of this research will be instrumental in the development of the next generation of computer-assisted surgical systems and medical training simulators. The integration of the resulting core computational technologies into such systems will create improved and new operational capabilities in computer-assisted medical training, diagnosis, and planning and execution of medical interventions. Other scientific applications requiring fast solutions to sparse linear systems of equations will also benefit from this research.