DOUG L. JAMES
Assistant Professor, Computer Science and Robotics
www

My primary research interest is multimodal interactive simulation of constrained continuous physical systems, with an emphasis on deformable systems for real time animation and force-feedback haptics. This is an interdisciplinary research area involving computer graphics, robotics, scientific computing, etc. It is also an enabling technology with wide application, e.g., to character animation, surgical simulation, virtual training and planning, visualization, telerobotics, robotic manipulation, and interactive entertainment.
Expert Simulation: Despite great advances in computing power, our everyday world remains filled with phenomena that are many orders of magnitude too complex to simulate interactively with standard "fast numerical methods." We are therefore interested in researching what I call "expert simulators," a term used to describe intelligent systems capable of simulating complex physical systems at minimal runtime costs. Our approach is to invoke massive precomputation to construct efficient data-driven physical models that can be used for low-cost simulation of particular systems under particular conditions. Our previous research has shown that data-driven simulation algorithms can easily produce million-fold speedups for deformable object simulations. Precomputed models also yield unique output-sensitive algorithms, and this is useful for, e.g., supporting force-feedback rendering of contact forces (at 1 kHz) without needing to simulate the entire system at runtime.
Physical Simulation on Graphics Hardware: Commodity graphics hardware has recently become programmable, and is growing in power much faster than general purpose processors. We are researching data-driven simulation algorithms that can exploit these new hardware capabilities. For example, our previous research has shown that physical deformation models can be precomputed and compiled into condensed data-formats optimized for synthesis in vertex and pixel shader hardware.
Reality-based Modeling: In addition to constructing data-driven physical models using numerical precomputation, they can also be directly estimated from real world measurements. This effectively allows us to exploit nature's supercomputer to import the realism of our everyday world into virtual spaces. Current research involves estimating interactive models of dynamic deformable systems.