Simulation
Multimodal Interactive Simulation
We would like to manipulate objects in simulated environments (for instance, surgical simulators and video games), as naturally as in the real world. This is challenging because the simulation has to balance the complexity of realistic contact physics and the real-time constraints of human interaction. Real physical contact is ``multimodal'' --- it produces sounds, forces, and object deformations that are important for human perception.Examples (shown on left) include visual/auditory contact dynamics simulation (with van den Doel and Kry) [.avi 1.7M] and visual/haptic deformation simulation (with James) [.avi 8.6M].
Our work in various aspects of interactive simulation (deformation, sounds, etc.) is described below. In addition, we're developing software and interfaces which tightly integrate different modalities and exploring perceptual constraints on the latency and synchronization in multimodal simulations.
Interactive Deformation
Doug James (Ph.D. student) and I have developed algorithms for very fast simulation of linear elastic deformable objects. The method exploits the coherence of typical interactions to achieve low latency using capacitance matrix algorithms. We use a boundary integral formulation which lends itself well to fast updates. Our implementation, in Java, is fast enough for real-time interaction with objects with hundreds of boundary faces, on ordinary PCs. D. L. James
and D. K. Pai, ``ArtDefo, Accurate Real Time Deformable
Objects,'' in Computer Graphics (SIGGRAPH 99 Conference
Proceedings), 1999. [pdf
1.2M]
Video: Real-time interaction using Virtual
Technologies' CyberTouch [.avi 3M].
More pictures, demos, etc. available on Doug's web
site.
Contact Sound Synthesis
Kees van den Doel (former Ph.D. student) and I developed a physically-based framework for the synthesis of contact sounds in interactive simulation. Rather than using sound samples, we model the sound response of objects due to contact. The sounds are synthesized based on user interaction, and provide cues to the object's material properites, shape, and the location of the contact. Our sound models can then be mapped onto geometric models -- the figure shows a sonified room modeled with our Sonic Explorer program. . Kees'
sound page has some interactive
demos and software.
K. van den Doel and D. K. Pai, ``The Sounds
of Physical Shapes,'' Presence, 7:4, The MIT Press, 1998. pp. 382--395.
[ps.gz 200k].
Dynamics with Smooth Surface Contact
Paul Kry and I developed an algorithm for dynamic simulation of contact between objects bounded by smooth surfaces (e.g., the popular family of subdivision surfaces). Our algorithm automatically parametrizes the contact manifold locally and simulates the dynamics in this manifold. This is useful because integration errors lie in the manifold allowing large steps to be taken without visible artifacts. It also makes it easier to compute forces during rolling and sliding without artifacts due to polyhedral approximation. See Paul's recent M.Sc. Thesis for
more details.
Video: pebble in a frictionless bowl [.avi 3.7MB], the famous
rattleback top, with friction[.avi 4.9MB]
Contact Response Maps
Accurate simulation of impact at interactive rates is difficult -- real contact events are extremely fast (typically < 0.1 ms for metals), and complex (elastic wave propagation inside the object can be a significant factor). Chris Ullrich (former M.Sc. student) and I introduced a fast simulation method using ``contact response maps'' which precompute Green's functions using a time-domain Boundary Element method. During simulation, contact force response can be computed rapidly using convolution.C. Ullrich and D. K. Pai, ``Contact Response Maps for Real Time Dynamic Simulation,'' in Proceedings of 1998 IEEE International Conference on Robotics and Automation, Leuven, Belgium, pp. 1950-1957, May 1998. [ps.gz 47k]
Numerical Issues in Multibody Dynamics
Simulation of tree-structured multibody systems, such as robot manipulators and humans, has a long history; but important numerical issues remain. For instance, we discovered a phenomenon called ``formulation stiffness'' - the fastest forward dynamics methods can have subtle interactions which slow down popular adaptive step-size integration methods. To analyze this phenomenon, we have developed a unified formulation which is of interest in its own right; the composite rigid body method and the articulated-body method are shown to be equivalent to different elimination methods to solve the same linear system, with the articulated body method taking advantage of sparsity.U. M. Ascher, D. K. Pai and B. Cloutier, ``Forward Dynamics, Elimination Methods, and Formulation Stiffness in Robot Simulation,'' International Journal of Robotics Research, 16:6, pp. 749--758, December 97. [ps.gz 71k]