Modeling
Modeling in HAVEN
At Rutgers, my focus is on modeling both human interaction and physical objects using the Rutgers Haptic, Auditory, and Visual ENvironment (HAVEN). We are currently building the HAVEN, funded in part by NSF Major Research Instrumentation grant EIA-0215887.Details coming soon...
Reality-based Modeling
For acquisition of models based on actual measurements of real objects, we developed ACME, the UBC Active Measurement Facility. Our goal is to acquire models of behavior (e.g. deformation, sound, contact textures) in addition to geometry and appearance.See my simulation page for research on modeling using only a priori knowledge of physical laws such as Navier's equation and classical mechanics.
Sound Modeling
Parameters of sound synthesis models of real world objects are difficult to compute, not only because PDEs have to be solved on complicated domains, but also because material properties and internal geometry are difficult to estimate. We have developed a technique for direct estimation of sound model parameters using measurements from ACME, with former M.Sc. student Josh Richmond. Josh developed a "sound effector" for ACME with which we can easily acquire a large number sound measurements for parameter estimation. J. L. Richmond and D. K. Pai,
``Active Measurement and Modeling of Contact Sounds,'' in
Proceedings of the 2000 IEEE International Conference on
Robotics and Automation , San Francisco, April 2000,
pp. 2146--2152.
[pdf 360k]
Video: contact sound acquisition
[1.3M.avi]
[10M.avi]
For example sounds, more videos, etc. see
Josh's
research page
Deformation Modeling
With Jochen Lang (Ph.D. student, co-supervised with Woodham), we have developed techniques for acquiring linear elastic deformation models (suitable for fast simulation). Jochen's software systematically applies known tractions to boundary vertices using a force-sensing robot arm, and measures global displacement of the surface using trinocular stereo vision. We can thus directly estimate the Green's functions which approximate the deformation response of the object. If necessary, we can also estimate the material constants of a homogeneous linear elastic object.Video: Toy tiger [.avi 5M]. Papers coming soon.
Contact Texture Modeling
By contact texture we mean the fine variation in the contact force due to friction and roughness. This is important for haptic displays and contact simulation. John Lloyd and I have developed techniques to automatically acquire these textures using ACME. A robot systematically explores the surface of object, stroking the surface in various directions, monitoring the contact force and motion. From this Coulomb friction and parameters for a stochastic roughness model are estimated. As with our sound modeling, we can adaptively sample the surface, automatically taking more measurements where the surface parameters change quickly. The figure on the right shows the Coulomb friction map from the bottle.J. E. Lloyd and D. K. Pai, ``Robotic Mapping of Friction and Roughness for Reality-based Modeling,'' to appear in proceedings of 2001 IEEE Intl. Conf. on Robotics and Automation.
Bayesian Estimation of Surfaces from Laser Scans
We have developed a Bayesian estimation method for measurement of both range and surface orientation using a laser range finder. The method not only provides more accurate estimates of range for dark surfaces that are difficult to measure, but also simultaneously provides estimates of surface normals.J. Lang and D. K. Pai, ``Bayesian Estimation of Distance and Surface Normal with a Time-of-Flight Laser Rangefinder,'' in proceedings of the Second International Conference on 3-D Digital Imaging and Modeling, Ottawa, October 4-8, 1999. [pdf 1.1M] [ps.gz 800k]