http://www.cds.caltech.edu/~doyle/MURI.html

"Mathematical Infrastructure for Robust Virtual Engineering"

Abstract (from proposal)

Virtual Engineering (VE) is a multidisciplinary engineering domain in which virtual reality interfaces, simulations, and integrated databases are combined to take complex systems from concept to design, including manufacture, operation, maintenance, and training. It could also include simulation-based decision support in policy making and real-time C3I. The enormous potential of VE is widely recognized, is being broadly pursued and heavily funded, and will take advantage of the expected availability of sophisticated VR human/computer interfaces, and parallel, distributed, high-performance computers, networks, and databases. Less appreciated is that a sound theoretical framework is needed to avoid compromising the potential power of VE.

Theory. Successful VE requires new mathematical and computational methods (VE theory, or VET) to integrate the currently diverse approaches for modeling, simulation, and analysis of hierarchical, multiresolution and variable granularity models of heterogeneous systems with interacting fluid, structural, material, chemical, electromagnetic, and embedded computer and electronic subsystems. We strongly endorse the view expressed in the MURI announcement that conventional methods of modeling and simulation (M&S) are inadequate for several reasons. A unifying theme and dominant technical issue in VE is that complexity is a byproduct of designing for reliable predictability in the presence of uncertainty.

A familiar example is smart weapons, where sensors, actuators, and computers are added to counter uncertainties in atmospheric conditions, release conditions, and target movement, thus trading off sensitivity to uncertainties in the environment versus a large number of new components. Because we build and interconnect the components, this tradeoff presents the potential for both enormous benefit as well as truly catastrophic failures. Explicit models of uncertainty are thus critical; as we build increasingly complex systems, evaluating these tradeoffs can be conceptually and computationally overwhelming. We need an integrated and coherent theory of modeling, analysis, simulation, testing, and model identification from data, as well as a suitable software architecture and engineering environment.

We believe this program must build on the success of robust control and dynamical systems, but the conventional scopes of these fields must be greatly expanded and integrated. Our research will focus on developing theoretical tools for analyzing uncertain simulation models that incorporate heterogeneous components, variable resolution and granularity, and computationally efficient algorithms. We will develop an abstract representation for VE component models which, by design, allows system-level questions to be answered in a robust and provable fashion, and allows models to be developed from first principles and systematically refined via experimentation and testing.

Software design.The computer graphics community has had great success creating geometric models and images with realistic appearance. Our goal is to achieve the same impact for modeling and simulating realistic physical behaviors. To these ends, we will be developing a new physically-based modeling language which allows uncertain, variable-resolution, heterogeneous models to be interconnected and simulated for the purposes of design and analysis. From theory to software, we will maintain robustness and scalability, and software modularity, model communicability, and code reusability. To facilitate rapid transition via existing collaborations, the theory will be initially be implemented as toolboxes in Matlab-like environments.

Testbeds.A series of experimental testbeds will leverage existing activities, with an emphasis on Unmanned Aerial Vehicles (UAVs). The flagship testbed, a formation flight UAV system ideally suited to test VE concepts, teams Caltech with UCLA, Rockwell, and NASA and will allow us to go from mission statement to conceptual design all the way to build and fly on a very short time scale. Although only about 20% of this program's budget (10% at Caltech, 10% at UCLA) is devoted to this testbed, we will leverage much larger activities at UCLA, Rockwell, and NASA, as well as flight control design activity in Caltech's PRET program. These larger efforts are already in place and initial flights are expected by summer, so our VE program will be able to concentrate on VE issues. We will also describe other complementary testbeds that will be used in this program.

Research team.We have an unusually strong team, with experts in dynamical systems, robust and nonlinear control, and computer graphics as well as individuals who are experts in different application domains. This interdisciplinary team has the mathematical depth and engineering experience to develop the theoretical and software tools required for success in VE research, as well as strong contact with applications through a combination of university experiments and industrial partners.