CHS: Small: Fast simulation of geometrically complex multibody systems in contact and self-contact

University Of Southern California, Los Angeles CA

Investigators

Abstract

The ability to simulate complex machinery in contact is broadly applicable to engineering practice. It can be used for virtual training, say in the operation of heavy machinery. Perhaps most importantly, it can be used to assemble and test complex mechanical structures in virtual reality (using a human-computer interface that includes haptic feedback). Such virtual prototyping, as it is commonly called, greatly shortens design cycles, decreases errors, improves product safety and saves millions of dollars in R&D costs. Applications can be found anywhere a complex structure must be designed and manufactured out of many component parts: airplanes, cars, trains, spaceships, power plants, buildings, tools, heavy equipment, etc. In this project the PI will develop computationally efficient collision detection and contact resolution methods that can accommodate complex systems consisting of many objects that are connected by joints and undergoing contact and self-contact. His goal is to devise algorithms that are sufficiently fast to accommodate high update rates (1,000 simulation steps per second for haptics, or more), and that scale to complex real-world mechanisms typically represented by millions of triangles, such as an internal combustion engine or an entire car engine compartment, an airplane landing gear or airplane doors, or excavator machines. Furthermore, whereas previous fast successful industrial penalty-based methods have typically been limited to pairs of objects in contact, in this research the PI's objective is to deal with more complex and realistic situations including rigid objects, joints, friction and self-contact. Fast simulation of multi-body systems in contact is challenging due to the severe computational and stability requirements imposed by complex geometry. Such simulations frequently involve distributed contact, that is to say contact involving many collision sites of varying surface areas and normal orientations that change rapidly over time. Because it is challenging for constraint-based methods to resolve such contact stably at high update rates, the Principal Investigator will exploit industry-proven penalty methods between points and implicit functions (distance fields or voxmaps), and he will extend the approach, which has to date been limited to pairs of objects in contact, to accommodate N >= 2 objects in arbitrary contact, as well as objects connected with joints and undergoing active control. The technical challenges include how to stably resolve and time-step distributed contact between N >= 2 objects, how to stably simulate and render 6-DOF distributed contact in the presence of constraints (joints), and how to handle self-contact and incorporate friction, all the while maintaining high update rates (or gracefully degrading them in case of extreme contact). Because the Principal Investigator's preliminary experience suggests that the discrete nature of current algorithms is an important limitation in practice, he will also investigate continuous collision detection between points and distance fields. Project outcomes will be transitioned to engineering practice via the PI's ongoing collaborations with a number of industrical leaders in high-tech virtual prototyping, and will advance the state of the art in computer graphics, haptics, robotics and virtual reality.