Intro
I'm a PhD student of Computer Science at University of California, Los Angeles. My research looks for ways to control physically simulated characters interactively. I'm interested in finding general methods for virtually embodied characters to learn complex behavior autonomously.
To compare the agility of state-of-the-art robots against that shown by our animal cousins is humbling. Part of this disparity is due to the mechanical advantages biological systems have versus our motors, sensors and materials. But even in a fully simulated environment, where such limitations are by-design (and so can be cheated with), it remains difficult to control articulated systems. What systems of control could produce artificial characters capable of fluid, animal-like interaction with their environment?
Approaches that work to control simulated systems may (or may not) turn out to be useful to control physical robots. However, control in simulation is an easier problem than physical control, so if you can't control a simulated character, you have little hope to control its physical realization. Working on control in simulation has other advantages as well: the morphology is easily modified, full knowledge of the system's state is available, and the cost of experiments is much lower. In addition, this is a problem with immediate application to the creation of virtual environments, computer games, and visual effects. Controllable, physically simulated characters promise realistic interaction with their environment and other characters. Ideally, changing physical properties of the character (such as resulting from a wound, or carrying a heavy load) would change the character's motion.