| Controllable, reactive human motion is essential in many video games and training environments. Characters in these applications often perform tasks based on modified motion data, but response to unpredicted events is also important in order to maintain realism. In this thesis, I approach the problem of motion synthesis for interactive, humanlike characters by combining dynamic simulation and human motion capture data. My control systems use trajectory tracking to follow motion capture data and a balance controller to keep the character upright while modifying sequences from a small motion library to accomplish specified tasks, such as throwing punches or swinging a racket. The system reacts to forces computed from a physical collision model by changing stiffness and damping terms. The freestanding, simulated humans respond automatically to impacts and smoothly return to tracking. I compare the resulting motion with video and recorded human data. |
