Twig: A Simple, AI-friendly, Character World for Believable Agents

Опубликовано 7 сентября 2016, 17:27

Interactive narrative and similar AI-intensive applications require characters to perform a wide range of actions and gestures, the details of which may be difficult to anticipate at authoring-time. In the game industry, character motions are typically generated by blending clips obtained through motion-capture or by hand-authoring key frames. While these techniques can produce highly realistic motion, creating large motion libraries can be prohibitively expensive for most universities, independent developers, and solo artists. In this talk, I will describe Twig, a fast, extensible, procedural animation system, built on XNA Game Studio, and designed for interactive narrative applications. Twig provides a simplified dynamic simulation specifically designed to be easy to control. Characters are controlled in much like physical puppets: by applying (linear) forces as needed directly to different body parts, rather than through joint torques and inverted pendulum control. The intrinsic physics of the body then finds positions for the remaining body parts and smooths the characterΓÇÖs overall motion. This ΓÇ£puppetryΓÇ¥ style of control provides the simplicity of kinematic control within an otherwise dynamic simulation. Although less realistic than motion capture or full biomechanical simulation, Twig produces compelling, responsive character behavior. Moreover, it is fast, stable, supports believable physical interactions between characters, and makes it easy to author new behaviors. I will also discuss applications of the system, including a simple scripted ΓÇ£web comicΓÇ¥ and a simulation of Reinforcement Sensitivity Theory, a neurophysiologically-based model of emotion and personality (in collaboration with Karl Fua, Andrew Ortony, and Bill Revelle).