A decision network framework for the behavioral animation of virtual humans

We introduce a novel framework for advanced behavioral modeling in virtual humans that features the first application of decision network techniques to computer graphics and addresses the challenging open problem of simulating the complex interactions of real people in urban settings. Based on hierarchical decision networks, our behavioral modeling framework handles uncertainty, interprets perceptual information, incorporates personality traits, and facilitates the autonomous control of behavior. Combining probability, decision, and graph theories, our framework yields autonomous human characters that can make nontrivial interpretations and arrive at intelligent decisions that depend on multiple considerations.; Unlike prior work in so-called "crowd simulation", we develop complex autonomous individuals that, in addition to motor and perceptual components, include broad behavioral repertoires that are much more challenging to model. In particular, our self-animating pedestrians can independently assess the interrelationships among all the relevant factors to make rational decisions in the presence of uncertainty.; Within our framework, we develop an emergency response behavior model, which enables virtual pedestrians to respond to an emergency situation in a variety of ways. We also develop a behavior model for establishing partnering relationships, an acquaintance behavior model enabling two characters to greet one another, and a collision avoidance model, all of which take into consideration personality traits, internal factors, and the interpretation of intentions.; Our virtual human simulator, which includes the aforementioned behavioral models, can automatically animate pedestrians in a large urban environment that interact with each other in a realistic manner. We demonstrate the potency of our decision net work framework in several behavioral animation scenarios involving interactions between autonomous pedestrians, including an elaborate, automatically-generated emergency response animation.