Modelling plausible steering behaviours for autonomous virtual humans in complex environments

Abstract

In order to create plausible virtual humans it is important to model their movement and interactions with their environment in an accurate and realistic manner. A lot of time and effort is spent by artists and engineers modeling user interactions with virtual agents with which the user acts directly. Virtual crowds, however, form an important component of virtual worlds. It would be unfeasible to author scripted behaviors and interactions for individual members of a virtual crowd. It is typical to rely on systems that allow for autonomous navigation and behavior. We propose some new algorithms and approaches to enhance the plausibility of crowd simulations. A key component of autonomous human behavior is how they avoid collisions. Many popularly adopted collision avoidance models take as absolute that motion is nonholonomic, or that velocity is expressed only in a forward direction, explicitly ignoring lateral velocities. This does not fully describe how humans actually avoid collisions, however. We propose a model that incorporates holonomic collision avoidance strategies, (i.e. sidestepping). We describe our model, built from measures and insights gleaned experimentally, that adds holonomic behaviors to the array of collision avoidance strategies that a virtual human can employ. From this research, we learned that holononric behavior is largely restricted to complex scenarios where time to collision is short.