On the Interface Between Steering and Animation for Autonomous Characters

Intelligence in autonomous agents is usually modeled with a hierarchy of layers, where each layer communicates with its neighboring layers through well-defined and often simplified interfaces. This is particularly true about the interface between crowd steering and motion synthesis – common interfaces include force vectors, velocity vectors, trajectory segments, or discrete behavior actions. By assuming one of these interfaces, the steering task and motion synthesis task are well isolated, but this also presents many problems. In particular, problems arise at the critical point where animations are the navigation decisions, for example, when a character chooses to side-step instead of turn, or when a character wants to fit through a tight space. In this project we propose that footprint data – position, orientation, and time duration – is a more appropriate interface. We illustrate this by comparing a behavior interface, vector interface, and a footprints interface. For the purposes of evaluating the interface, we assume “ideal” steering intelligence and “ideal” animation quality in all cases, by hand-constructing illustrative examples. Furthermore, we have developed a prototype space-time planner based on footprints, that allows precise steering in both space and time.