Autonomy and Proactive Perception for Virtual Actors

摘要

This study lies in the context of virtual engineering and human information systems. We propose to model and implement the behaviour of believable virtual agents, using ideas from psychology (cognitive maps, affordances) and neurophysiology (active perception, movement prediction). Virtual worlds are peopled with autonomous entities improvising in free interaction. Autonomization of a model consists in giving to it a sensorimotor interface and also a decision module so that it could adapt its reactions to inner and extern stimuli. We propose in this article the basis for a behavioral model imitating human beings' perceptive operation. The psychological notion of "affor-dance" will help us in the construction of fuzzy cognitive maps for believable virtual human behaviour specification. Sensus Alain Berthoz, neurophysiologist, perception is not only an interpretation of sensorial messages: it is also an internal simulation of the action and an anticipation of the consequences of this simulated action. Following neu-rophysiological experiments on hippocampus in which was observed oscillations permitting prediction of trajectories, our virtual actor uses fuzzy cognitive maps in an imaginary space and simulate a behaviour. This simulation in the simulation allows him to predict the consequences of actions. The expected benefit from our affordance-based and proactive model consists in elaborating a believable virtual helmsman within the framework of a virtual sailing ship. We have implemented such a virtual actor in the multi-agent environment oRis.