Reasoning of Intelligent Virtual Agents with Exceptional Behavior in Rare Scenarios

Abstract

This paper presents a concept for temporal reasoning. Temporal reasoning enables non-player characters (NPCs) to infer new knowledge based on previously acquired information and apply it within their decision making process. The approach aims to enhances the credibility of NPC behavior by allowing them to make flawed decisions based on ambiguous or misleading information. As a proof of concept a pedestrian crossing scenario with occluded vision was realized and evaluated inside the FIVIS bicycle simulator system. It could be demonstrated that so far rare simulation scenarios such as traffic accidents could be simulated in a more plausible and nondeterministic nor randomized way. These scenarios only occur if certain conditions are met and even in these cases only if the decision making process has been influenced by incomplete, error prone or ambiguous input data, so that proper decisions were difficult to derive. That means our approach is able to model human failures and ambiguity in the decision making process.