Identifying virtual episodes based on character animation traces

Recognizing and captioning the occurrence of virtual episodes can add descriptive capabilities to games and simulations featuring human-like agents. We introduce a captioning heuristic for multi-character animations. The input of our algorithm consists of the traces (lowest-level procedure names) of each character's animation, such as walking, running, talking, reaching, etc. To identify virtual episodes from these traces we pre-authored episode-centric trees called Core Components Trees (CCT). We compute a vagueness measure over each possible match between episode CCTs and the given trace inputs using fuzzy logic, and derive the best match to describe and caption the perceived episodes.