Generation of virtual character interaction logic driven by multimodal behavioural data

The generation of adaptive interaction logic for virtual characters remains challenging, as traditional rule-driven methods often produce rigid and contextually insensitive behaviours.To overcome this, we present the multimodal meta-generation network, a multimodal behaviour data-driven framework that synthesises natural and socially appropriate interaction logic from streams including speech, posture, and facial expression.The framework employs cross-modal temporal alignment and hierarchical reinforcement learning to fuse asynchronous signals and enable joint strategy planning with action execution.A causal reasoning module is integrated to enhance social rationality.Experiments on public multimodal interaction datasets demonstrate that our method significantly outperforms baseline models, achieving an F1-score of 0.795 in accuracy and a human subjective score of 4.3 out of 5.0 in naturalness.This research provides a practical solution for deploying adaptive virtual characters in fields such as the metaverse, intelligent education, and remote collaboration.