MULTIMODAL INFORMATION PROCESSING AND BEHAVIOUR GENERATION OF HUMANOID ROBOTS BASED ON PALM 2 MODEL AND MULTIMODAL TRANSFORMER ARCHITECTURE, 648-664. SI

In response to the low efficiency of behaviour decision-makingand generation for humanoid robots, single data modality, anddifficulty in effectively integrating information from differentmodalities, this article combines Pathways Language Model 2(PaLM 2) and multimodal Transformer architecture to study themultimodal information processing and behaviour generation ofhumanoid robots.