[1Q3-GS-11] Robot and real worlds: Multimodal information

[1Q3-GS-11-02] Influence of Communication on Multimodal Category Formation between Agents

〇Kazuma Furukawa¹, Yoshinobu Hagiwara¹, Tadahiro Taniguchi¹ (1. Univ. of Ritsumeikan)

Keywords:AI, Symbol Emergence

Humans categorize objects by integrating various information obtained from the objects through sensory modalities such as visual, auditory, and haptic modalities. In this regard, Hagiwara et al. proposed a model that forms categories and signs between two agents manipulating objects. However, they only considered the case where each agent forms categories from a single modality. In this paper, we extend this previous model using multimodal latent dirichlet allocation (MLDA) so that each agent can form categories from multiple modalities. In particular, we investigate the effect of communication on category formation when each agent forms categories using multiple modalities. We conducted experiments in which one or both agents lacked a modality and communicated with each other during the categorization process. The results show that communication improved the accuracy of the categorization of agents lacking a modality.