Research on Real World Understanding by Robots Based on Multimodal Categorization

点击翻译按钮获取中文摘要

In this research project, we address the problem of "robot intelligence" from a viewpoint of how robots can truly understand things. Here, we define the "true understanding" as prediction through categories (concepts), which are formed by categorizing past experiments in an unsupervised way. Therefore, the unsupervised multimodal categorization algorithm is required for robots in order to form the basis of understanding. We have developed some multimodal categorization algorithms based on statistical clustering methods and showed the possibility of the truly understanding by robots.