Improved and Scalable Online Learning of Spatial Concepts and Language Models with Mapping

Autonomous Robots manuscript No. (will be inserted by the editor) Abstract We propose a novel online learning algorithm, Keywords Online learning · Place categorization · called SpCoSLAM 2.0, for spatial concepts and lexical Scalability · Semantic mapping · Lexical acquisition · acquisition with high accuracy and scalability. Previously,Unsupervised Bayesian probabilistic model we proposed SpCoSLAM as an online learning algorithm based on unsupervised Bayesian probabilistic model that integrates multimodal place categorization, 1 Introduction lexical acquisition, and SLAM. However, our previous algorithm had limited estimation accuracy owing Robots operating in various human environments must to the influence of the early stages of learning, and increased adaptively and sequentially acquire new categories for computational complexity with added training places and unknown words related to various places as data. Therefore, we introduce techniques such as fixedlag well as the map of the environment (Kostavelis and rejuvenation to reduce the calculation time while Gasteratos, 2015). It is desirable for robots to acquire maintaining an accuracy higher than that of the previous place categories and vocabulary autonomously based algorithm. The results show that, in terms of estimation on their experience because it is difficult to manually accuracy, the proposed algorithm exceeds the design spatial knowledge in advance. Related research previous algorithm and is comparable to batch learning. Our approach will contribute to the realization interest in recent years. However, conventional approaches of long-term spatial language interactions between in most of these studies are limited insofar as humans and robots.