Simultaneous Learning of Relative and Absolute Spatial Concepts without Any Prior Distinction
Rikunari Sagara, Ryo Taguchi, Akira Taniguchi, Tadahiro Taniguchi · 2021 IEEE 10th Global Conference on Consumer Electronics (GCCE) · 2021
This study proposes a learning method for relative and absolute spatial concepts using spoken user utterances. Service robots that assist humans in performing daily activities must be able to learn and understand spatial concepts and their linguistic representations. Spatial concepts are divided into two types: relative spatial concepts (e.g., front and right) and absolute spatial concepts (e.g., kitchen and corridor). Robots are required to learn both types of concepts. Therefore, we propose a method by which a robot can learn both relative and absolute spatial concepts without prior knowledge of words. The method is formulated using a probabilistic model to mutually complement the uncertainty of locations and spoken user utterances. The experimental results demonstrate that both relative and absolute spatial concepts can be learned under the condition that the robot does not know which type of concept is uttered. In addition, we show that spoken user utterances are segmented with high accuracy.