[1Q5-GS-11] Robot and real worlds: Fundamental theory

[1Q5-GS-11-01] Considering the Environment for Implicit Command Understanding by Service Robots

Using Encoder-Decoder Network and Probabilistic Logic Programming

〇Taichi Naito¹, Naoya Hirota¹, Yoshinobu Hagiwara¹, Naoto Iwahashi², Tadahiro Taniguchi¹ (1. Ritsumeikan University, 2. Okayama Prefectural University)

Keywords:Language understanding, Planning, Implicature, Probabilistic Logic Programming, Encoder-Decoder Network

The purpose of this study is getting the goal description that a robot has to achieve from the user's utterances, and the action sequence of the robot to reach that goal taking into consideration the environment. The aim is to enable a robot to understand utterance commands including not only explicit instructions but also implicit instructions, and to plan a suitable action sequence according to the environment. The proposed method uses Sequence-to-Sequence (seq2seq) to get a goal description from the user's utterances, and Logic Programming to plan the action sequence of the robot in each environment. In our experiment, we compared the proposed method with a previous seq2seq method, with and without considering the environment, by evaluating the resulting action sequences. The results demonstrated that the method enables a robot to understand commands with implicit instructions, and to plan a suitable action sequence in each considered environment.