[2P4-GS-11] Robot and real worlds: Social application

[2P4-GS-11-01] Symbol Grounding based on Spatial Semantic Representation of Various Parking Instructions for Autonomous Vehicles

〇Nana Otawara¹, Hiroshi Tsukahara², Atsushi Keyaki², Ichiro Kobayashi¹ (1. Ochanomizu University, 2. Denso IT Laboratory, Inc.)

Keywords:Autonomous driving, Belief propagation, Symbol grounding

It is necessary to connect objects in the real world to the objects expressed as words in operating commands. In this study, we attempt to realize such correspondence relationship, (i.e., symbol grounding) between them. In particular, we focus on the parking operation of a car. In previous work, the method deals with particular parking instructions, and the number of those instructions is small. Therefore, we have developed a method of automatically generating a graphical model which infers correspondence relationship to ground parking commands onto the objects observed by the sensors of a car. We conducted two kinds of experiments: one is the method of training each structure of parking instructions one by one, and the other is the method of training all data together. Through the experiments, we have estimated the precision for grounding between driving commands and the objects observed by the sensors of a car, and confirmed that our proposed method is useful.