BROWSE
Detail
Conference Paper
A Design of a Dataset for Human-Like Question Generation under Uncertainty for Interactive Robots
Cited - time in 
Share 
Share
- Authors
- Minjung Shin, Minsu Jang, Miyoung Cho, Jeh-Kwang Ryu
- Issue Date
- 2023-06
- Citation
- International Conference on Ubiquitous Robots (UR) 2023, pp.456-458
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Asking questions is essential for humans and robots in learning and interaction. To enable robots to ask insightful questions, we first describe the cognitive process behind the inquisitive nature of humans and then propose a dataset, CAWS (Curious About Weird Scene), consisting of images, textual descriptions, and questions. We employ a text-to-image generative model to produce images from textual descriptions with factual inconsistencies to incur cognitive uncertainties for robots in understanding the images. Our study highlights the significance of considering humans’ inquiry process under uncertainty. We also claim that the proposed dataset can be utilized to improve interactive robot agents and their ability to come up with human-like and insightful questions.
- KSP Keywords
- Cognitive processes, Generative models, Human-like, Interactive robot, Question generation, learning and interaction