ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper End-to-End Learning-Based Non-Verbal Behavior Generation of Social Robots
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Woo-Ri Ko, Minsu Jang, Jaeyeon Lee, Jaehong Kim
Issue Date
2021-11
Citation
International Conference on Social Robotics (ICSR) 2021: Workshop, pp.1-2
Language
English
Type
Conference Paper
Abstract
In order for users to feel familiar with social robots, it is important for social robots to generate non-verbal robot behaviors, such as handshakes. However, the traditional approaches of reproducing pre-coded motions allow users to easily predict the robot’s reaction, giving the impression that the robot is a machine and not a real agent. To enable social robots to learn multiple human-like behaviors from human-human interactions, we proposed an end-to-end learning-based behavior generation method. The Seq2Seq architecture consisting of two long shortterm memory units was adopted. One is for encoding user behavior and the other is for generating the next robot behavior. To demonstrate the effectiveness of our method, two experiments were performed using a humanoid robot, Pepper, in a simulated environment. Experimental results showed that the robot can generate five social behaviors, i.e. bow, stand, handshake, hug, and block face corresponding to user behavior, and adjust its behavior according to the user’s posture.
KSP Keywords
Behavior Generation, End to End(E2E), End-to-end learning, Human interaction, Human-like, Humanoid Robot, Learning-based, Multiple human, Nonverbal behavior, Pre-coded, Robot behavior