Researchers at the Electronics and Telecommunication Research Institute (ETRI), Korea, develop a new deep learning model to produce engaging social behaviors like hugging or shaking someone’s hand in robots. Their research on the new deep learning model is presented in a paper on arXiv.
According to the researchers, deep learning techniques have shown exciting results in computer vision and natural language processing. They wanted to apply deep learning to social robotics to allow robots to learn social behavior from human interactions.
The deep-learning model’s architecture combines the sequence-to-sequence model introduced by Google researchers with generative adversarial networks (GAN). GAN is a machine learning model where two neural networks compete to become more accurate in the prediction. The architecture in the deep learning model is trained on the AIR-Act2Act dataset, which consists of 5000 human interactions occurring in 10 different scenarios.
Korean researchers have tested the new deep-learning model on a simulated version of Pepper, a humanoid robot. The model generates five non-verbal behaviors for robots: bowing, staring, shaking hands, blocking their face, and hugging. The new deep learning model will help to make social robots more adaptive and socially responsive. In the future, the new deep learning model can be tested on many robotics systems like home service robots, guide robots, delivery robots, educational robots, and more.