Robot Learning from Humans in Everyday Life Scenarios


Robots need to be able to learn about novel environments and acquire new capabilities during deployment. Robot learning from humans is a paradigm to enable the human user to teach robots certain information and skills without programming knowledge. In this chapter, we provide an overview of this domain and present some of our work as concrete examples. First, we address grounded language learning with the goal to create connections between words and references (e.g., objects, locations) in social environments. We present our incremental word learning systems using the Pepper robot. Following that, we introduce to learning low-level actions from demonstrations. We present our systems with an industrial robotic arm and a dexterous robotic hand. Then, we address the role of the teacher in the learning process. We investigate the human factors that are important for facilitating the learning process and present the results of our user studies. We conclude with open challenges and opportunities for further research.

Chapter in Trust in Robots published by TU Wien Academic Press