Keynote

1. Human-Robot interaction with Large Language Models
2. Trustworthy human-robot interaction

Is the autonomous social robot within reach?

Abstract
The recent breakthroughs in Generative AI offer fantastic opportunities to research novel concepts for intelligent embodied agents. In this keynote, I will introduce our findings in designing interactive robot agents that learn from humans, and that can exploit their acquired knowledge in different situations. We designed a Virtual Playground, in which we conducted user studies to understand the efficiency of robot curiosity as well as of multi-modal cues with respect to an explainable interaction. To close the gap towards behavior generation, I will introduce recent research in exploiting Large Language Models (LLMs) for robot task and motion planning. We combined reasoning, planning, and motion generation, and introduced a novel concept for correcting errors during planning and execution. I’ll show several results both in simulations and real-world tasks for tasks like block arrangement, cocktail, and pizza preparation. I will then discuss our recent concept of “Attentive Support”, in which we made the step from LLM-based autonomous problem-solving capabilities to human-robot group constellations and conclude with my view of interesting future research questions.

Biography
Michael Gienger received the diploma degree in Mechanical Engineering from the Technical University of Munich, Germany, in 1998. Until 2003, he was research assistant at the Institute of Applied Mechanics of the TUM and received his PhD degree with a dissertation in the field of robotics. After this, he joined the Honda Research Institute Europe in Germany, where he currently holds a position as a Chief Scientist and Competence Group Leader in the field of robotics. His research interests include mechatronics, robotics, whole-body control, imitation learning, and human-robot interaction.

Trustworthy human-robot interaction

Abstract

Biography