Socially-Aware Reinforcement Learning for Personalized Human-Robot Interaction

Hannes Ritschel

Proceedings of the 17th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2018)

This research in the context of Human-Robot Interaction explores how to tailor the robot behavior to the human's individual preferences in real-time. Algorithmically, Reinforcement Learning is the method of choice as it allows the robot to explore and learn autonomously. Instead of relying on task-related data, the proposed approach is primarily based on human social signals, which occur all the time and provide valuable information which cannot be extracted from the task itself. Including social signal data in the Reinforcement Learning framework enables us to adapt robot behavior depending on the current user behaviors without additional interaction.