Reinforcement Learning Approaches in Social Robotics.

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Social robotics involves physically embodied robots designed for human interaction.
• Reinforcement learning (RL) is a machine learning paradigm where agents learn through trial-and-error interactions with an environment.
• The synergy between RL and social robotics is promising for developing robots capable of real-world human interaction.

Purpose of the Study:

• To provide a comprehensive survey of reinforcement learning approaches in social robotics.
• To analyze and categorize existing RL methods based on techniques and reward mechanisms.
• To explore the application of RL in real-world human-robot interaction scenarios.

Main Methods:

• Systematic review and categorization of reinforcement learning techniques used in social robotics.
• Analysis of reward function designs, including interactive, intrinsically motivated, and task-driven methods.
• Grouping of studies based on communication mediums for reward formulation and evaluation metrics.

Main Results:

• Categorization of RL approaches by method, reward design, and communication medium for reward formulation.
• Identification of three main reward themes: interactive RL, intrinsically motivated methods, and task performance-driven methods.
• Discussion of benefits, challenges, evaluation methods, and future research directions for RL in social robotics.

Conclusions:

• Reinforcement learning is a suitable framework for enabling social robots to learn optimal behaviors through user interaction.
• Effective reward mechanism design and communication strategies are crucial for successful RL in social robotics.
• Further research is needed to address real-world challenges and explore under-investigated RL approaches in this field.