Deep Q-network for social robotics using emotional social signals

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Social robots require sophisticated interaction capabilities beyond pre-programmed responses.
• Current methods often overlook human emotional states, a crucial aspect of social cues.
• Developing autonomous social robots necessitates learning and adapting to human emotional feedback.

Purpose of the Study:

• To develop a Deep Reinforcement Learning (DRL) system for social robots to autonomously decide on appropriate behaviors based on human emotional states.
• To enhance human-robot interaction by enabling robots to perceive and react to human emotions.
• To create a novel DRL architecture, Social Robotics Deep Q-Network (SocialDQN), for socially intelligent robot behavior.

Main Methods:

• Utilized Deep Reinforcement Learning (DRL), combining Reinforcement Learning (RL) and Deep Learning (DL).
• Proposed the Social Robotics Deep Q-Network (SocialDQN) architecture for robot behavior learning.
• Extracted human emotional states from facial characteristics for robot perception, validated using the SimDRLSR simulator.

Main Results:

• The developed DRL system successfully learned to maximize rewards.
• The SocialDQN architecture enabled the robot to exhibit socially acceptable behavior.
• The system demonstrated satisfactory learning performance in simulated social interactions.

Conclusions:

• DRL is effective for creating social robots that can learn and adapt to human emotional states.
• The SocialDQN architecture provides a viable method for teaching robots socially appropriate responses.
• Integrating emotional state recognition significantly improves robot perception and interaction quality.