A study of interactive robot architecture through the practical implementation of conversational android.

Area of Science:

• Human-Robot Interaction
• Artificial Intelligence
• Natural Language Processing

Background:

• Daily dialogue systems differ from task-oriented ones, lacking clear purpose or information.
• Generating natural conversation requires adapting dialogue management strategies.
• Existing systems are not optimized for open-ended, human-like daily interactions.

Purpose of the Study:

• To determine the optimal dialogue system architecture for natural daily conversations.
• To implement an autonomous android robot for demonstrating daily dialogue capabilities.
• To explore the design principles for human-like conversational agents.

Main Methods:

• Defined 'desire' parameters for the android to guide conversation.
• Developed a dialogue management system that adapts internal states based on desire and human behavior.
• Implemented a system for selecting contextually appropriate dialogue topics.
• Conducted experiments involving first-time interactions between the robot and human partners.

Main Results:

• The developed android successfully maintained daily dialogue for approximately 10 minutes in initial encounters.
• A multimodal Turing test indicated that 50% of participants perceived the android's behavior as humanlike, with some suspecting remote control.
• The implemented system demonstrated effectiveness in facilitating natural, extended conversations.

Conclusions:

• The proposed system construction method is effective for realizing natural daily dialogue.
• The study provides insights into the necessary system architecture for autonomous conversational androids.
• Further research can build upon this architecture to enhance human-robot conversational capabilities.