In situ bidirectional human-robot value alignment

Area of Science:

• Human-Robot Interaction
• Artificial Intelligence
• Cognitive Science

Background:

• Social coordination relies on bidirectional communication, with individuals acting as both listeners and speakers.
• Misalignment in human-robot values can hinder collaborative performance in complex tasks.
• Robots must effectively infer human intent and explain their decision-making processes.

Purpose of the Study:

• To investigate fostering effective bidirectional communication between humans and robots.
• To achieve value alignment in human-robot teams through explainable AI (XAI).
• To develop a system where robots predict user values and explain their decisions.

Main Methods:

• Proposed an XAI system integrating a cooperative communication model to infer human values from in situ feedback.
• Simulated human mental dynamics and used graphical models to predict optimal explanations.
• Conducted psychological experiments to validate the computational framework.

Main Results:

• Demonstrated that real-time mutual understanding between humans and robots is achievable.
• Showcased the effectiveness of a learning model based on bidirectional communication.
• Validated the core components of the proposed computational framework.

Conclusions:

• Bidirectional communication is key for human-robot value alignment.
• The developed interaction framework supports communicative XAI systems.
• This approach advances future human-machine teaming systems.