Human-Robot Mutual Adaptation in Shared Autonomy.

Area of Science:

• Human-Robot Interaction
• Robotics
• Artificial Intelligence

Background:

• Shared autonomy (SA) combines human input with robot control for task completion.
• Improving SA performance requires robots to guide users effectively while maintaining trust.
• Challenges in SA include unobservable human/robot goals and only robot actions altering the environment.

Purpose of the Study:

• To develop and evaluate a principled human-robot mutual adaptation formalism for SA.
• To enable robots to adapt to human users' preferences and adaptability levels.
• To enhance team performance and user trust in SA settings.

Main Methods:

• Integrated a bounded-memory human adaptation model into a partially observable stochastic decision model.
• Developed a formalism enabling robot adaptation to adaptable humans and compliance with stubborn users.
• Compared the proposed approach against a robot strictly following user preferences in human subject experiments.

Main Results:

• The mutual adaptation formalism significantly improved human-robot team performance.
• User trust in the robot was maintained at high levels.
• The proposed method outperformed the baseline approach of the robot strictly following user preferences.

Conclusions:

• Human-robot mutual adaptation is a viable formalism for enhancing SA.
• Robots can effectively guide users towards better strategies while preserving trust.
• This approach offers a more effective SA strategy than simple user preference adherence.