Teaching robots social autonomy from in situ human guidance

Area of Science:

• Robotics
• Human-Robot Interaction
• Machine Learning

Background:

• Balancing robot autonomy and human control is crucial in social robotics, especially in sensitive fields like therapy and education.
• High autonomy risks system failure and ethical concerns, while human expertise is vital for accountability and optimal performance.

Purpose of the Study:

• To evaluate SPARC (supervised progressively autonomous robot competencies), an approach for robots to learn autonomous behaviors from human demonstrations.
• To address the challenge of integrating robot autonomy with human oversight in complex interaction scenarios.

Main Methods:

• A field study was conducted using SPARC, which employs online machine learning for progressive skill acquisition.
• The robot learned social policies in a child-tutoring setting through in situ human demonstrations and guidance.
• The system allowed for human supervision to be maintained when necessary.

Main Results:

• The robot successfully acquired legible and congruent social policies in a high-dimensional, complex environment.
• Effective learning was achieved with a limited number of human demonstrations.
• The SPARC approach demonstrated rapid learning of both social and domain-specific policies.

Conclusions:

• SPARC enables robots to learn complex social behaviors efficiently by leveraging human expertise.
• This paradigm facilitates the development of adaptable autonomous systems for challenging human-robot interaction scenarios.
• The generic nature of SPARC makes it applicable to diverse applications requiring nuanced human-robot collaboration.