Learning Shared Control by Demonstration for Personalized Wheelchair Assistance.

Area of Science:

• Robotics
• Human-Robot Interaction
• Machine Learning

Background:

• Designing personalized assistance for users in robotics is a significant challenge.
• Current methods often require extensive training data or predefined rules.

Purpose of the Study:

• To develop a method for robots to learn shared control policies from single human demonstrations.
• To enable robots to provide personalized and adaptive assistance to users.

Main Methods:

• Training a Gaussian process (GP) regression model on a single human demonstration.
• Learning a shared control policy to regulate the level of robot assistance based on user actions and environmental state.
• Comparing the learned policy against human assistance in an experimental study.

Main Results:

• The learned policy successfully emulated human shared control, matching assistance levels and locations.
• User effort was comparable to human-human assistance settings.
• User joystick movements became less jerky, while robot commands increased in jerkiness.
• Robotic assistance slightly increased task completion time but maintained similar distances to obstacles as human assistance.

Conclusions:

• The proposed one-shot learning method effectively enables robots to provide personalized shared control.
• The system demonstrates the potential for robots to learn and adapt assistance based on minimal human input.
• This approach offers a promising direction for developing more intuitive and effective assistive robots.