Learning compliant manipulation through kinesthetic and tactile human-robot interaction.

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Robot Learning from Demonstration (RLfD) is crucial for practical robot applications.
• Current RLfD methods often focus on kinematics, using stiff position control.
• Stiff control is insufficient for tasks requiring contact or response to perturbations.

Purpose of the Study:

• To address the challenge of teaching robots task-specific compliance in RLfD.
• To develop interfaces enabling human teachers to demonstrate compliance variations through physical interaction.
• To enhance robot adaptability for complex, real-world tasks.

Main Methods:

• Developed novel interfaces for human-guided compliance adjustment during robot task execution.
• Utilized physical interaction to convey desired compliance changes to the robot.
• Validated the approach on 7 DoF Barrett WAM and KUKA LWR robot manipulators.

Main Results:

• Successfully demonstrated the ability of robots to learn and adapt compliance based on human physical guidance.
• Validated the effectiveness of the proposed interfaces in two distinct experimental setups.
• Gathered user feedback on the usability of the approach for non-expert users.

Conclusions:

• The proposed method effectively allows robots to learn variable compliance for tasks requiring physical interaction.
• Human physical guidance is a viable and intuitive method for teaching robot compliance.
• The approach shows promise for making robots more adaptable and useful in diverse daily-life scenarios.