Quick Setup of Force-Controlled Industrial Gluing Tasks Using Learning From Demonstration

Area of Science:

• Robotics
• Machine Learning
• Control Systems

Background:

• Programming robots for in-contact tasks, such as industrial gluing, is complex.
• Existing methods often require multiple demonstrations or lack adaptability.

Purpose of the Study:

• To present a novel framework for programming in-contact robot tasks using learning by demonstration.
• To develop a unified controller for seamless demonstration and execution phases.
• To adapt controller parameters to task geometry and human behavior for enhanced performance.

Main Methods:

• A unified controller structure supporting admittance control for demonstration and parallel force/position control for execution.
• Online estimation of task constraint geometry.
• Adaptive controller gains tuned to human demonstration behavior.
• A method for minimizing contact forces during motion mode switching.

Main Results:

• Successful demonstration of the framework on an industrial gluing task.
• Accurate online estimation of curved surface geometry.
• Adaptive controller achieved higher accuracy and shorter demonstration times compared to off-the-shelf controllers.
• Execution controller effectively reduced impact forces and maintained constant process force while adapting to surface geometry.

Conclusions:

• The proposed framework enables high-quality robot behavior programming for in-contact tasks with minimal demonstrations.
• The adaptive controller design enhances both the learning and execution phases of robot tasks.
• The system demonstrates robustness in handling transitions between free and in-contact motion.