MKT-GMM: A Motion Knowledge Transferring Framework for Robot Trajectory Adaptation to Variable Via-Points.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Learning from Demonstration (LfD) enables robots to learn tasks from human examples.
• Current LfD methods struggle with adaptability due to a lack of transferable motion knowledge for varying task constraints.

Purpose of the Study:

• To propose a new motion representation framework for enhanced robotic skill acquisition.
• To develop a Motion Knowledge Transferring Gaussian Mixture Model (MKT-GMM) for trajectory generalization.

Main Methods:

• Encoding demonstration trajectories into a Gaussian Mixture Model (GMM) to represent motion primitives.
• Transferring learned motion knowledge by adapting GMM parameters via affine transformations and constraint-error minimization.
• Utilizing Gaussian Mixture Regression (GMR) for smooth motion reconstruction and a pseudo via-point mechanism for robustness.

Main Results:

• The MKT-GMM framework successfully captures transferable motion knowledge from source tasks.
• The method enables reliable trajectory generalization to new, unseen tasks without additional demonstrations.
• Experiments on handwriting and pick-and-place tasks demonstrate effective skill transfer under varying configurations.

Conclusions:

• The proposed framework enhances the adaptability of LfD by enabling robots to generalize learned skills.
• This approach facilitates more robust and efficient robotic skill acquisition across diverse tasks and environments.