An enhanced teaching interface for a robot using DMP and GMR.

Area of Science:

• Robotics
• Human-Robot Interaction
• Machine Learning

Background:

• Robot programming often requires specialized skills.
• Teleoperation and learning from demonstration offer intuitive alternatives.
• Generalizing movements from limited data is a key challenge.

Purpose of the Study:

• To develop an enhanced teaching interface for robots.
• To enable accurate movement modeling and generalization from human demonstrations.
• To validate the approach on both Baxter and KUKA iiwa robot platforms.

Main Methods:

• Movement data captured using Kinect v2 sensor.
• Dynamic Movement Primitives (DMP) for movement modeling.
• Dynamic Time Warping (DTW) and Gaussian Mixture Models (GMM) for data preprocessing and pattern generation.
• Gaussian Mixture Regression (GMR) for trajectory synthesis.

Main Results:

• The system successfully captured and modeled robot movements.
• The proposed method achieved accurate generalization across multiple demonstrations.
• Reduced position errors were observed in the synthesized 3D trajectories.
• The interface was effectively tested on both Baxter and KUKA iiwa robots.

Conclusions:

• The enhanced teaching interface provides an effective method for robot programming.
• The combination of DMP, DTW, GMM, and GMR enables robust movement learning and generalization.
• The approach is versatile and applicable to different robot platforms and tasks.