A novel trajectory learning method for robotic arms based on Gaussian Mixture Model and k-value selection algorithm.

Area of Science:

• Robotics
• Machine Learning
• Artificial Intelligence

Background:

• Gaussian Mixture Models (GMMs) are effective for robotic arm trajectory imitation learning.
• A key challenge is initializing GMMs, specifically determining the optimal number of Gaussian kernels (k-value).
• Suboptimal k-value selection leads to reduced model performance in trajectory generation.

Purpose of the Study:

• To propose a novel trajectory learning method for robotic arms.
• To accurately determine the optimal k-value for Gaussian Mixture Models.
• To enhance the accuracy and efficiency of robotic arm trajectory generation.

Main Methods:

• A novel k-value selection algorithm combining the elbow method, exponential functions, correction terms, and weight adjustments.
• K-means clustering for GMM parameter initialization using the optimal k-value.
• Expectation-Maximization algorithm for refining GMM parameters.
• Gaussian Mixture Regression for trajectory generation.

Main Results:

• The proposed method significantly improves robotic arm trajectory accuracy.
• Demonstrated over 15% improvement in accuracy compared to traditional GMM approaches.
• Reduced Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) in trajectory generation.

Conclusions:

• The novel k-value selection algorithm effectively optimizes GMM initialization for robotic arm trajectory learning.
• The proposed method enhances both accuracy and efficiency in generating robotic arm trajectories.
• This approach offers a promising solution for advanced robotic manipulation tasks.