Comparison of Graph Fitting and Sparse Deep Learning Model for Robot Pose Estimation.

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Real-time robot arm tracking is crucial for automation and industrial applications.
• Accurate measurement of robot state (angles) is essential for precise control.
• Existing methods may face challenges with real-time performance and accuracy.

Purpose of the Study:

• To develop a simple and robust computer vision system for real-time robot arm tracking.
• To evaluate the performance of different machine learning approaches for robot pose estimation.
• To leverage RGB-D camera data for accurate robot state measurement.

Main Methods:

• The system integrates image preprocessing with machine learning techniques.
• Two machine learning approaches were compared: point cloud fitting and sparse 3D convolutional neural network (CNN) fitting.
• Sparse CNNs were utilized for processing point cloud data transformed into sparse images.

Main Results:

• The developed system enables real-time robot tracking.
• The sparse CNN approach demonstrated effectiveness in processing 3D depth data.
• Tracking accuracy was found to be comparable to the inherent accuracy of the depth sensor.

Conclusions:

• The proposed computer vision system offers a robust solution for real-time robot arm tracking.
• The use of sparse CNNs provides an efficient method for handling 3D point cloud data in robot tracking.
• The system achieves high accuracy, making it suitable for practical robotic applications.