Convolutional neural network-based pose mapping estimation as an alternative to traditional hand-eye calibration.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Industrial robot automation relies heavily on vision systems.
• Accurate hand-eye calibration is essential for robot end-effector and camera relationship determination.
• Traditional calibration methods face limitations with parallel robots' motion range and accuracy.

Purpose of the Study:

• To develop a robust hand-eye calibration method for parallel robots.
• To address accuracy issues in traditional calibration techniques for robots with limited motion.
• To enable precise vision-guided automated assembly tasks.

Main Methods:

• Proposed a pose, nonlinear mapping estimation method for hand-eye calibration.
• Developed a 1-D pose estimation convolutional neural network (PECNN).
• PECNN enables end-to-end mapping of target object pose variations to robot end pose variations.

Main Results:

• The PECNN-based hand-eye calibration method demonstrated high accuracy.
• Experimental validation confirmed the method's effectiveness.
• The approach is suitable for vision-guided parallel robots in automated assembly.

Conclusions:

• The proposed hand-eye calibration method is accurate and applicable to parallel robots.
• The method shows promise for enhancing automated assembly with vision-guided robots.
• The technique is adaptable to most parallel and tandem robots.