Monocular Visual Measurement System Uncertainty Analysis and One-Step End-End Estimation Upgrade.

Area of Science:

• Robotics
• Computer Vision
• Manufacturing Engineering

Background:

• Monocular visual measurement and vision-guided robotics are crucial in automated manufacturing, especially aerospace assembly.
• Multi-source error accumulation from visual measurement, hand-eye calibration, and robot calibration degrades final assembly accuracy.

Purpose of the Study:

• To develop an uncertainty analysis method for monocular visual measurement systems in assembly pose estimation.
• To propose a direct, one-step solution for robot and hand-eye calibration problems informed by uncertainty analysis.
• To construct a high-performance, end-to-end pose estimation convolutional neural network (OECNN).

Main Methods:

• Developed an uncertainty analysis method for monocular visual measurement systems, detailing uncertainty propagation paths and input values.
• Applied nonlinear mapping estimation for a direct, one-step solution to robot and hand-eye calibration challenges.
• Constructed and validated the one-step end-to-end pose estimation convolutional neural network (OECNN).

Main Results:

• The OECNN directly maps target object pose variation to positioner drive volume variation.
• Uncertainty analysis provided key insights for improving assembly pose estimation precision.
• Experimental validation confirmed the high accuracy and applicability of the proposed one-step method.

Conclusions:

• The proposed uncertainty analysis methodology offers a valuable reference for complex system uncertainty analysis.
• The one-step end-to-step pose estimation method significantly enhances accuracy in automated assembly tasks.
• The approach is particularly suitable for high-precision applications like aircraft assembly using vision-guided robots.