Estimation of Flat Object Deformation Using RGB-D Sensor for Robot Reproduction.

Area of Science:

• Robotics
• Computer Vision
• Mechanical Engineering

Background:

• Estimating and replicating complex object deformations, particularly for flat objects like folded planes, remains a challenge in robotics.
• Accurate spatial understanding of an object's deformed state is crucial for robotic manipulation and reproduction.

Purpose of the Study:

• To develop a system capable of estimating the deformation process of a folded plane.
• To generate robot input data for replicating the estimated deformation on similar objects.
• To improve upon conventional methods for non-rigid point matching and plane detection.

Main Methods:

• Utilized RGB and depth data processing.
• Employed a weighted graph clustering method for non-rigid point matching and clustering.
• Implemented a refined region growing method for plane detection using a novel offset error.
• Introduced a sliding checking model to identify bending lines and inter-plane relationships.

Main Results:

• The system achieved an average angular error of approximately 1.59 degrees in deforming paper objects.
• This accuracy is comparable to the human eye's angular discrimination threshold.
• The system accurately captures spatial information of bending lines and planes for folded objects.

Conclusions:

• The developed system effectively estimates and reproduces the deformation of folded flat objects.
• The core techniques demonstrate significant improvements over existing methods.
• Future work aims to extend the system for robotic reproduction of general object deformations.