6DoF assembly pose estimation dataset for robotic manipulation

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Robotic assembly demands advanced cognition and dexterity.
• Current perception tools offer robot capabilities like detection and pose estimation.
• Existing datasets lack essential information for robotic assembly pose estimation.

Purpose of the Study:

• To address the limitations of current datasets for robotic assembly pose estimation.
• To provide a dataset with crucial information for training perception models.
• To facilitate the development of novel algorithms for robotic assembly manipulation.

Main Methods:

• The study presents a dataset of 2 simulated object assembly scenes.
• The dataset includes RGB-D images, 3D mesh files, and ground truth assembly poses.
• The dataset is designed as an extension for the State-of-the-Art BOP format.

Main Results:

• The new dataset enables adaptation of existing perception models for assembly tasks.
• It supports the development of novel algorithms for estimating assembly pose.
• The dataset provides object 3D meshes and point clouds in canonical view frames.

Conclusions:

• The developed dataset is crucial for advancing robotic assembly capabilities.
• It bridges the gap in existing datasets for pose estimation in robotic manipulation.
• This resource will accelerate research in robot perception and assembly tasks.