Model-Free Transformer Framework for 6-DoF Pose Estimation of Textureless Tableware Objects.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Estimating the six-degree-of-freedom (6-DoF) pose of tableware is crucial for robotic manipulation in restaurants.
• Conventional methods struggle with textureless, uniform objects like plates and bowls, hindering autonomous grasping.
• Existing approaches often rely on texture cues or specific 3D models, which are not always applicable.

Purpose of the Study:

• To develop a model-free and texture-free 6-DoF pose estimation framework for robotic tableware manipulation.
• To overcome the limitations of traditional pose estimation methods in restaurant environments.
• To enable reliable autonomous grasping and collection of tableware by service robots.

Main Methods:

• A transformer encoder architecture is utilized for pose estimation, processing geometry-based features from depth images.
• Features include surface vertices and rim normals, providing strong structural priors without texture.
• The pipeline integrates object detection/segmentation with a pretrained video foundation model and transformer-based pose prediction.

Main Results:

• The proposed method achieved an average rotational error of 3.53 degrees and a translational error of 13.56 mm across ten tableware types.
• Real-world deployment on a mobile robot platform demonstrated successful autonomous recognition and collection of tableware.
• The geometry-driven approach proved practical for service robotics applications.

Conclusions:

• The novel geometry-based framework effectively addresses the challenge of 6-DoF pose estimation for textureless tableware.
• This approach enhances the capabilities of service robots in restaurant environments.
• The model-free and texture-free method offers a practical solution for autonomous robotic manipulation.