A 6D Object Pose Estimation Algorithm for Autonomous Docking with Improved Maximal Cliques

Area of Science:

• Robotics and Computer Vision
• 3D Perception and Reconstruction

Background:

• Accurate 6D object pose estimation is crucial for autonomous docking operations.
• Maximal cliques-based methods often suffer from inefficiencies and inaccuracies.
• A need exists for faster and more precise pose estimation algorithms.

Purpose of the Study:

• To develop a fast and accurate 6D object pose estimation algorithm for autonomous docking.
• To overcome the limitations of existing maximal cliques-based pose estimation methods.
• To enhance the reliability of robotic perception systems.

Main Methods:

• Proposed a novel algorithm integrating feature space and spatial compatibility constraints for 6D pose estimation.
• Utilized Laplacian filtering to reduce graph size by resampling high-frequency signal nodes.
• Employed truncated Chamfer distance for evaluating candidate pose alignment and selected optimal transformation matrix.
• Refined pose estimation using a point-to-plane Iterative Closest Point (ICP) algorithm.

Main Results:

• Achieved high recall rates: 94.5% on 3DMatch, 62.2% on 3DLoMatch, and 99.1% on KITTI datasets.
• Demonstrated low errors on the autonomous docking dataset: 0.96° rotation and 5.82 cm localization error.
• The proposed algorithm significantly outperformed existing methods in experimental evaluations.

Conclusions:

• The developed algorithm offers a fast and accurate solution for 6D object pose estimation in autonomous docking.
• Integration of feature and spatial constraints effectively improves pose estimation performance.
• The approach validates its effectiveness and potential for real-world robotic applications.