LRBO2: Improved 3D Vision Based Hand-Eye Calibration for Collaborative Robot Arm

2025-04-30

AI Summary

To address the inefficiency of conventional collaborative robot hand-eye calibration—namely, its reliance on external calibration targets, multiple robotic motions, and prolonged execution time—this paper proposes a target-free, single-pose point-cloud-based hand-eye calibration method. The approach leverages point-cloud registration to synthesize simulation data and jointly employs Iterative Closest Point (ICP) and deep feature matching to estimate the rigid-body transformation between the camera and robot end-effector. Furthermore, cross-platform kinematic modeling and robust optimization ensure compatibility across 14 robot models from nine major manufacturers, including KUKA, Universal Robots, and Franka Emika. Experimental results demonstrate that the method achieves commercial-grade accuracy while completing calibration in only a few seconds—substantially outperforming traditional approaches in both speed and practicality. The source code is publicly available.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Biometrics, Face, Gesture & Pose
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Hand-eye calibration is a common problem in the field of collaborative robotics, involving the determination of the transformation matrix between the visual sensor and the robot flange to enable vision-based robotic tasks. However, this process typically requires multiple movements of the robot arm and an external calibration object, making it both time-consuming and inconvenient, especially in scenarios where frequent recalibration is necessary. In this work, we extend our previous method, Look at Robot Base Once (LRBO), which eliminates the need for external calibration objects such as a chessboard. We propose a generic dataset generation approach for point cloud registration, focusing on aligning the robot base point cloud with the scanned data. Furthermore, a more detailed simulation study is conducted involving several different collaborative robot arms, followed by real-world experiments in an industrial setting. Our improved method is simulated and evaluated using a total of 14 robotic arms from 9 different brands, including KUKA, Universal Robots, UFACTORY, and Franka Emika, all of which are widely used in the field of collaborative robotics. Physical experiments demonstrate that our extended approach achieves performance comparable to existing commercial hand-eye calibration solutions, while completing the entire calibration procedure in just a few seconds. In addition, we provide a user-friendly hand-eye calibration solution, with the code publicly available at github.com/leihui6/LRBO2.

Problem

• Eliminates need for external calibration objects
• Improves hand-eye calibration speed and convenience
• Validates method across multiple collaborative robot brands

Innovation

• Eliminates need for external calibration objects
• Uses point cloud registration for alignment
• Achieves fast calibration in seconds