Abstract

This paper presents an environment-agnostic automatic LiDAR-RGB-D calibration method for a Unitree G1 humanoid platform. The proposed pipeline reconstructs depth-camera observations into 3D point clouds and aligns them directly with LiDAR point clouds, avoiding the need for calibration targets or manual adjustment. The method applies explicit preprocessing, coarse simulated annealing, polish simulated annealing, a symmetric nearest-neighbor alignment objective, and final point-to-plane local refinement. Experimental results show that the coarse simulated annealing stage provides most of the calibration correction, while the polish stage and local refinement further improve convergence and alignment accuracy. Overall, this framework demonstrates a practical target-free approach for automatic 3D LiDAR-RGB-D alignment on humanoid robotic platforms.

Document Type

Article

Author's School

McKelvey School of Engineering

Author's Department

Electrical and Systems Engineering

Class Name

Electrical and Systems Engineering Undergraduate Research

Language

English (en)

Date of Submission

4-24-2026

Share

COinS