Reproduce “RING++: Roto-translation Invariant Gram for Global Localization on a Sparse Scan Map” on OpenResearch

About this paper

Xuecheng Xu, Sha Lu, Jun Wu, Haojian Lu, Qiuguo Zhu, Yiyi Liao, Rong Xiong, Yue Wang

RING++ introduces a learning-free framework for robust global localization using LiDAR scans, achieving roto-translation invariant place recognition and globally convergent pose estimation, enabling effective operation on sparse scan maps under significant viewpoint differences. The method demonstrates superior performance in multi-session and generalization scenarios, enhancing localization success rates and reducing errors compared to existing baselines.

Key insights

• The combination of Radon Transform and Discrete Fourier Transform magnitudes can systematically derive representations (TING, RING) with provable translation and roto-translation invariance.
• Multi-channel feature aggregation, when features satisfy specific equivariance/invariance criteria, can significantly improve place recognition performance without compromising the framework's core invariance properties.
• Globally convergent rotation and translation solvers, derived from equivariant representations, provide robust initial pose estimates crucial for the success of local refiners like ICP, especially under large viewpoint changes.

Summary by alphaXiv.