Register Any Point: Scaling 3D Point Cloud Registration by Flow Matching

2025-12-01

AI Summary

Point cloud registration aims to align unposed multi-view point clouds into a common coordinate system; however, conventional correspondence-based methods suffer from poor generalization under low-overlap, cross-scale, and cross-sensor conditions. This paper introduces the first end-to-end conditional generative framework for registration based on flow matching, which directly learns per-point velocity fields to implicitly model continuous motion trajectories toward the target coordinate system—eliminating the need for explicit correspondences. The method integrates a lightweight local feature extractor with test-time rigid constraints to balance computational efficiency and geometric consistency. It achieves state-of-the-art performance on challenging benchmarks including low-overlap and multi-view registration. Furthermore, it demonstrates strong cross-domain generalization in real-world applications such as relocalization, multi-robot SLAM, and multi-temporal map fusion, significantly outperforming prior approaches in robustness and adaptability.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Motion & Tracking
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Point cloud registration aligns multiple unposed point clouds into a common frame, and is a core step for 3D reconstruction and robot localization. In this work, we cast registration as conditional generation: a learned continuous, point-wise velocity field transports noisy points to a registered scene, from which the pose of each view is recovered. Unlike previous methods that conduct correspondence matching to estimate the transformation between a pair of point clouds and then optimize the pairwise transformations to realize multi-view registration, our model directly generates the registered point cloud. With a lightweight local feature extractor and test-time rigidity enforcement, our approach achieves state-of-the-art results on pairwise and multi-view registration benchmarks, particularly with low overlap, and generalizes across scales and sensor modalities. It further supports downstream tasks including relocalization, multi-robot SLAM, and multi-session map merging. Source code available at: https://github.com/PRBonn/RAP.

Problem

• Aligns multiple unposed point clouds into a common frame.
• Directly generates registered point clouds via conditional generation.
• Achieves state-of-the-art results with low overlap and cross-modality generalization.

Innovation

• Uses conditional generation for point cloud registration
• Directly generates registered point cloud via flow matching
• Employs lightweight feature extractor and rigidity enforcement