InLiER: Learning-Free Heterogeneous LiDAR Place Recognition via Intermediate Mixed-Radix Structural Keypoint Tokenization

2026-07-18

AI Summary

This work addresses the challenge of place recognition under heterogeneous LiDAR configurations—characterized by varying fields of view, resolutions, and scanning patterns—where existing descriptors suffer from strong sensor coupling that limits performance. The authors propose a training-free, efficient pipeline that encodes local 3D geometry into compact (<2 KB), rotation-invariant tokens via height-sliced keypoint extraction and mixed-radix tokenization. These tokens enable a unified representation compatible across diverse LiDAR types. Place recognition is further refined through yaw-aware re-ranking followed by token-guided 6-DoF geometric verification. Without requiring any learning, the method achieves state-of-the-art performance among handcrafted approaches on the HeLiPR benchmark and real-world scenarios, outperforming learning-based baselines in most cross-device configurations.

Abstract

LiDAR place recognition supports loop closure, relocalization, and multi-agent map management. As robotic platforms increasingly combine LiDARs with different fields of view, resolutions, and scanning patterns, existing descriptors degrade because they are tightly coupled to sensor-specific characteristics. We present InLiER, a learning-free pipeline based on an intermediate tokenization step. Height-sliced keypoints from structural elements receive mixed-radix token IDs encoding height, radial distance, local shape, and azimuth from local 3D geometry, in a compact sub-2KB representation. The same vocabulary is reorganized across three retrieval stages: height-ceiling histogram intersection for fast rotation-invariant shortlisting, binary bitmask alignment for yaw estimation and reranking, and token-guided geometric verification for 6-DoF pose estimation. InLiER achieves state-of-the-art performance on the HeLiPR dataset and in real-world field experiments, among modern handcrafted methods and outperforms the learning-based baseline on most cross-sensor configurations.

Problem

• LiDAR place recognition
• heterogeneous sensors
• cross-sensor
• loop closure
• relocalization

Innovation

• learning-free
• heterogeneous LiDAR
• mixed-radix tokenization
• place recognition
• rotation-invariant retrieval