Research Analyzer
← Back ICRA 2026

SHeRLoc: Synchronized Heterogeneous Radar Place Recognition for Cross-Modal Localization

Hanjun Kim, Minwoo Jung, Wooseong Yang, Ayoung Kim

PDF

AI summary

Key figure (auto-extracted from paper)
SHeRLoc boosts heterogeneous radar place recognition recall@1 from below 0.1 to 0.9 by synchronizing radar cross-section data and aggregating features via optimal transport.
Radar Place Recognition Heterogeneous Sensors Cross-Modal Localization Optimal Transport Radar Cross-Section Autonomous Navigation

Problem

Current radar place recognition methods are confined to homogeneous sensors and fail to generalize across diverse radar types due to varying fields of view, noise patterns, and data formats, leaving cross-modal localization unresolved.

Approach

The method synchronizes spinning and 4D radar scans using radar cross-section polar matching and multi-view projections, then aggregates features hierarchically with an optimal transport network to generate rotationally robust descriptors for metric learning.

Key results

  • First deep network specifically designed for heterogeneous radar systems
  • RCS polar synchronization and multi-view projection pipeline bridges modality gaps
  • HOLMES feature aggregation overcomes speckle noise while preserving global scene context
  • Order-of-magnitude performance gain (recall@1 <0.1 to 0.9) across diverse and extreme conditions

Why it matters

Enables robust, cross-modal localization for autonomous vehicles and robots operating in adverse weather and dynamic environments where traditional sensors fail.

Abstract

Despite the growing adoption of radar in robotics, the majority of research has been confined to homogeneous sensors, overlooking the integration and cross-modality chal- lenges inherent in heterogeneous radar. This leads to significant difficulties in generalizing across diverse radar types, with modality-aware approaches that could leverage the complemen- tary strengths of heterogeneous radar remaining unexplored. To bridge these gaps, we propose SHeRLoc, the first deep network tailored for heterogeneous radar, which utilizes radar cross- section polar matching to align multimodal radar data. Our hi- erarchical optimal transport-based feature aggregation generates rotationally robust multi-scale descriptors. By employing FFT- similarity-based data mining and adaptive margin-based triplet loss, SHeRLoc enables FOV-aware metric learning. SHeRLoc achieves an order of magnitude improvement in heterogeneous radar place recognition, increasing recall@1 from below 0.1 to 0.9, and paves the way for cross-modal localization.

Index terms

Localization SLAM Range Sensing

Related papers