Research Analyzer
← Back ICRA 2026

TransLocNet: Cross-Modal Attention for Aerial-Ground Vehicle Localization with Contrastive Learning

Phu Pham, Damon Conover, Aniket Bera

PDF

AI summary

Key figure (auto-extracted from paper)
TransLocNet reduces aerial-ground localization error by up to 63% by fusing LiDAR and aerial imagery through cross-modal attention and contrastive learning.
aerial-ground localization cross-modal attention contrastive learning LiDAR bird's-eye view autonomous navigation

Problem

Existing cross-view localization methods struggle with large viewpoint shifts and modality gaps between ground-level LiDAR and overhead imagery, often relying on simplistic fusion that fails under noisy real-world conditions.

Approach

The framework projects ground LiDAR scans into a bird’s-eye-view representation and aligns them with aerial images using bidirectional cross-modal attention, followed by a probabilistic likelihood decoder and contrastive learning to enforce shared feature embeddings.

Key results

  • Reduces localization error by up to 63% and orientation error by up to 81% on CARLA and KITTI benchmarks
  • Achieves sub-meter and sub-degree accuracy in both synthetic and real-world settings
  • Introduces bidirectional cross-modal attention for robust feature fusion across large viewpoint gaps
  • Integrates contrastive learning to enforce shared semantic embeddings, improving cross-modal alignment

Why it matters

Enables reliable global positioning for autonomous vehicles and robots in GNSS-denied environments like urban canyons and dense forests.

Abstract

Aerial–ground localization is difficult due to large viewpoint and modality gaps between ground-level LiDAR and overhead imagery. We propose TransLocNet, a cross-modal attention framework that fuses LiDAR geometry with aerial semantic context. LiDAR scans are projected into a bird’s-eye- view representation and aligned with aerial features through bidirectional attention, followed by a likelihood map decoder that outputs spatial probability distributions over position and orientation. A contrastive learning module enforces a shared embedding space to improve cross-modal alignment. Experiments on CARLA and KITTI show that TransLocNet outperforms state-of-the-art baselines, reducing localization er- ror by up to 63% and achieving sub-meter, sub-degree accuracy. These results demonstrate that TransLocNet provides robust and generalizable aerial–ground localization in both synthetic and real-world settings.

Index terms

Aerial Systems: Perception and Autonomy Localization Deep Learning for Visual Perception

Related papers