Research Analyzer
← Back ICRA 2026

Learning to Anchor Visual Odometry: KAN-Based Pose Regression for Planetary Landing

Leizheng Shu

PDF

AI summary

Key figure (auto-extracted from paper)
Fusing a lightweight Kolmogorov–Arnold Network with visual odometry eliminates drift and achieves globally consistent, real-time 6-DoF pose estimation for lunar landing, cutting translation and rotation errors by up to 45%.
Visual localization Kolmogorov–Arnold Networks planetary landing visual odometry pose regression bundle adjustment

Problem

Visual odometry accumulates unbounded drift over long trajectories, while traditional absolute visual localization fails in texture-sparse, low-light lunar terrain, creating a critical gap for reliable autonomous planetary landing.

Approach

The framework tightly couples high-frequency monocular visual odometry with sparse absolute pose anchors generated by a Kolmogorov–Arnold Network regressor, fusing them via constrained bundle adjustment to correct drift while preserving local motion precision.

Key results

  • KAN-based pose regressor achieves high accuracy with remarkable parameter efficiency
  • Hybrid VO–absolute localization yields globally consistent real-time trajectories at ≥15 FPS
  • Tailored mask data augmentation improves robustness to severe sensor occlusions
  • Reduces average translation and rotation error by 32% and 45% respectively on synthetic and real lunar datasets

Why it matters

Enables reliable, resource-constrained autonomous navigation for future lunar and planetary landing missions where traditional sensors fail.

Abstract

Accurate and real-time 6-DoF localization is mission-critical for autonomous lunar landing, yet existing ap- proaches remain limited: visual odometry (VO) drifts unbound- edly, while map-based absolute localization fails in texture- sparse or low-light terrain. We introduce KANLoc, a monoc- ular localization framework that tightly couples VO with a lightweight but robust absolute pose regressor. At its core is a Kolmogorov–Arnold Network (KAN) that learns the complex mapping from image features to map coordinates, producing sparse but highly reliable global pose anchors. These anchors are fused into a bundle adjustment framework, effectively canceling drift while retaining local motion precision. KANLoc delivers three key advances: (i) a KAN-based pose regressor that achieves high accuracy with remarkable parameter efficiency, (ii) a hybrid VO–absolute localization scheme that yields globally consistent real-time trajectories (≥15 FPS), and (iii) a tailored data augmen- tation strategy that improves robustness to sensor occlusion. On both realistic synthetic and real lunar landing datasets, KANLoc reduces average translation and rotation error by 32% and 45%, respectively, with per-trajectory gains of up to 45%/48%, outperforming strong baselines.

Index terms

Space Robotics and Automation Localization Deep Learning for Visual Perception

Related papers