GAP-GDRNet: Geometry-aware monocular 6D pose estimation for spacecraft using synthetic geometric supervision

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

GAP-GDRNet, a geometry-aware attention-enhanced framework, addresses monocular RGB-based 6D pose sensing for spacecraft. The method follows the geometry-guided direct regression paradigm of GDR-Net and modifies two points in the pipeline: an attention-based feature refinement (AFR) module is placed before dense geometric prediction, and a patch-level geometric self-attention (PGSA) module is inserted into Patch-PnP. AFR reinforces global spacecraft structure and local weak-texture cues, while PGSA relates downsampled geometric patches before final pose regression. Training utilizes a Blender-based synthetic dataset of 50,000 images from one CAD model, providing target masks, dense model-coordinate maps, and 6D pose labels. GAP-GDRNet achieves a rotation error of 1.96°, translation error of 0.0165 m, and ADD@0.02 m accuracy of 95.16% on this dataset, outperforming the reproduced GDR-Net baseline by 3.88 percentage points while running at 35.97 FPS. Supplementary evaluations on T-LESS and LM-O datasets also show BOP AR gains of 6.8 and 3.1 percentage points, respectively.

Key takeaway

For robotics engineers developing pose estimation for non-cooperative spacecraft, GAP-GDRNet offers a robust approach. You should consider integrating attention-based feature refinement (AFR) and patch-level geometric self-attention (PGSA) into your vision pipelines. This enhances accuracy, especially under challenging conditions like weak illumination or heavy occlusion, while maintaining a practical inference speed of 35.97 FPS. Explore synthetic data generation to overcome real-world data scarcity for training.

Key insights

GAP-GDRNet enhances monocular 6D spacecraft pose estimation by integrating attention-based feature refinement and patch-level self-attention into the GDR-Net pipeline.

Principles

• Spacecraft pose estimation benefits from global and local feature refinement.
• Patch-level attention improves geometric evidence aggregation.
• Synthetic data is crucial for training robust spacecraft pose estimators.

Method

GAP-GDRNet uses AFR for global/local feature enhancement before dense geometric prediction and PGSA within Patch-PnP to model patch-level geometric relations for 6D pose regression.

In practice

• Use attention modules to refine features in low-texture environments.
• Employ synthetic data generation for hard-to-label scenarios.
• Integrate self-attention for cross-region dependency modeling.

Topics

• 6D Pose Estimation
• Spacecraft Rendezvous
• Monocular Vision
• Attention Mechanisms
• Synthetic Data Generation
• GDR-Net

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• GAP-GDRNet: Geometry-Aware Monocular Visual Pose Sensing on a Single-Target Synthetic Spacecraft Dataset — GAP-GDRNet enhances 6D spacecraft pose sensing by integrating attention mechanisms for robust geometric feature extraction.
• NeRF-based Spacecraft Reconstruction from Close-Range Monocular Imagery Under Illumination Variability and Pose Uncertainty — NeRFs can be extended with per-image embeddings and pose corrections to robustly reconstruct spacecraft under challenging conditions.
• OffNadirLoc: Benchmark and Framework for Challenging UAV-to-Satellite Geo-Localization under Large Off-Nadir Views — Large off-nadir UAV-to-satellite geo-localization requires structure-aware feature weighting and multi-view coherent learning for robustness.
• Semantic Prior Guided One-View 6D Pose Estimation for Novel Objects — OneViewAll enables accurate, efficient 6D pose estimation for novel objects using a single RGB-D view and hierarchical semantic priors.
• GS-RealBlur: A Flexible Data Acquisition Framework for Real-World Image Deblurring — GS-RealBlur enables flexible, realistic real-world deblurring data acquisition via 3D reconstruction and blur-aware pose refinement.
• Honey, I Shrunk the Arc de Triomphe! — A new dataset and method mitigate scale-collapse in monocular depth estimation for distant, unconstrained scenes.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.