SMARTPOSE: Development of a Sample-Efficient, Model-Agnostic, Robust, Two-Stage POSE Estimator for Unknown Satellites
Yash Kishorbhai Joshi, Suresh Sundaram
AI summary
Problem
Accurate relative pose estimation for unknown, non-cooperative satellites is critical for on-orbit servicing but hindered by the lack of 3D CAD models, scarce training data, and extreme lighting variations. Existing model-free methods lack the accuracy and robustness required for precision tasks, while model-based methods fail without geometric priors.
Approach
The framework first uses an appearance-aware 3D reconstruction module to build a 3D model from limited images and generate diverse synthetic training data for a CNN pose predictor. It then refines the initial pose estimate in a second stage by iteratively minimizing photometric error through differentiable rendering conditioned on the target image's appearance.
Key results
- Novel appearance-aware 3D reconstruction module that adapts to extreme lighting and reflective surfaces
- Two-stage predictor-corrector framework enabling sample-efficient training and high-accuracy pose refinement
- State-of-the-art accuracy and robustness on SPEED+ and URSO Soyuz datasets, reducing rotation error by 80%
- Strong generalization across simulation-to-reality domain shifts without requiring prior 3D CAD models
Why it matters
Enables reliable autonomous close-proximity operations like debris removal and servicing for unknown satellites where traditional methods fail.
Abstract
Accurate relative pose estimation is critical for autonomous close-proximity satellite operations, such as on- orbit servicing and debris removal. However, this task remains highly challenging for unknown, non-cooperative targets due to the unavailability of geometric priors, scarce training data, and the extreme variations in lighting and backgrounds inherent to the space environment. Existing approaches for pose estimation can be categorized into: 1. model-based methods, which achieve higher accuracy but assume access to 3D CAD models of target satellites; and 2. model-free methods, which can be used for unknown satellites but typically suffer from reduced accuracy and robustness. To bridge this gap, this paper introduces a sample-efficient, model-free pose estimation framework that achieves high accuracy and robustness for unknown satellites while demonstrating strong generalization under the uncertain conditions inherent to the space environment. The proposed method utilizes a novel appearance aware 3D reconstruction to generate satellite model from images accounting for different lighting conditions during the training. This model is then used to generate a large, diverse dataset to train a pose predictor network (stage 1). The predicted pose is refined using the 3D reconstruction by utilizing the appearance information of the target image along with differentiable rendering (stage 2). Evaluated across SPEED+ and URSO Soyuz datasets, our approach achieves state-of-the-art accuracy and proves highly robust to test-time domain shifts, notably reducing rotation error by 80% on the challenging URSO Soyuz dataset.