TLDR: This paper introduces an adaptive Unscented Kalman Filter (UKF) framework for robust relative pose estimation of uncooperative spacecraft like ENVISAT. It features a dual-adaptive strategy that dynamically tunes both measurement noise (R) and process noise (Q) covariances online. This improves estimation accuracy and consistency, especially during challenging conditions like measurement outages (e.g., eclipses), by integrating a CNN for marker detection and advanced noise adaptation techniques.
Space exploration and operations are becoming increasingly complex, especially when it comes to interacting with uncooperative objects like defunct satellites or space debris. Accurately determining the relative position and orientation (known as relative pose) of these objects is a critical challenge for missions such as Active Debris Removal (ADR), on-orbit servicing, and space situational awareness. Traditional methods often struggle with the unpredictable dynamics, severe visibility issues, and uncertain sensor performance inherent in these scenarios.
A new research paper, “Adaptive Relative Pose Estimation Framework with Dual Noise Tuning for Safe Approaching Maneuvers,” introduces a comprehensive solution to this problem. Authored by Batu Candan and Simone Servadio, this work presents an innovative pipeline that combines advanced computer vision techniques with an adaptive nonlinear filtering approach to enhance the robustness and accuracy of relative pose estimation.
The Challenge of Space Debris and Uncooperative Targets
The European Space Agency’s (ESA) ENVISAT satellite, a large and complex derelict spacecraft, serves as a key benchmark target for ADR missions. Its non-cooperative nature and tumbling motion make precise relative pose estimation incredibly difficult. While recent advancements in vision-based navigation, particularly with Convolutional Neural Networks (CNNs), have improved object detection in space imagery, these visual outputs are still prone to significant uncertainty due to sensor limitations, jitter, and occlusions. Fusing these noisy and intermittent detections into a reliable navigation solution remains a central hurdle.
A Dual-Adaptive Filtering Approach
The core of this framework lies in its use of an Unscented Kalman Filter (UKF), chosen for its ability to handle nonlinear relative dynamics without linearizing the system. What sets this approach apart is its novel dual adaptive strategy, which dynamically tunes two critical noise parameters within the UKF: the measurement noise covariance (R) and the process noise covariance (Q).
-
Adaptive Measurement Noise (R) Tuning: Traditional filters assume a fixed level of sensor noise, which doesn’t reflect the real-world variability in space. This framework introduces an online adaptation strategy for the measurement noise covariance matrix (R). It uses a Multiple Tuning Factor (MTF) approach, which is an innovation-based adaptive tuning method. This means the filter continuously assesses the consistency of incoming measurements and adjusts its confidence in them in real-time. If measurements are unexpectedly noisy or inconsistent, the filter becomes more cautious, preventing overconfident updates and ensuring better consistency.
-
Adaptive Process Noise (Q) Tuning: The process noise covariance (Q) accounts for unmodeled dynamics or maneuvers. Previous methods often relied on manual adjustments or prior knowledge, especially during periods of no measurements, like satellite eclipses. This new method dynamically tunes Q using a forward-looking cross-covariance matrix, inspired by a technique called Rauch–Tung–Striebel (RTS) smoothing. This allows the filter to automatically respond to dynamic model mismatches or loss of visibility, such as during an eclipse, without needing manual intervention or batch processing.
Integrated System Architecture
The complete pipeline begins with a lightweight CNN architecture called TinyCornerNET, specifically designed for detecting structural markers (corners) on the ENVISAT satellite from images captured by a chaser spacecraft. These 2D corner coordinates are then converted into 3D measurements using a camera model. These 3D measurements are then fed into the UKF, which estimates the full relative pose, including position, velocity, attitude, and angular velocity.
The system was rigorously evaluated through high-fidelity simulations using a realistic ENVISAT model. The performance was compared against ground truth under various conditions, including scenarios with measurement outages, such as eclipses. The results demonstrated that the dual-adaptive filter significantly improved estimation accuracy and consistency, especially during challenging periods when visual data was sparse or completely absent. For instance, during an eclipse, the adaptive filter maintained much tighter error bounds compared to non-adaptive filters, which showed growing divergence between predicted uncertainty and actual errors.
Also Read:
- Gaze-Guided Robots: Enhancing Efficiency and Robustness with Human-Inspired Vision
- Optimizing Information Freshness in Drone-Powered Sensing and Communication Networks
Advancing Space Operations Safety
This comprehensive approach offers an enhanced solution for robust onboard relative navigation, significantly advancing the capabilities required for safe proximity operations during ADR missions. By dynamically adjusting to varying measurement uncertainties and unmodeled dynamics, the framework ensures reliable performance even under the most challenging spaceborne conditions. This work contributes to the critical need for Active Debris Removal, which is vital for the long-term sustainability and safety of the Low Earth Orbit (LEO) space environment, helping to avoid scenarios like Kessler’s syndrome.
For more detailed information, you can read the full research paper here.
