FlexKalmanNet: A Modular AI-Enhanced Kalman Filter Framework Applied to Spacecraft Motion Estimation

Moritz D. Pinheiro-Torres Vogt,Markus Huwald,M. Khalil Ben-Larbi,Enrico Stoll

2024-05-06

Abstract:The estimation of relative motion between spacecraft increasingly relies on feature-matching computer vision, which feeds data into a recursive filtering algorithm. Kalman filters, although efficient in noise compensation, demand extensive tuning of system and noise models. This paper introduces FlexKalmanNet, a novel modular framework that bridges this gap by integrating a deep fully connected neural network with Kalman filter-based motion estimation algorithms. FlexKalmanNet's core innovation is its ability to learn any Kalman filter parameter directly from measurement data, coupled with the flexibility to utilize various Kalman filter variants. This is achieved through a notable design decision to outsource the sequential computation from the neural network to the Kalman filter variant, enabling a purely feedforward neural network architecture. This architecture, proficient at handling complex, nonlinear features without the dependency on recurrent network modules, captures global data patterns more effectively. Empirical evaluation using data from NASA's Astrobee simulation environment focuses on learning unknown parameters of an Extended Kalman filter for spacecraft pose and twist estimation. The results demonstrate FlexKalmanNet's rapid training convergence, high accuracy, and superior performance against manually tuned Extended Kalman filters.

Robotics,Earth and Planetary Astrophysics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the difficult problem of parameter tuning in spacecraft relative motion estimation. Specifically, although the traditional Kalman Filter (KF) is efficient in handling noise compensation, it requires extensive tuning of the system model and the noise model, which is often very time - consuming and complex in practical applications. Especially when dealing with nonlinear systems, this tuning becomes even more difficult.

To solve this problem, the paper proposes a new framework named FlexKalmanNet. This framework, by combining the Deep Fully Connected Neural Network (DFCNN) with the Kalman Filter, can directly learn the parameters of any Kalman Filter from the measurement data without manual tuning. In addition, FlexKalmanNet also has the flexibility to use different Kalman Filter variants, thereby improving its adaptability and performance.

The paper uses the experimental data in NASA's Astrobee simulation environment to demonstrate the fast training convergence, high precision of FlexKalmanNet in learning the unknown parameters of the Extended Kalman Filter (EKF), and its superior performance compared to the manually - tuned EKF. These characteristics make FlexKalmanNet an effective tool for solving the parameter tuning problem in spacecraft relative motion estimation.