Neural Aided Kalman Filtering for UAV State Estimation in Degraded Sensing Environments

Problem
This paper addresses the challenge of accurate state estimation for unmanned aerial vehicles (UAVs) operating in degraded sensing environments, where traditional Kalman filter variants fail due to their assumptions being violated by nonlinear dynamics, noisy measurements, and unknown control inputs. The authors highlight the limitations of existing methods in handling these conditions and propose a novel approach that integrates neural networks with Kalman filtering. This work is presented as a preprint and has not yet undergone peer review.

Method
The core technical contribution is the Bayesian Neural Kalman Filter (BNKF), which combines a trained Bayesian Neural Network (BNN) with a Kalman correction step. The BNN models uncertainty by representing weights as distributions, allowing for the generation of predictive means and uncertainties through Monte Carlo sampling. The BNKF framework enhances traditional Kalman filtering by incorporating Bayesian uncertainty directly into covariance propagation, which is crucial for robust state estimation in high-noise scenarios. The authors utilize synthetic nonlinear UAV flight data for training and evaluation, employing five-fold cross-validation to assess performance across varying radar noise levels and sampling rates. The ensemble variant, BNKFe, is also introduced to further improve precision in high-noise conditions, albeit with a slight tradeoff in accuracy.

Results
The BNKF demonstrates superior performance compared to the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) across multiple metrics, including accuracy, precision, and truth containment. Specific results indicate that BNKF achieves a 15% improvement in accuracy and a 20% increase in precision over EKF under high noise conditions. The ensemble variant BNKFe shows a 10% enhancement in precision at the cost of a 5% reduction in accuracy. The runtime analysis reveals that BNKF incurs minimal inference overhead, making it suitable for real-time applications.

Limitations
The authors acknowledge that the performance of BNKF is contingent on the quality of the training data and the BNN architecture. They also note that while the method improves robustness in high-noise environments, it may still struggle with extreme outliers or highly dynamic scenarios not represented in the training set. Additionally, the reliance on Monte Carlo sampling introduces computational overhead, which, while minimal, could be a concern in resource-constrained environments. The paper does not address the scalability of the approach to larger state spaces or more complex UAV dynamics.

Why it matters
The integration of Bayesian neural networks with Kalman filtering represents a significant advancement in state estimation for UAVs, particularly in challenging environments where traditional methods falter. This work has implications for various applications in aerospace operations, including surveillance, reconnaissance, and autonomous navigation, where reliable state estimation is critical for decision-making. The proposed BNKF framework could serve as a foundation for future research into hybrid models that leverage the strengths of both neural networks and classical filtering techniques, potentially leading to more robust and adaptable systems in dynamic and uncertain environments.

Authors: Akhil Gupta, Erhan Guven
Source: arXiv:2604.28107
URL: https://arxiv.org/abs/2604.28107v1