A switching Kalman filter approach to online mitigation and correction of sensor corruption for inertial navigation
Artem Mustaev, Nicholas Galioto, Matt Boler, John D. Jakeman, Cosmin Safta, Alex Gorodetsky · 2024
Read A switching Kalman filter approach to online mitigation and correction of sensor corruption for inertial navigation by Artem Mustaev, Nicholas Galioto, Matt Boler, John D. Jakeman, Cosmin Safta, Alex Gorodetsky (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-06601.txt.
Introduction
Artem Mustaev and five co-authors present a method for handling corrupt sensor data in navigation in this 2024 paper. Inertial navigation tracks motion from inside a vehicle, with outside sensors correcting drift. When an outside sensor fails, standard practice throws its data away.
Their method instead keeps the bad data and runs several models of the sensor at once. A switching filter weighs which model best fits, spotting the moment a sensor goes bad. Parameter tricks let the system model the fault itself and correct for it. Tests on balloon flights and a shuttle reentry case recovered the true path even under heavy sensor bias. The authors also map the settings where the method works best and where it fails.
Catalog record
| CREATOR | Artem Mustaev, Nicholas Galioto, Matt Boler, John D. Jakeman, Cosmin Safta, Alex Gorodetsky |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-06601
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A switching Kalman filter approach to online mitigation and correction of sensor corruption for inertial navigation Artem Mustaev, Nicholas Galioto, Matt Boler, John D. Jakeman, Cosmin Safta, Alex Gorodetsky Abstract This paper introduces a novel approach to detect and address faulty or corrupted external sensors in the context of inertial navigation by leveraging a switching Kalman Filter combined with parameter augmentation. Instead of discarding the corrupted data, the proposed method retains and processes it, running multiple observation models simultaneously and evaluating their likelihoods to accurately identify the true state of the system. We demonstrate the effectiveness of this approach to both identify the moment that a sensor becomes faulty and to correct for the resulting sensor behavior to maintain accurate estimates. We demonstrate our approach on an application of balloon navigation in the atmosphere and shuttle reentry. The results show that our method can accurately recover the true system state even in the presence of significant sensor bias, thereby improving the robustness and reliability of state estimation systems under challenging conditions. We also provide a statistical analysis of problem settings to determine when and where our method is most accurate and where it fails. I Introduction I-A Motivation External corrections to inertial navigation systems (INS) are essential to realize highly performant vehicle navigation systems across a variety of domains [ 7 ] . However, without external sensor readings, the drift of Inertial Measurement Units (IMU) cause an accumulation of errors in estimates of the system state. External measurements such as those provided by global positioning systems (GPS) can be used to correct the drift inherent in IMU data. However, GPS and other sensors operating external to the vehicle are not immune to failure or corruption. For example, external threats such as spoofing attacks can introduce biases that lead to inaccurate state estimates having severe consequences, for example altering the vehicle course or even fully take over the control of a vehicle [ 10 , 2 , 9 ] . Consequently, as the reliance critical systems on GPS grows, we must develop robust methodologies to detect, mitigate, and recover from GPS corruption and failure. I-B Existing Approaches A variety of approaches exist for performing state estimation under the possibility of sensor failure, and these approaches include both preventive and reactive methods. Preventive measures use secure communication protocols and other protective strategies to limit the instances of GPS corruption [ 8 , 15 ] . However, these methods do not eliminate the risk entirely, and reactive mitigation strategies are necessary. While they are important, we do not focus on preventative approaches in this paper. In contrast, typical reactive approaches involve detecting the presence of corruption and then discarding the corrupted GPS measurements altogether [ 11 ] . While this approach prevents misleading information from affecting the state estimation, it also results in the loss of potentially valuable information that could still contribute to accurate state estimation if appropriately managed. This approach, although common, limits the robustness of state estimation systems, especially in environments where GPS signals are intermittent or degraded. Recent advancements in reactive methods have focused on developing more sophisticated methodologies that can continue to use GPS measurements when they are corrupted. Notably, particle filters have been employed to estimate the posterior distribution of the state by using a set of weighted particles, each representing a possible state of the system [ 12 ] . This approach employs a hypothesis based method to verify whether the difference between GPS measurement and prediction of the IMU measurement is greater than certain threshold set based on the GPS accuracy. The algorithm proceeds with normal particle filtering for estimating the state until data corruption is detected. The particle weights are then dynamically adjusted or a model describing the assumed corruption process is added, thereby decreasing the influence of corrupted measurements. To mitigate the downsides of weighted filters, alternative approaches use model-based learning to learn the parametrized corruption bias model [ 6 ] to mitigate spoofing. For example, the above mentioned reference compares the likelihood of receiving the current measurement based on previous states to a set threshold to identify if spoofing is present. Once corruption is detected, the system assigns a cost function to the current measurement based on the model parameters. It then uses iterative techniques, such as the golden search method [ 3 ] , to find a set of parameters that minimize the cost function and, therefore, are most likely to be the actual spoofing parameters. Once the parameters of the bias model are retrieved, the corruption is removed from the measurements, and state estimation proceeds as normal. This approach is significantly superior to the ones discussed previously as it identifies the corruption and recovers the initial measurements of the GPS, hence making use of all the available information for state estimation. However, it can be expensive to implement and extend to other corruption models beyond specific forms considered in [ 6 ] . I-C Paper Contributions In this paper, we introduce a novel approach that combines the Switching Kalman Filter (SKF) with parameter augmentation techniques to improve state estimation in the presence of corrupted measurements. Rather than discarding corrupted GPS data, our method retains and processes this information by running the Kalman filter with multiple observation models. Like the approach described in [ 6 ] , this paper aims to model sensor corruption as a parametric function, estimate the parameters, and then perform estimation with the corrected GPS measurements. However, our paper makes several significant contributions that further improve this idea. Unlike the discussed system, which deals with static systems, our approach performs online state estimation of a dynamical system. The SKF and parameter augmentation techniques identify the start of corruption, instead of relying on the previously discussed threshold-based methods which may be challenging to tune. Our method requires hypothesizing a single (or several) potential parametric model for the GPS bias, where the corruption is characterized by specific parameters that define the nature and extent of the bias over time. The SKF is designed to operate across multiple potential observation models — each one corresponding to an initiation of sensor fault — and switch between them as it assesses the likelihood of each model being correct at any given time. This allows the system to pinpoint the exact time step at which the GPS data transitions from an uncorrupted state to a corrupted one. The model then identifies the parameters of the bias model using a parameter augmentation approach, where the bias parameters are treated as additional state variables within the Kalman filter. The filter simultaneously estimates the state of the system and the parameters of the corruption model, continuously refining these estimates as new data is processed. This dual estimation process ensures that both the system state and the nature of the GPS corruption are accurately tracked, enabling the recovery of the true system state even in the presence of significant GPS bias. I-D Applications In this paper we demonstrate the effectiveness of the proposed methodology using two representative applications: balloon navigation in a moving atmosphere and shuttle reentry. These scenarios are representative of the significant challenges faced by existing state estimation methods when used with potentially corrupted GPS