Simple Science

Cutting edge science explained simply

# Computer Science# Robotics# Artificial Intelligence

Improving Vehicle Localization with Advanced GNSS Techniques

A new method enhances vehicle localization by addressing GNSS signal challenges.

― 5 min read


GNSS LocalizationGNSS LocalizationAdvancesin urban areas.New method enhances vehicle positioning
Table of Contents

Global navigation satellite systems (GNSS) are crucial for accurately locating vehicles in transportation systems. However, in some environments, like city canyons, GNSS signals can be distorted. This can lead to challenges in Vehicle Localization, making it unreliable. The proposed method focuses on detecting non-line-of-sight (NLOS) receptions-situations where GNSS signals are not directly received because they are blocked by buildings or other structures. It also predicts errors in GNSS observations.

The Problem with GNSS

GNSS provides high accuracy in vehicle positioning, especially when combined with correction data. Yet, its effectiveness is often hindered in places with poor satellite visibility or high interference. Traditional methods often fail to address issues like NLOS receptions and multipath effects, where signals bounce off objects before reaching the receiver. This can result in faulty location data, leading to unsafe conditions in transportation systems.

There are several existing methods to deal with these issues. Some rely on statistical models to filter out bad measurements, while others use techniques like receiver autonomous integrity monitoring (RAIM) to check for consistency among GNSS measurements. However, these methods can still struggle in complex environments like urban areas, where many factors can affect signal quality.

Learning-Based Solutions

Recently, more advanced learning-based techniques have gained attention for their ability to classify NLOS signals and predict errors. These methods use data from various sensors and apply machine learning algorithms to improve GNSS signal accuracy. Deep Learning, in particular, can capture complex patterns in data, making it suitable for analyzing GNSS signals over time and across different contexts.

This approach involves training a model using labeled datasets from various locations to recognize patterns associated with both reliable and faulty GNSS signals. The goal is to create a system that can automatically identify NLOS receptions and predict errors in GNSS observations.

Dataset Collection

To train and evaluate the proposed system, data was collected in urban areas. The dataset includes measurements from an inertial measurement unit (IMU), lidar point clouds, and GNSS signals. This data allows the model to learn from real-world conditions and improve its predictions.

Labeling the data is a crucial step. It involves identifying which measurements are reliable and which are affected by NLOS conditions. This labeling is done using a multi-sensor fusion approach, combining information from IMUs and lidar with GNSS data.

Proposed Network Design

The proposed system is designed to analyze GNSS observations as a time series data. It employs a neural network combining long short-term memory (LSTM) networks with a transformer-like attention mechanism. This combination allows the model to better handle the unique challenges of GNSS data, capturing both time-related and spatial information.

The neural network processes the features extracted from GNSS observations, focusing on those that are most relevant for detecting NLOS conditions and predicting errors. The attention mechanism enhances the LSTM’s ability to evaluate different satellites' context, improving the overall performance of the model.

Training and Evaluation

The model was trained using data from different cities and evaluated both with in-distribution and out-of-distribution datasets. This evaluation helps assess how well the model performs when faced with data it hasn't seen before. The goal is to ensure the model can generalize well and provide accurate predictions across various scenarios.

The training process involves adjusting the model to minimize errors in predicting NLOS conditions and Pseudorange errors. Using both classical machine-learning methods and deep learning, the results from various models were compared.

Performance Metrics

The performance of the proposed method is evaluated using metrics like precision and recall. These metrics provide insight into how well the model identifies NLOS signals and predicts errors. A high precision indicates that when the model predicts NLOS, it's usually correct, while high recall means the model successfully identifies most of the actual NLOS cases.

Results demonstrate that the proposed deep learning model outperforms classical machine learning methods, particularly in challenging environments. The advanced model excels at maintaining accuracy even when trained on imbalanced datasets, where one class of data (e.g., NLOS signals) is less frequent than the other.

Feature Importance

Examining which features are most important for the model reveals that deep learning models effectively consider all features rather than relying on only a few. This balanced approach helps improve the model’s ability to generalize and avoid overfitting, a common issue in machine learning.

The features selected for this analysis include several key aspects of the GNSS observations, such as elevation and azimuth angles, signal strength, and pseudo-range residuals. Each feature provides valuable information about the quality of the GNSS signal and potential issues.

Vehicle Localization

To demonstrate the practical usefulness of this method, experiments were conducted using real-world GNSS data from urban environments. By integrating NLOS detections into a state estimation algorithm, it was possible to achieve more consistent vehicle localization results.

When NLOS observations were filtered out, the trajectory estimation improved significantly. This suggests that the proposed model does a better job of identifying signals that could lead to inaccurate localization. By distinguishing between reliable and unreliable signals, the model contributes to safer vehicle navigation in urban areas.

Conclusion

The proposed transformer-enhanced LSTM network shows promise in detecting NLOS signals and predicting GNSS pseudorange errors. By evaluating the model against both classical methods and new deep learning approaches, it becomes clear that the proposed system achieves better generalization and accuracy.

The results indicate that deep learning models can effectively handle the complexities of GNSS data in challenging environments. Future work will focus on improving dataset balance and expanding models to include advanced features, potentially increasing accuracy and performance even more.

By integrating this method into real-time systems, it will be possible to enhance the reliability of vehicle localization, making transportation safer and more efficient in urban settings.

Original Source

Title: Learning-based NLOS Detection and Uncertainty Prediction of GNSS Observations with Transformer-Enhanced LSTM Network

Abstract: The global navigation satellite systems (GNSS) play a vital role in transport systems for accurate and consistent vehicle localization. However, GNSS observations can be distorted due to multipath effects and non-line-of-sight (NLOS) receptions in challenging environments such as urban canyons. In such cases, traditional methods to classify and exclude faulty GNSS observations may fail, leading to unreliable state estimation and unsafe system operations. This work proposes a deep-learning-based method to detect NLOS receptions and predict GNSS pseudorange errors by analyzing GNSS observations as a spatio-temporal modeling problem. Compared to previous works, we construct a transformer-like attention mechanism to enhance the long short-term memory (LSTM) networks, improving model performance and generalization. For the training and evaluation of the proposed network, we used labeled datasets from the cities of Hong Kong and Aachen. We also introduce a dataset generation process to label the GNSS observations using lidar maps. In experimental studies, we compare the proposed network with a deep-learning-based model and classical machine-learning models. Furthermore, we conduct ablation studies of our network components and integrate the NLOS detection with data out-of-distribution in a state estimator. As a result, our network presents improved precision and recall ratios compared to other models. Additionally, we show that the proposed method avoids trajectory divergence in real-world vehicle localization by classifying and excluding NLOS observations.

Authors: Haoming Zhang, Zhanxin Wang, Heike Vallery

Last Update: 2023-10-12 00:00:00

Language: English

Source URL: https://arxiv.org/abs/2309.00480

Source PDF: https://arxiv.org/pdf/2309.00480

Licence: https://creativecommons.org/licenses/by/4.0/

Changes: This summary was created with assistance from AI and may have inaccuracies. For accurate information, please refer to the original source documents linked here.

Thank you to arxiv for use of its open access interoperability.

More from authors

Similar Articles