Abstract:Modern navigation solutions are largely dependent on the performances of the standalone inertial sensors, especially at times when no external sources are available. During these outages, the inertial navigation solution is likely to degrade over time due to instrumental noises sources, particularly when using consumer low-cost inertial sensors. Conventionally, model-based estimation algorithms are employed to reduce noise levels and enhance meaningful information, thus improving the navigation solution directly. However, guaranteeing their optimality often proves to be challenging as sensors performance differ in manufacturing quality, process noise modeling, and calibration precision. In the literature, most inertial denoising models are model-based when recently several data-driven approaches were suggested primarily for gyroscope measurements denoising. Data-driven approaches for accelerometer denoising task are more challenging due to the unknown gravity projection on the accelerometer axes. To fill this gap, we propose several learning-based approaches and compare their performances with prominent denoising algorithms, in terms of pure noise removal, followed by stationary coarse alignment procedure. Based on the benchmarking results, obtained in field experiments, we show that: (i) learning-based models perform better than traditional signal processing filtering; (ii) non-parametric kNN algorithm outperforms all state of the art deep learning models examined in this study; (iii) denoising can be fruitful for pure inertial signal reconstruction, but moreover for navigation-related tasks, as both errors are shown to be reduced up to one order of magnitude.

What problem does this paper attempt to address?

The paper attempts to address the issue of performance degradation in navigation solutions using low-cost MEMS inertial sensors (particularly accelerometers) in inertial navigation systems without external signal sources, due to the influence of noise sources. Specifically, the paper focuses on how to reduce noise in accelerometer signals during stationary states through data-driven methods, thereby improving the accuracy of the navigation system. ### Background and Problem Description Modern navigation solutions heavily rely on the performance of standalone inertial sensors, especially when no external signal sources are available. In such cases, inertial navigation solutions may degrade over time, particularly when using low-cost consumer-grade inertial sensors. Traditionally, model-based estimation algorithms have been used to reduce noise levels and improve the quality of useful information, thereby directly enhancing navigation solutions. However, the optimality of these methods is difficult to guarantee due to differences in manufacturing quality, process noise modeling, and calibration accuracy among different sensors. ### Limitations of Existing Methods Most current inertial denoising models are model-based methods, and recently some data-driven methods have been proposed, mainly for denoising gyroscope measurements. However, the task of denoising accelerometers is more challenging because the projection of gravity on the accelerometer axes is unknown. This makes accelerometer denoising an underexplored area. ### Main Contributions of the Paper To address this issue, the paper proposes several learning-based methods and compares their performance with existing denoising algorithms. The main contributions include: 1. Proposing a learning-based accelerometer denoising method that outperforms traditional signal processing denoising techniques. 2. Evaluating the denoising effect through Static Coarse Alignment (SCA), showing that the learning-based method excels in pure inertial signal reconstruction and reduces angular errors in navigation-related tasks by nearly an order of magnitude. 3. Conducting a comprehensive review of related literature, clarifying the contributions of this paper. ### Methods and Experiments The paper details the proposed learning methods, including unidirectional two-layer LSTM, bidirectional single-layer RNN, bidirectional single-layer GRU, and k-nearest neighbors algorithm (kNN). These methods were validated on simulated and experimental datasets, with evaluation metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Peak Signal-to-Noise Ratio (PSNR), and Relative Absolute Error (RAE). ### Results Analysis Experimental results show that learning-based methods significantly outperform traditional signal processing methods in denoising performance, particularly in pure signal reconstruction and navigation-related tasks. Through visual and quantitative analysis, the paper demonstrates the performance advantages of these methods at different time scales. In conclusion, the paper effectively addresses the noise issue in low-cost MEMS inertial sensors without external signal sources by proposing learning-based accelerometer denoising methods, thereby improving the accuracy and reliability of navigation systems.

Data-Driven Denoising of Stationary Accelerometer Signals

Data-Driven Denoising of Accelerometer Signals

Data-driven denoising of stationary accelerometer signals

Physics-Informed Data Denoising for Real-Life Sensing Systems

Denoising IMU Gyroscopes with Deep Learning for Open-Loop Attitude Estimation

Unsupervised Noise Reductions for Gravitational Reference Sensors or Accelerometers Based on the Noise2Noise Method

A data-driven approach for denoising GNSS position time series

A hybrid data-driven and learning-based method for denoising low-cost IMU to enhance ship navigation reliability

Using Inertial Sensors for Position and Orientation Estimation

A Learning-Based Gyros Denoising Method for Underwater Navigation

Inertial Navigation Meets Deep Learning: A Survey of Current Trends and Future Directions

Unsupervised noise reductions for gravitational reference sensors or accelerometers based on Noise2Noise method

A MEMS Gyroscope Noise Suppressing Method Using Neural Architecture Search Neural Network

DoorINet: Door Heading Prediction through Inertial Deep Learning

Deep Learning Assisted Inertial Dead Reckoning and Fusion

Towards Learning-Based Gyrocompassing

Enhancing Pure Inertial Navigation Accuracy through a Redundant High-Precision Accelerometer-Based Method Utilizing Neural Networks

A MEMS IMU De-Noising Method Using Long Short Term Memory Recurrent Neural Networks (LSTM-RNN)

A Learning-based approach for bias elimination in low cost gyroscopes

Deep Learning for Inertial Sensor Alignment

Hybrid Deep Recurrent Neural Networks for Noise Reduction of MEMS-IMU with Static and Dynamic Conditions