Enhancement of Neural Inertial Regression Networks: A Data-Driven Perspective

TL;DR

The paper addresses the lack of standardized benchmarks for neural inertial regression by systematically evaluating 13 data-driven techniques across six real-world IMU datasets, totaling $1079$ minutes of data at $120$-$200$ Hz. It analyzes three broad strands—network architectural design (including multi-head variants), data augmentation (rotation, additive bias, additive noise), and data preprocessing (denoising, normalization, detrending)—using a consistent baseline CNN+Bi-LSTM+FC model trained with Adam on an RTX 4090. The key findings show that rotation and Gaussian noise augmentation yield the most consistent improvements (average RMSE gains around $7\%$ and $6\%$ respectively), while normalization generally harms performance; multi-head designs offer robustness with Head2 being more stable overall. The study also provides benchmarking strategies and dataset-specific recommendations, offering practical guidance for researchers and practitioners aiming to improve neural inertial regression systems in diverse real-world settings.

Abstract

Inertial sensors are integral components in numerous applications, powering crucial features in robotics and our daily lives. In recent years, deep learning has significantly advanced inertial sensing performance and robustness. Deep-learning techniques are used in different domains and platforms to enhance network performance, but no common benchmark is available. The latter is critical for fair comparison and evaluation in a standardized framework as well as development in the field. To fill this gap, we define and thoroughly analyze 13 data-driven techniques for improving neural inertial regression networks. A focus is placed on three aspects of neural networks: network architecture, data augmentation, and data preprocessing. Extensive experiments were made across six diverse datasets that were collected from various platforms including quadrotors, doors, pedestrians, and mobile robots. In total, over 1079 minutes of inertial data sampled between 120-200Hz were analyzed. Our results demonstrate that data augmentation through rotation and noise addition consistently yields the most significant improvements. Moreover, this study outlines benchmarking strategies for enhancing neural inertial regression networks.

Figures (10)

• Figure 1: Baseline network architecture. The output differs between each task and dataset.
• Figure 2: Taxonomy tree of data-driven techniques used for enhancing neural inertial networks.
• Figure 3: Head2 architecture. One head receives accelerometer data and the other receives gyroscope data.
• Figure 4: Head3 architecture. Each head receives the accelerometer and gyroscope readings along the x,y, and z axes.
• Figure 5: RMSE improvement in percentages for each dataset using the multi-head architecture.