NeurIT: Pushing the Limit of Neural Inertial Tracking for Indoor Robotic IoT

2024-04-13

arXiv.org

AI Summary

To address the low accuracy and magnetic interference susceptibility of low-cost IMU-based inertial tracking in indoor robotic IoT applications, this paper proposes TF-BRT, an end-to-end sequential modeling framework. TF-BRT introduces a novel time-frequency block recurrent Transformer architecture for joint time- and frequency-domain feature learning, and is the first to systematically incorporate differential modeling of magnetometer measurements in the body frame to suppress environmental magnetic distortions. It fuses multi-source IMU signals without requiring external aids (e.g., vision, UWB, or prior maps). Under pure inertial conditions, TF-BRT achieves a mean pose error of only 1.0 m over 300 m trajectories—outperforming state-of-the-art methods by 48.21% in unseen environments and surpassing Google Tango in textureless scenarios. The code and dataset are publicly released.

Technology Category

• Intelligent Robots: State Estimation
• Computer Vision: Motion & Tracking
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Inertial tracking is vital for robotic IoT and has gained popularity thanks to the ubiquity of low-cost Inertial Measurement Units (IMUs) and deep learning-powered tracking algorithms. Existing works, however, have not fully utilized IMU measurements, particularly magnetometers, nor maximized the potential of deep learning to achieve the desired accuracy. To enhance the tracking accuracy for indoor robotic applications, we introduce NeurIT, a sequence-to-sequence framework that elevates tracking accuracy to a new level. NeurIT employs a Time-Frequency Block-recurrent Transformer (TF-BRT) at its core, combining the power of recurrent neural network (RNN) and Transformer to learn representative features in both time and frequency domains. To fully utilize IMU information, we strategically employ body-frame differentiation of the magnetometer, which considerably reduces the tracking error. NeurIT is implemented on a customized robotic platform and evaluated in various indoor environments. Experimental results demonstrate that NeurIT achieves a mere 1-meter tracking error over a 300-meter distance. Notably, it significantly outperforms state-of-the-art baselines by 48.21% on unseen data. NeurIT also performs comparably to the visual-inertial approach (Tango Phone) in vision-favored conditions and surpasses it in plain environments. We believe NeurIT takes an important step forward toward practical neural inertial tracking for ubiquitous and scalable tracking of robotic things. NeurIT, including the source code and the dataset, is open-sourced here: https://github.com/NeurIT-Project/NeurIT.

Problem

• Enhancing neural inertial tracking accuracy for robotic IoT
• Utilizing magnetometers and deep learning more effectively
• Improving robustness in diverse indoor environments

Innovation

• Time-Frequency Block-recurrent Transformer for feature learning
• Body-frame differentiation of magnetometers reduces error
• Robust performance in diverse indoor environments