Wi-Fi-Based Indoor Localization and Navigation: A Robot-Aided Hybrid Deep Learning Approach.

Area of Science:

• Robotics and Artificial Intelligence
• Wireless Communication and Networking

Background:

• Indoor localization and navigation are critical for mobile devices and network applications.
• Wi-Fi technology offers potential due to widespread infrastructure, but annotated data is often scarce.
• Existing methods like trilateration and machine learning rely heavily on labeled Wi-Fi observations.

Purpose of the Study:

• To develop a novel data collection strategy for indoor localization and navigation.
• To design deep learning models leveraging both labeled and unlabeled Wi-Fi data.
• To enhance the efficiency and accuracy of indoor positioning systems.

Main Methods:

• A robot-aided data collection strategy to acquire limited high-quality labeled and abundant unlabeled Wi-Fi data.
• Two deep learning models based on variational autoencoders for distinct localization and navigation tasks.
• A hybrid learning approach combining supervised, unsupervised, and semi-supervised learning for model training.

Main Results:

• The proposed approach effectively learns from unlabeled data, showing incremental performance improvements.
• Models achieved promising localization and navigation accuracy in complex indoor environments with obstacles.
• The hybrid learning strategy maximized the utility of collected data, including unlabeled datasets.

Conclusions:

• The robot-aided data collection and hybrid learning approach significantly enhances indoor localization and navigation.
• Deep learning models, trained with diverse data strategies, demonstrate robust performance.
• This work addresses the challenge of limited annotated data in Wi-Fi-based indoor positioning systems.