ChatTracer is an LLM-powered real-time Bluetooth device tracking system that comprises three key components: an array of Bluetooth sniffing nodes, a database, and a fine-tuned LLM.
The key highlights and insights are:
ChatTracer's design is based on the observation that commercial Apple/Android devices always broadcast hundreds of BLE packets per minute even in their idle status, enabling real-time tracking.
ChatTracer introduces a reliable and efficient BLE packet grouping algorithm that combines physical-layer features (time gap, AoA, RSS, CFO) and payload features (vendor, model, status, activity) to identify packets from the same device.
ChatTracer employs two techniques for fine-tuning the LLM: Supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF). SFT focuses on improving the LLM's localization accuracy, while RLHF enhances the LLM's overall performance.
Experimental results show that ChatTracer outperforms existing model-based and learning-based Bluetooth localization approaches, achieving a median localization error of 41 cm in a 60 m2 apartment, 58 cm in a 75 m2 laboratory, and 98 cm in a 200 m2 shopping mall.
ChatTracer provides an intelligent interface for end users to access real-time sensory information, similar to ChatGPT.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Qijun Wang,S... at arxiv.org 04-01-2024
https://arxiv.org/pdf/2403.19833.pdfDeeper Inquiries