Real-time Asset Management and Localization with Machine Learning and Bluetooth Low Energy Tags

Effectively managing assets is crucial for businesses, yet asset management can be expensive, time-consuming, and labor intensive. Improving asset management is extremely desirable to increase efficiency and profitability. Nearly 92% of companies invest in technologies to help manage assets. Technologies such as barcodes and Radio Frequency Identification (RFID) tags have long been used for asset tracking. However, these strategies are not able to precisely localize assets and cannot do so in real-time. Other strategies utilizing triangulation have been attempted using radio parameters such as RSSI and time of flight. However, these strategies are highly sensitive to environmental factors. Also, models made from radio parameters often generalize poorly. In this paper, we develop an improved asset tracking method leveraging Bluetooth, RSSI, and Packet Reception Rate. This approach overcomes many limitations of traditional asset tracking methods and enables assets to be accurately tracked in real-time. By training models specific to a deployment, we drastically reduce our sensitivity to environmental factors and improve accuracy. We demonstrate localization accuracies of 85% to 94% and show that these models can be easily trained. Additionally, the models require very minimal training data (only 30-40 samples per room) in order to achieve localization accuracies.