RF Asset Localization Using ML for Non-Line-of-Sight Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inaccurate location tracking of assets in material handling environments due to reflections of RF beacon signals, leading to uncalibrated and erroneous positioning, which can result in productivity losses.
Innovation Solution
A central server system that utilizes RF beacons and machine learning models to differentiate between line-of-sight and non-line-of-sight locations, training on metadata and corrected location data from inertial sensors or mobile computers to predict accurate asset positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RF tags are used for asset tracking in material handling environments, then location tracking capability is provided, but location accuracy deteriorates when RF tags are out of Line of Sight of beacons
Solution Approach 1:
A machine learning model is introduced as an intermediary component that receives RF signal metadata from beacons and predicts accurate asset locations even when the RF tag is out of Line of Sight. The ML model acts as a mediator between the unreliable RF signal data and the required accurate location information, using patterns learned from calibrated locations to infer positions in uncalibrated regions.
Solution Approach 2:
The patent replaces the traditional mechanical/sensor-based tracking approach (requiring additional sensors like inertial or GPS systems) with an information-processing approach using machine learning algorithms. Instead of adding more physical sensors to improve accuracy, the system uses computational models to substitute for the missing sensor data in NLOS conditions.
2Measurement precision
If additional sensors like inertial and GPS systems are added to improve location accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent extracts the location prediction function from the physical sensor system and implements it through a machine learning model that processes existing RF signal metadata. By taking out the need for additional sensors and implementing the solution through software intelligence, the system achieves high measurement precision without the complexity of extra hardware components.
Solution Approach 2:
The invention substitutes mechanical sensor systems (inertial sensors, GPS receivers) with an information-processing system using machine learning algorithms. The ML model processes RF signal metadata and calibrated location data to predict accurate positions, replacing the need for additional physical sensing mechanisms while maintaining or improving accuracy.
3Measurement precision
If machine learning models are trained using metadata from both LOS and Non-LOS locations, then location prediction accuracy for uncalibrated locations improves, but processing time increases
Solution Approach 1:
The machine learning model is trained in advance using metadata from both LOS and Non-LOS locations, storing the learned patterns and relationships. During actual operation, the model can quickly make predictions by applying these pre-learned patterns rather than performing complex real-time analysis, thus reducing processing time while maintaining high accuracy for uncalibrated locations.
Data Source
Figure 1
Figure 1
Figure 3
AI summary
Various embodiments disclose a method for tracking assets. Method includes determining one or more locations of an asset within an indoor environment based on metadata associated an RF signal received from an RF tag associated with the asset. The method further includes identifying a first set of locations of the one or more locations. The method further includes identifying a second set of locations of the one or more locations, wherein the second set of locations corresponds to uncalibrated locations of the asset within indoor environment. Additionally, the method includes receiving a third set of locations of the asset. Furthermore, the method includes training a machine learning (ML) model based on the first set of locations, the second set of locations, and the third set of locations, and the metadata associated with the RF signal. The ML model predicts a fourth set of locations of another asset.