US-20260127503-A1 - ASSET LOCATING METHOD

US20260127503A1US 20260127503 A1US20260127503 A1US 20260127503A1US-20260127503-A1

Abstract

A method for training a machine learning model for locating an asset in an environment when the machine learning model is executed on a computer system. The method includes receiving data corresponding to a plurality of training signals. The plurality of training signals are received at a plurality of receivers from a plurality of training tags located in the environment. The environment has a plurality of zones, and each of the training tags is associated with a zone in which it is located. The method further includes generating training data for first and second zones of the plurality of zones, wherein the training data for the first and second zones includes values of the training signals received from training tags located in the first and second zones, respectively, each value associated with the receiver at which a respective training signal was received. The method further includes training, using the training data for the first and second zones, a machine learning model to output a zone as a determined location of an asset in the environment based on an input which includes data corresponding to one or more signals received at one or more receivers from a tag associated with the asset.

Inventors

Aymen BAHROUN
Hedi FENDRI
Lamyae LAHLOU
Luca Gradassi

Assignees

Nagravision Sàrl

Dates

Publication Date: 20260507
Application Date: 20260105
Priority Date: 20230707

Claims (16)

1 . A computer-implemented method for training a machine learning model for locating an asset in an environment when the machine learning model is executed on a computer system, the method comprising: receiving data corresponding to a plurality of training signals, wherein the plurality of training signals are received at a plurality of receivers from a plurality of training tags located in the environment the environment having a plurality of zones, and wherein each of the training tags is associated with a zone in which it is located; generating training data for a first zone of the plurality of zones, wherein the training data for the first zone includes values of the training signals received from training tags located in the first zone, each value associated with the receiver at which a respective training signal was received; generating training data for a second zone of the plurality of zones, wherein the training data for the second zone includes values of the training signals received from training tags located in the second zone, each value associated with the receiver at which a respective training signal was received; and training, using the training data for the first zone and the training data for the second zone, a machine learning model to output a zone as a determined location of an asset in the environment based on an input which includes data corresponding to one or more signals received at one or more receivers from a tag associated with the asset.
2 . The computer-implemented method of claim 1 , wherein the data corresponding to the plurality of training signals comprises, for each training signal, the value of the training signal, information identifying the training tag from which the training signal was received, and information identifying the receiver at which the training signal was received.
3 . The computer-implemented method of claim 1 , wherein the values of the training signals are RSSI values.
4 . The computer-implemented method of claim 1 , wherein the training signals are received during a predefined scanning period.
5 . The computer-implemented method of claim 4 , wherein the training tags and/or the receivers are static within the environment for the duration of the predefined scanning period.
6 . The computer-implemented method of claim 1 , wherein the data corresponding to the training signals comprises a plurality of sets of data corresponding to training signals, each set comprising data corresponding to training signals received from a respective training tag at a respective receiver.
7 . The computer-implemented method of claim 6 , further comprising filtering the received data corresponding to the training signals, wherein filtering the received data corresponding to the training signals comprises: selecting, from each set of data corresponding to the training signals, the data corresponding to the training signal with the maximum value in that set; and discarding the data corresponding to the other training signals.
8 . The computer-implemented method of claim 1 , wherein multiple training tags are located in each zone of the plurality of zones, and the training tags in each zone are evenly spaced over the respective zones.
9 . The computer-implemented method of claim 1 , further comprising receiving, at the plurality of receivers, the plurality of training signals from the plurality of training tags located in the environment.
10 . The computer-implemented method of claim 1 , further comprising: splitting the training data for the first and second zones into a training set and a testing set; training the machine learning model using the training set; and testing the trained machine learning model using the testing set.
11 . The computer-implemented method of claim 1 , wherein: generating the training data for the first zone comprises generating a plurality of first zone arrays of the values of the training signals received from training tags located in the first zone, each first zone array including values of the training signals received from a different respective receiver; and generating the training data for the second zone comprises generating a plurality of second zone arrays of the values of the training signals received from training tags located in the second zone, each second zone array including values of the training signals received from a different respective receiver.
12 . The computer-implemented method of claim 11 , wherein: generating the training data for the first zone comprises, if one of the receivers has not received any training signals from a training tag located in the first zone, assigning a minimum observed value to the first zone array corresponding to that receiver; and generating the training data for the second zone comprises, if one of the receivers has not received any training signals from a training tag located in the second zone, assigning a minimum observed value to the second zone array corresponding to that receiver.
13 . The computer-implemented method of claim 9 , wherein: generating the training data for the first zone further comprises generating an N×1 first zone array, wherein N is the number of receivers, and each value in the N×1 first zone array is a value of a training signal received from a training tag located in the first zone at each different receiver; generating the training data for the second zone may further comprise generating an N×1 second zone array, wherein N is the number of receivers, and each value in the N×1 second zone array is a value of a training signal received from a training tag located in the second zone at each different receiver; and the machine learning model is trained using the N×1 first zone array and the N×1 second zone array.
14 . The computer-implemented method of claim 2 , wherein the values of the training signals are RSSI values.
15 . A system for training a machine learning model for locating an asset in an environment, the system comprising one or more processors, and memory storing thereon instructions that, as a result of being executed by the one or more processors, cause the system to perform the computer-implemented method of: receiving data corresponding to a plurality of training signals, wherein the plurality of training signals are received at a plurality of receivers from a plurality of training tags located in the environment the environment having a plurality of zones, and wherein each of the training tags is associated with a zone in which it is located; generating training data for a first zone of the plurality of zones, wherein the training data for the first zone includes values of the training signals received from training tags located in the first zone, each value associated with the receiver at which a respective training signal was received; generating training data for a second zone of the plurality of zones, wherein the training data for the second zone includes values of the training signals received from training tags located in the second zone, each value associated with the receiver at which a respective training signal was received; and training, using the training data for the first zone and the training data for the second zone, a machine learning model to output a zone as a determined location of an asset in the environment based on an input which includes data corresponding to one or more signals received at one or more receivers from a tag associated with the asset.
16 . A computer-implemented method for locating an asset in an environment, the method comprising: training a machine learning model according to the method of: receiving data corresponding to a plurality of training signals, wherein the plurality of training signals are received at a plurality of receivers from a plurality of training tags located in the environment the environment having a plurality of zones, and wherein each of the training tags is associated with a zone in which it is located, generating training data for a first zone of the plurality of zones, wherein the training data for the first zone includes values of the training signals received from training tags located in the first zone, each value associated with the receiver at which a respective training signal was received, generating training data for a second zone of the plurality of zones, wherein the training data for the second zone includes values of the training signals received from training tags located in the second zone, each value associated with the receiver at which a respective training signal was received, and training, using the training data for the first zone and the training data for the second zone, a machine learning model to output a zone as a determined location of an asset in the environment based on an input which includes data corresponding to one or more signals received at one or more receivers from a tag associated with the asset; receiving data corresponding to one or more signals received at one or more receivers from a tag associated with the asset; inputting the data corresponding to the one or more signals received from the tag into the trained machine learning model; and determining, based on the output of the trained machine learning model, a zone in the environment as the determined location of the asset in the environment.

Description

FIELD OF THE INVENTION The present invention relates to a method for training a machine learning model for locating an asset in an environment. The present invention also relates to a method for locating an asset in an environment and particularly, although not exclusively, to a method for locating an asset in an environment using machine learning techniques. BACKGROUND In many environments, there is a need to efficiently locate an asset in the environment. For example, in an indoor environment, such as a warehouse, there may be multiple assets (e.g., objects) in the indoor environment and the correct asset may need to be found quickly. Conventionally, humans were tasked with finding the correct asset in an environment such as a warehouse. Often, they did not know where the assets were, and so had to search the environment until they found the correct asset. Sometimes, maps of the environment, which mapped the location of different assets in the environment, could be used. The user could then follow the map to locate the required asset. However, this approach is labour intensive, and thus inefficient. It is known to track assets using GPS (Global Positioning System). A GPS tag positioned on the asset communicates with relevant satellites such that the asset can be located wherever they are positioned around the world. GPS provides live tracking of assets, and is thus especially useful for tracking assets in transit, such as rental cars. However, GPS tracking can be costly to implement, especially in a warehouse or indoor environment. Furthermore, GPS tracking of assets cannot provide an accurate or precise location in an indoor environment such as a warehouse. For example, GPS trackers may only be accurate to approximately >6m. In an indoor environment, although the GPS tracking may find an approximate location of the asset, a human would still need to manually search for and find the asset within that area. This leads to inefficient asset location. GPS tracking can also be costly to implement in some situations. It is also known to use radio frequency (RF) technology to locate assets in an environment. In particular, a radio frequency tag may be positioned on the asset, and the location of the asset may be determined based on Relative Signal Strength Indicator (RSSI) values measured by a plurality of receivers, and using a positioning algorithm such as trilateration or Time Difference of Arrival (TDOA) techniques. Bluetooth Low Energy (BLE) or Near Field Communication (NFC) technology may be used to locate assets in this way. Although these approaches are less costly than GPS, they are still not precise or accurate enough to provide efficient location of assets in an indoor environment such as a warehouse. This lack of accuracy of the RF technology techniques results from the assumption that RF propagation is in free space, without obstacles. However, in reality, there are often objects such as furniture, walls, and/or people in the environment, which impact radio signal propagation due to diffraction, reflection and/or scattering of the RF signals. This is known as RF multipath propagation/interference, and is particularly prevalent in indoor environments. Multipath interference occurs when an RF signal arrives at a receiver via two or more routes. This results in the total length of each signal path, and thus the time delay and phase of each received signal, to be different. This can lead to the two or more signals arriving in phase or out of phase, depending on the paths taken, meaning the signal power (RSSI) increases or decreases depending on the path taken. For stationary assets at a fixed location in a fixed environment, the multipath interference will be substantially constant over time when the assets' signals are measured from a given point in space, leading to large errors in the location measurement, and thus reduced accuracy. These RF propagation limitations can result in conventional location algorithms, such as BLE RSSI and trilateration, having a low locations precision of 5-10 m. In order to improve the accuracy of the location measurement, the number of receivers used to measure RSSI values from a radio frequency tag can be increased. However, such techniques still result in limited accuracy, and involve a complex deployment often with a vast quantity of receivers. The present invention has been devised in light of the above considerations. SUMMARY OF THE INVENTION According to a first aspect, there is provided a method for training a machine learning model for locating an asset in an environment when the machine learning model is executed on a computer system, the method comprising: receiving data corresponding to a plurality of training signals, wherein the plurality of training signals are received at a plurality of receivers from a plurality of training tags located in the environment, the environment having a plurality of zones, and wherein each of the training tags is associated wit