User equipment and process for performing control in a set of user equipment

Abstract

A user equipment that exercises control over the set of user equipment is provided with a model improvement module that is dedicated to updating the prediction model used to infer the state prediction values. The model improvement module alternates between a training mode, in which a new prediction model to be implemented next is determined, and an idle mode. A trade-off is obtained between limited resource requirements and accuracy of state knowledge. In the idle mode, the model improvement module can evaluate the validity of the currently used prediction model.

Description

Technical Field

[0001] The present invention relates to a user equipment adapted to perform control in a set of user equipments. The present invention also relates to a corresponding process and a set of user equipments. The present invention has a number of applications, particularly in the field of the Internet of Things.

Background Art

[0002] There is a system composed of a plurality of user equipments, where each user equipment determines its own state based on the measurement values of at least one of the other user equipments. In this way, the knowledge of the respective states of some or all of the user equipments is gradually constructed and continuously updated, so that it becomes possible to control the operation of one, some or all of the user equipments in a consistent manner. When some of the user equipments are controlled, their control may aim to achieve a global mission, and in such a case, it is called distributed control.

[0003] One example of such an application is the flight control of a set of several unmanned aerial vehicles, also known as flying drones, for example when these drones are intended to adopt a predetermined relative position. However, there are also many other applications, and the state definitions considered for each user equipment to perform control may vary depending on each application. In the case of the application to drones, each drone constitutes a separate user equipment, and its state is composed of a position in space, and such a position can be measured by a positioning system. In many other applications, control based on measurement values related to a plurality of user equipments, such as network synchronization, is effective. In the case of this latter application, wireless devices are user equipments, and their states are respective reference time points.

[0004] In such a system, each user device iteratively updates its own state based on knowledge of the states of several other user devices and, in some cases, based on knowledge of the state of only one other user device. In such a control system, better accuracy is achieved when determining the state of each user device compared to determining the states of the user devices individually. For this purpose, each user device needs to obtain continuous measurements of the states of some of the other user devices, which may or may not include a measurement of its own state. However, in practice, for many reasons, one or some of the state measurements required to update the state of one of the user devices are not available at the time of performing the update. To avoid preventing the state update process or simply to improve its accuracy, the state measurements are supplemented with state prediction values within the user device currently updating its own state. Each state prediction value is based on previous state measurements available within the user device performing the state update, and the prediction model issues the state prediction value from these previous state measurements. Next, for each of the user devices used to perform the update, respective state estimate values are determined based on the state measurements and state prediction values obtained for them. Next, state update values are determined from these estimate values. Instead of implementing the prediction model using the previous state measurements, it is possible to implement the prediction model based on statistical values derived from these previous state measurements.

[0005] A user device adapted to perform control in a set of user devices, wherein the user device under consideration updates its own state based on state measurements and state prediction values obtained for a subset of the user devices, in a known manner, The user device is storage means adapted to store at least some of the state measurements obtained for each user device of the subset or statistical values derived from these state measurements obtained for each user device of the subset, A prediction module adapted to infer a state prediction value for any of a subset of user devices by implementing a prediction model using state measurement values or statistical values retrieved from a memory means; A state estimation module adapted to issue a state estimation value for each of the subset of user devices by combining the state measurement values and the state prediction values of the subset of user devices; A resource control module adapted to select configuration parameters for controlling the operation of the resources of the user device based on the state estimation value issued by the state estimation module; Similarly, a state control module adapted to update the state of the user device under consideration based on the state estimation value issued by the state estimation module; Comprising.

[0006] The state measurement value can be obtained inside the user device under consideration, for example, when the user device under consideration includes a measurement module suitable for measuring one or some of the states of other user devices. Alternatively, the state measurement value may be received via communication by the user device under consideration from any suitable external means equipped with each user device and including measurement means suitable for measuring the respective state of this user device.

[0007] When the statistical values are stored in the memory means, these statistical values may have been provided by the state estimation module at an earlier time.

[0008] Such an architecture of the user device is shown in FIG. 1, in which the reference signs have the following meanings. 1, 2, 3, 4 Four user devices 10 Resources allocated to the user device, which may include wireless-type communication resources in some cases 13 Memory means of user device 1 as described above 14 Prediction module of user device 1 15 State estimation module of user device 1 16 Resource control module of user equipment 1 17 State control module of user equipment 1

[0009] In the example shown, user equipment 1 receives state measurement values from user equipment 2 and 3, but not from user equipment 4. Therefore, y 1 k , y 2 k and y 3 k indicate the state measurement values obtained for each of user equipment 1, 2, and 3 at time point k. These state measurement values are supplied to the state estimation module 15. The previous state measurement values obtained for user equipment 1, 2, and 3 before time point k are stored in the storage means 13 and supplied to the prediction module 14, and the prediction module 14 infers the state prediction values related to time point k and user equipment 1, 2, and 3. It is repeated that it is possible but not essential to use the state measurement values and prediction values related to this user equipment 1 within user equipment 1.

[0010] X 1 k , X 2 k and X 3 k indicate the state estimation values issued by the state estimation module 15 for user equipment 1, 2, and 3, respectively, at time point k. These state estimation values are obtained by combining the state Measurement value and the state prediction value in a known manner.

[0011] When the configuration parameter is intended to control the communication resources implemented between user equipment, the resource control module 16 can be a communication control module.

[0012] In some cases, in order to select the configuration parameter, the resource control module 16 can also take into account the requirements issued by the state control module 17, for example, when the state of the user equipment is changing rapidly.

[0013] Finally, x 1 k and x 1 k+1 represent the true state of the user equipment 1 at time points k and k + 1, respectively. The state x 1 k+1 is obtained by changing x 1 k through activation of the actuator of the user equipment 1, other inputs such as external command commands, etc. after the operation of the state control module 17. In some cases, the state control module 17 of the user equipment 1 can operate in cooperation with other user equipment so as to form a distributed control system.

[0014] In such a control operation, the accuracy of the state generated by the state control module 17 depends greatly on the accuracy of the state Measurement value and the state prediction value used. In addition, the accuracy of each state prediction value depends on both the measurement value input to the prediction model and the effectiveness of the prediction model used. However, a prediction model that is effective at one point in time may no longer be effective at a later point in time. Since the effectiveness of the prediction model is thus limited, the accuracy of the state prediction value may be low.

[0015] The state measurement values, that is, y in FIG. 1 1 k , y 2 k and y 3 k are more accurate and thus more reliable when each is obtained from several basic state measurement values all related to time point k. For this purpose, the collection frequency of the basic state measurement values should be higher than the state update frequency, and the state measurement values input to the state estimation module 15 and the storage means 13 for time point k are, respectively, the average of the basic state measurement values related to the same time point k. Such an average is calculated by the average calculation module 12 from the basic state measurement values received by the user equipment 1 or provided by its internal measurement module 11 if it exists.

[0016] In this specification, the prediction model implemented by the prediction module 14 may perform a joint estimation of the states of a subset of user devices, i.e., user devices 1, 2, and 3 in the example of FIG. 1, or may be composed of individual models individually related to the subset of user devices.

[0017] The paper "Reinforcement Learning-Based Control and Networking Co-Designs for Industrial Internet of Things" published in IEEE Journal on Selected Areas in Communications, Vol. 38, No. 5, May 2020 by H. Xu et al. discloses training the communication control module, i.e., module 16 in FIG. 1, through reinforcement learning so as to make an optimal decision when the current state of the system is given. The optimal decision is to recover the maximum number of basic state measurement values while avoiding congestion, especially the congestion of the communication means connecting the user devices to each other for transmitting the measurement means or state measurement values.

[0018] Starting from this situation, one object of the present invention is to improve the accuracy of the state estimation value determined by the user device for implementing control while reducing the resource requirements.

[0019] A secondary object of the present invention is to enable such improvement to be implemented from an existing user device architecture. SUMMARY OF THE INVENTION

[0020] To meet these objects or other objects, a first aspect of the present invention, in addition to the storage means, prediction module, state estimation module, resource control module, and state control module enumerated above, - a model improvement module adapted to alternately implement a training mode and an idle mode, A user device comprising the same is proposed.

[0021] In the training mode, the model improvement module determines a prediction model to be next implemented by the prediction module from the state measurement values or statistical values obtained for a subset of user devices, or from the state estimation values issued by the state estimation module.

[0022] The model improvement module is further configured to input the prediction model determined in the training mode to the prediction module. In this way, the input prediction model is used by the prediction module when the model improvement module is in the idle mode.

[0023] According to another feature of the present invention, the model improvement module is further adapted to control the resource control module to adjust the reliability of the state measurement values obtained for each user device in the subset to a level higher than another level of reliability effective during a period dedicated to the idle mode, during a preceding or subsequent period dedicated to the training mode.

[0024] Therefore, in the implementation of the control, while providing a state prediction value with sufficient reliability, fewer resources required during the period dedicated to the idle mode are reduced. The reliability of such a state prediction value is provided by the training mode, in which the prediction model is trained using more resources. In this way, the prediction model continuously used by the prediction module becomes more reliable, and thus the quality of the state estimation value issued by the state estimation module is improved. The present invention thus improves the trade-off between the resource requirement and the accuracy of the state estimation value used by the state control module.

[0025] In some cases, in the idle mode, since the prediction model is pre-determined by the model improvement module during a period dedicated to the training mode, the model improvement module can evaluate the effectiveness of this prediction model currently implemented by the prediction module.

[0026] In a possible embodiment of the present invention, the resource is adapted to enable improvement of the accuracy of each state measurement value directly issued by the state measurement means used. In such a case, the accuracy of the state measurement value results in the reliability of the state measurement value.

[0027] In another possible embodiment of the present invention, each state measurement value of any subset of user devices stored by the storage means or used by the state estimation module can be obtained from one or several basic state measurement values related to the same user device. Next, the model improvement module is further adapted to control the resource control module to adjust the frequency at which the basic state measurement values are obtained for each user device in the subset such that the value of this frequency during a period dedicated to the training mode is higher than another value of the frequency effective during a preceding or subsequent period dedicated to the idle mode. In such an embodiment of the present invention, the reliability of the state measurement value is improved by increasing the number of basic state measurement values resulting in one state measurement value.

[0028] For some applications of the present invention, one or several of the following additional features can be advantageously implemented. - The user devices are adapted to communicate with each other by implementing a wireless communication mode and wirelessly receive at least one state measurement value or basic state measurement value of a subset of the user devices. At this time, the resources of the set of user devices can include wireless communication resources. - The user equipment is adapted such that the state of any of the subset of user equipment includes the coordinate values of the positions of the subset of user equipment. In such a case, the user equipment is further adapted to operate using the basic state measurement values provided by at least one positioning system. Next, the basic state measurement values of each user equipment in the subset can be wirelessly transmitted between user equipment or transmitted within each user equipment. - The model improvement module is adapted to switch from the training mode to the idle mode when the current determination of the prediction model satisfies the matching criterion in the training mode between the state measurement value and the state prediction value inferred using this currently determined prediction model, or after the user equipment receives a request to release some resources, or at the time issued by the scheduling algorithm. - The model improvement module is subject to the following conditions, namely, the matching criterion in the idle mode is no longer satisfied between the state measurement value and the state prediction value issued by the prediction module, after receiving information that a certain amount of resources are available, after receiving information that the reliability of the state measurement value exceeds a predetermined threshold, at the time issued by the scheduling algorithm, when at least one of the above occurs, it is adapted to switch back from the idle mode to the training mode. - Based on the state estimation value issued by the state estimation module and, in some cases, also based on the requirements from the state control module, the configuration parameters selected by the resource control module are suitable for controlling at least one of the frequency of basic state measurement, the transmission frequency of these basic state measurement values, the modulation and coding scheme, and at least one transmission power implemented between user equipment.

[0029] A second aspect of the present invention proposes a process for performing control in a set of user equipment, wherein each user equipment updates its own state based on state measurement values and state prediction values obtained for a subset of the user equipment. To this end, each user equipment performs the following steps, namely, storing at least some of the state measurement values obtained for each user equipment in the subset, or storing statistical values derived from these state measurement values obtained for each user equipment in the subset; inferring a state prediction value for any one of the user equipment in the subset by implementing a prediction model using the pre-stored state measurement values or statistical values; issuing an estimated value for each state of the user equipment in the subset by combining the state measurement values and state prediction values of the user equipment in the subset; selecting configuration parameters for controlling the operation of the resources of the user equipment based on the state estimated values; similarly, updating the state of the user equipment based on the state estimated values; and execute.

[0030] According to the present invention, at least one of the user equipment called model improvement user equipment operates alternately in a training mode and an idle mode.

[0031] In the training mode, the prediction model is determined from the state measurement values or statistical values obtained for each user equipment in the subset, or from the state estimated values issued for each user equipment in the subset.

[0032] Next, the process further includes updating the prediction model implemented to infer the state prediction value according to the prediction model determined in the training mode.

[0033] Also, the configuration parameters are adjusted such that the reliability of the state measurement values obtained for each user equipment in the subset is at a higher level than another level of reliability effective during a period dedicated to the idle mode, either preceding or succeeding a period dedicated to the training mode.

[0034] In some cases, in the idle mode, since this prediction model is predetermined during a period dedicated to the training mode, the effectiveness of the currently implemented prediction model is evaluated.

[0035] In a possible embodiment of the present invention, for any of the user equipments in the subset, each state measurement value stored or used to issue one of the state estimation values can be obtained from one or several basic state measurement values. In that case, the frequency at which the basic state measurement values are obtained for each user equipment in the subset can be adjusted such that the value of this frequency during a period dedicated to the training mode is higher than another value of the frequency effective during a period dedicated to the idle mode, either preceding or succeeding.

[0036] The model improvement user equipment may be according to the first aspect of the invention.

[0037] In a manner similar to the first aspect of the invention, the following features can be implemented for the process of the second aspect of the invention. - The state measurement value or the basic state measurement value can be wirelessly transmitted to the model improvement user equipment by at least one of the user equipments in the subset. - The state of any of the user equipments in the subset can include the coordinate values of the position of the user equipment in the subset. In such a case, the basic state measurement value can be provided by at least one positioning system. Next, the basic state measurement value for each user equipment in the subset can be wirelessly transmitted to the model improvement user equipment or can be transmitted inside the model improvement user equipment. - When the matching criterion for the training mode is satisfied between the state measurement value and the state prediction value inferred using this prediction model by the currently determined prediction model, or after the model improvement user device receives a request to release some resources, or at the time issued by the scheduling algorithm, the model improvement user device can switch from the training mode to the idle mode. - The model improvement module is subject to the following conditions, namely, the matching criterion for the idle mode is no longer satisfied between the state measurement value and the state prediction value used for issuing the state estimation value, after receiving information that a certain amount of resources is available, after receiving information that the reliability of the state measurement value exceeds a predetermined threshold, at the time issued by the scheduling algorithm, when at least one of the above occurs, it can switch back from the idle mode to the training mode. - The configuration parameters can be selected based on the state estimation value and, optionally, also based on the control requirements, in order to control at least one of the frequency of basic state measurements, the transmission frequency of these basic state measurement values, the modulation and coding method, and at least one transmission power implemented between user devices.

[0038] Finally, the third aspect of the present invention proposes a set of user devices adapted to perform control, and at least one of these user devices is according to the first aspect of the invention.

[0039] These features and other features of the present invention will now be described with reference to the accompanying drawings, which relate to preferred but non-limiting embodiments of the present invention.

Brief Description of the Drawings

[0040]

Figure 1

Figure 2

Figure 3

Figure 4

Best Mode for Carrying Out the Invention

[0041] For the sake of clarity of the description to be detailed below, when reference signs shown in different ones of the drawings, including FIG. 1 already described, are the same, they refer to the same single or plural elements having the same function.

[0042] Here, in the present invention, a specific application of controlling a set of unmanned aerial vehicles, generally called drones, to have a desired relative flight position will be described. For such an example, each drone constitutes a separate user device introduced above, and the respective current flight positions of each drone are its current states. Therefore, the state of drone i at time k is x i k as shown. Here, i is an integer index that separately identifies each drone in the entire set.

[0043] However, the present invention can be applied to many systems other than the drone set, and the type of user device depends on each system and the states considered for these user devices.

[0044] FIG. 2 shows, for example, a simple system in which the flight positions of each of four drones are controlled using distributed control. The drones are labeled with reference signs 1, 2, 3, and 4 corresponding to the integer index i introduced above. For each drone i, the corresponding darkly shaded circle is its true position at time k: x i kis shown. As an illustrative but non-limiting example, the target of distributed control implemented in the drone set is to form and maintain a specified rectangular pattern having the relative positions of drones 1 to 4.

[0045] Each drone i can be equipped with its own GPS (GPS represents "Global Positioning System") receiver that constitutes the measurement module of this drone i. Alternatively, any other positioning system may be used. For this reason, in the case of drone 1, its measurement module 11 is composed of a GPS positioning receiver. For this reason, each drone i can, in principle, collect at least one basic measurement value of its own position related to any time point k, preferably several such measurement values.

[0046] Each drone i can also be provided with communication means (not shown) designed to communicate with other drones. At this time, the communication means can be made part of the resource 10 in addition to the measurement module 11. These resources 10 can also include other communication components such as a central node or a master node, etc., and these can be at least partially of a wireless type, especially a wireless type. For this reason, each drone can transmit the basic measurement value of its own position provided by its GPS receiver to at least some of the other drones in the set.

[0047] Alternatively, other systems for measuring the positions of the drones may be used. For example, each drone i can be equipped with appropriate means for measuring the respective positions of at least some of the other drones. Yet another possibility is that each drone i wirelessly receives the respective position measurement values of at least some of the other drones from an external position measurement system. It will be understood that the present invention is applicable regardless of what the actual measurement means and communication model used to provide the necessary position measurement values for at least one of the drones are.

[0048] Assume that drone 1 does not receive the position measurement values of drone 4 and only receives the position measurement values of drones 2 and 3, and the description of drone 1 will be continued. For this reason, drones 1, 2, and 3 form a subset within the entire drone set, and this is used to perform control on drone 1. Let j be an integer index that identifies the drones in this subset. That is, j = 1, 2, or 3. When there are multiple basic position measurement values related to time point k and drone j, they are averaged by the averaging module 12 in order to issue the resulting position measurement value. This position measurement value is also related to time point k and drone j here, and is y j k is. When only one basic position measurement value is collected for time point k and drone j, this directly constitutes the position measurement value y j k In addition, in some cases, no basic measurement values may be collected for time point k and drone j. Due to measurement errors and noise, each position measurement value y j k may be different from the corresponding true position x j k In FIG. 2, the position measurement values are represented as white circles for each drone i, and each dotted circle represents its target relative position with respect to other drones.

[0049] The basic position measurement value of drone 1 can be internally provided by the GPS receiver 11, and can also be averaged by the averaging module 12 of drone 1 to provide the position measurement value y 1 k However, it is not essential for drone 1 to use its own position measurement value, and drone 1 may use only the position measurement values of drones 2 and 3. In the following, the position of drone 1 is measured by the measurement value y 1 k is considered to be available in addition to the position measurement values y 2 k and y 3 k

[0050] For the drone 1 at time points k-1, k-2,... before time point k, y 1 k-1 , y 1 k-2 ..., y 2 k-1 , y 2 k-2 ... and y 3 k-1 , y 3 k-2 ... Assume that the position measurement values of drones 1, 2, and 3, which are indicated by y, are collected in advance. Here, k, k-1, k-2,... are the consecutive time points at which the drone 1 updates its position and obtains the true positions x 1 k+1 , x 1 k , x 1 k-1 ... Based on the previous position measurement values y 1 k-1 , y 1 k-2 ..., y 2 k-1 , y 2 k-2 ... and y 3 k-1 , y 3 k-2 ... stored by the memory means 13, the prediction module 14 infers the position prediction values at time point k for drones 1 to 3 respectively. Alternatively, the memory means 13 can store the statistical distribution of each drone j obtained from the previous position measurement values y j k-1 , y j k-2 ... Such a statistical distribution can be transmitted by the state estimation module 15 to the memory means 13. The position prediction values are inferred by the prediction module 14 using the prediction model discussed below. The position measurement values and position prediction values related to time point k are all related to the respective position estimation values X 1 k , X 2 k and X 3 kAll of these are supplied to the state estimation module 15 that combines them together for issuing. The remaining part of the control operation remains unchanged with respect to the prior art already described.

[0051] Position measurement value y j k 、y j k-1 、y j k-2 The reliability of each of... can be improved by increasing the number of basic position measurement values related to drone j and the point of interest and combined through averaging by the averaging module 12. Therefore, it is necessary that the collection frequency of the basic position measurement values is higher than the state update frequency. Adjusting the collection frequency of the basic position measurement values involves the configuration parameters of the drone set as the measurement frequency and the transmission bandwidth. For this reason, the position measurement value y used for implementing distributed control j k 、y j k-1 、y j k-2 ... can be increased, but at the cost of increased resource requirements.

[0052] Another issue relates to the effectiveness of the prediction model implemented by the prediction module 14. In fact, this prediction model constructs assumptions about the movements of drones 1, 2, and 3 and provides a prediction of its own position at time point k from the previous positions. However, this prediction model may no longer be valid for any reason such as a new external command, the influence of at least one drone actuator, any interference event, etc.

[0053] The operation of the overall control can be supported by an equation derived from a known Bayesian framework. This framework combines the position measurement value and the position prediction value to obtain the updated position estimate X j k . This framework is for the position estimate X j kIndicates that the accuracy of [[ID=]] depends greatly on the reliability of the measured values and predicted values. As shown above, the reliability of the measured values can be improved by increasing the frequency with which the basic measured values are input to at least some of the average calculation modules 12 of drone j. However, if the effectiveness of the prediction model implemented by the prediction module 14 is insufficient, the overall control becomes vulnerable. The improvements of the present invention described below with reference to FIG. 3 aim to reduce or eliminate this vulnerability.

[0054] As shown in FIG. 3, the control system mounted on drone 1, and preferably on other drones as well, is completed with an additional module 18 called a model improvement module. For this reason, drone 1 is referred to as a model improvement user device in the summary part of this specification. Thus, the architecture of FIG. 1 including modules 11 - 17, known from the prior art, is maintained but combined with the new module 18. This module 18 is designed according to the present invention to improve the trade - off between resource demand and prediction reliability. According to the present invention, the model improvement module 18 receives the position measurement value y from the storage means 13 in parallel with the prediction module 14 and is connected to control the resource control module 16. The model improvement module 18 may alternatively be connected to receive the position estimate X or the statistical value of the position measurement value from the state estimation module 15 in parallel with the resource control module 16 and the state control module 17, instead of being connected to receive the position measurement value y from the storage means 13. The model improvement module 18 is also connected to the prediction module 14 to update the prediction model actually implemented by the prediction module 14 and monitor the accuracy of the resulting position prediction value. Most conveniently, the model improvement module 18 can be implemented as a software module. j k to receive and is connected to control the resource control module 16. The model improvement module 18 may alternatively be connected to receive the position measurement value y j k from the storage means 13. Instead, it is connected to receive the position estimate X j k or the statistical value of the position measurement value from the state estimation module 15 in parallel with the resource control module 16 and the state control module 17. The model improvement module 18 is also connected to the prediction module 14 to update the prediction model actually implemented by the prediction module 14 and monitor the accuracy of the resulting position prediction value. Most conveniently, the model improvement module 18 can be implemented as a software module.

[0055] The model improvement module 18 is designed to operate alternately in a training mode and an idle mode. As described above, the model improvement module 18 operates in either mode while control continues to be implemented. The training mode is dedicated to issuing a new prediction model that will be used next by the prediction module 14. This new prediction model will provide a more accurate position prediction value. In fact, the training mode results in an update of the prediction model to be implemented by the prediction module 14 in order to take into account changes that may occur in the operation of the set of drones, such as external command commands or perturbations. At the end of each period dedicated to the training mode, the updated prediction model is input by the model improvement module 18 into the prediction module 14 and used in place of the current prediction model. Next, the model improvement module 18 switches to the idle mode and can, but is not required to, monitor the effectiveness of the newly implemented prediction model by the prediction module 14 in the idle mode. Since the prediction model was trained immediately before the preceding period dedicated to the training mode, its effectiveness is more certain and the resulting position prediction value is more accurate and reliable. During the next period dedicated to the idle mode, the model improvement module 18 can check the effectiveness of this prediction model by comparing each position measurement value y j k with the corresponding position prediction value issued by the prediction module 14.

[0056] In the training mode, a sufficient number of position measurement values need to be collected. Thereby, the prediction model currently being trained by the model improvement module 18 becomes sufficiently accurate. For this purpose, the model improvement module 18 forces the resource control module 16 in the training mode so that basic position measurements are performed and the basic position measurement values are transmitted at a sufficiently high frequency. Conversely, during the idle mode, since the prediction model is sufficiently accurate for estimating the position measurement values, the requirement for a sufficiently large number of basic position measurement values no longer holds. Therefore, it is possible to reduce the demand for measurement and communication resources during the period dedicated to the idle mode. In particular, the frequency of basic position measurements and the transmission frequency of basic position measurement values controlled by the resource control module 16 can be significantly reduced with respect to the training mode. Therefore, the transmission load of communication resources can be reduced, but due to the reliability of the prediction model provided by the training mode, as a result of such reduction, the accuracy of the position estimation value is hardly lost. This advantage of the present invention particularly applies when the communication between user equipment, that is, drones, is of the wireless type. At this time, according to the present invention, a reduction in wireless load can be obtained.

[0057] Therefore, during each period dedicated to the training mode, the model improvement module 18 controls the resource control module 16 to obtain sufficient basic position measurement values for training a new prediction model. At this time, the obtained position measurement values y j k , y j k-1 , y j k-2 ... are more reliable. The cost therefor is the resource requirement in the training mode for collecting basic position measurement values at a sufficiently high frequency. This resource requirement is satisfied during each training mode period by forcing appropriate configuration parameters of the drone set. For the exemplary applications of FIGS. 1 and 3, the reliable position measurement values y thus collected for drones 2 and 3 and inside drone 1 j kis used by the state estimation module 15 to obtain a more accurate position estimate value X j k and is also used in parallel in the model improvement module 18 to train a new prediction model. At this point, the prediction module 14 still implements the prediction module provided by the model improvement module 18 at the end of the preceding period dedicated to the training mode. During each period dedicated to the training mode, the state estimation module 15 mostly depends on the position measurement values. This is because these position measurement values are sufficient in number and accuracy.

[0058] FIG. 4 shows the prediction model being trained by the model improvement module 18 in the training mode. Such an expression is common to those skilled in the art of artificial intelligence. As shown in the left part of FIG. 4, the input of the model is the position measurement values y of drones 1, 2, and 3 at at least one preceding time point k−1 in the operation of the entire control system in the case of the application example of FIG. 2 j k-1 In some cases, these position measurement values y j k-1 can be replaced with the position estimate values X issued for drones 1, 2, and 3 and for time point k−1 j k-1 and / or the associated statistical distribution. As shown by z in the right part of the figure j k the output of the prediction model is the position prediction values for these drones 1, 2, and 3 and for the next time point, i.e., k. The labels used for training are the position measurement values obtained for time point k: y 1 k y 2 k y 3 k In different embodiments of the present invention, the prediction model may handle drones j collectively or may be composed of individual prediction models related to drones j separately and respectively.

[0059] The end of each training mode period can be triggered in various ways. The first way is that the judgment criterion is satisfied by the prediction model currently being trained, and this trained prediction model provides sufficient consistency between the predicted position value z j k and the measured position value y j k This is what it means. Such a judgment criterion is called the training mode consistency judgment criterion in the summary part of this specification. For example, the following judgment criterion: norm ||y j k -z j k || is less than a predetermined threshold for several subsequent time points k, can be implemented. Another possibility to stop the training mode period is that drone 1 receives a request for resources available from any other element of the system, including one of the other drones. Such a request may be issued because another function to be implemented within the drone set requires sufficient communication bandwidth. Yet another possibility to stop the training mode period is a command issued by a scheduling algorithm. Such a scheduling algorithm is dedicated to distributing communication resources among various functions of the drone set, based on the assignment rules of these functions and their respective resource requirements, whatever these functions may be. At the end of the training mode period, the newly trained prediction model is input into the prediction module 14 and replaces the previous prediction model. Next, the prediction module 14 operates using this new prediction model, and the model improvement module 18 switches to the idle mode.

[0060] During the idle mode period, the model improvement module 18 can continuously check the effectiveness of the prediction model currently being implemented by the prediction module 14. For this purpose, the model improvement module 18 compares the position prediction value obtained as a result from the prediction model currently being implemented by the prediction module 14 with the measured position value, that is, y 1 k 、y2 k and y 3 k is compared with. Since the effectiveness of the prediction model used is improved, the control operation during the idle mode period is supplied with a reduced number of basic position measurement values. Thereby, the idle mode requires fewer measurement and communication resources than the training mode, and thus more resource amounts can be allocated to functions other than distributed control. In particular, the frequency of basic position measurement performed for each drone and the transmission of basic position measurement values between drones is set to a higher value in the training mode compared to this frequency value effective in the idle mode through the resource control module 16 by the model improvement module 18. The idle mode period is the position measurement value y j k can continue until the coincidence criterion is no longer satisfied between the position prediction value issued by the prediction module 14. This coincidence criterion for the idle mode may be similar to the coincidence criterion for the training mode, but does not necessarily have to be similar. Alternatively, the idle mode period may continue until drone 1 receives information that a certain amount of resources are available, or until a switching command is issued by the scheduling algorithm. Next, the model improvement module 18 switches back to the training mode. Also, in the idle mode, the model improvement module 18 continuously evaluates the reliability of the position measurement value y j k and switches to the training mode when it is determined that this reliability is good enough to enable training of a new prediction model. Such a switching trigger can be facilitated by increasing the frequency of basic position measurement when possible in the idle mode. Here too, in the idle mode, there may be a requirement for the model improvement module 18 to temporarily increase the frequency of basic position measurement in order to obtain a measurement result with improved reliability.

[0061] The present invention is not limited to the field of controlling a set of drones, and can be advantageously applied to many other systems composed of a plurality of user devices. The number of user devices in the system and the number of these devices actually used within one user device as a subset for implementing decentralized control can be arbitrary. Therefore, the number 4 of user devices in the system and the number 3 of user devices in the subset are not limiting and are used only for illustrative purposes.

Claims

1. A user device adapted to perform control in a set of user devices, wherein the user device updates its own state based on state measurement values and state prediction values obtained for a subset of the user devices, the user device is, storage means adapted to store at least some of the state measurement values obtained for each user device in the subset, or to store statistical values derived from the state measurement values obtained for each user device in the subset; a prediction module adapted to infer a state prediction value for any one of the user devices in the subset by implementing a prediction model using the state measurement values or statistical values retrieved from the storage means; a state estimation module adapted to issue a state estimation value for estimating the state at the time of estimation for each of the user devices in the subset by combining the state measurement values and the state prediction values of the user devices in the subset related to the time of estimation; a resource control module adapted to select configuration parameters for controlling the operation of the resources of the user device based on the state estimation value issued by the state estimation module; similarly, a state control module adapted to update the state of the user device based on the state estimation value issued by the state estimation module; A user device comprising: the user device is, further comprising a model improvement module adapted to alternately perform a training mode and an idle mode, in the training mode, the model improvement module trains a prediction model to be next implemented by the prediction module from the state measurement values or statistical values obtained for the user devices in the subset, or from the state estimation values issued by the state estimation module, the model improvement module is further configured to input the prediction model trained in the training mode into the prediction module, whereby when the model improvement module is next in the idle mode, the prediction model input into the prediction module by the model improvement module is used by the prediction module, The model improvement module is further adapted to control the resource control module to adjust the reliability of the state measurement values obtained for each user device of the subset to a level higher than another level of reliability effective during a preceding or subsequent period dedicated to the idle mode during a period dedicated to the training mode, a user device characterized by that.

2. Since the prediction model is pre-trained by the model improvement module during a period dedicated to the training mode, the model improvement module is further adapted to evaluate the effectiveness of the prediction model currently implemented by the prediction module during a period dedicated to the idle mode, the user device according to claim 1.

3. For any one of the user devices of the subset, each state measurement value stored by the storage means or used by the state estimation module is adapted to be obtained from one or several basic state measurement values, The model improvement module is further adapted to control the resource control module to adjust the frequency at which the basic state measurement values are obtained for each user device of the subset such that the value of the frequency during a period dedicated to the training mode is higher than another value of the frequency effective during the preceding or subsequent period dedicated to the idle mode, the user device according to claim 1 or 2.

4. The state of any one of the user devices of the subset is adapted to include coordinate values of the position of the user device of the subset, the user device according to claim 1 or 2.

5. Adapted to operate using the basic state measurement values provided by at least one positioning system, the basic state measurement values for each user device of the subset are wirelessly transmitted between the user devices or transmitted inside each user device, the user device according to claim 3.

6. The model improvement module is adapted to switch from the training mode to the idle mode when the consistency judgment criterion of the training mode is satisfied between the state measurement value and the state prediction value inferred using the prediction model currently being trained, or after a request to release some resources is received by the user equipment, or at the time issued by the scheduling algorithm, for the user equipment according to claim 1 or 2.

7. The model improvement module is subject to the following conditions, namely, the consistency judgment criterion of the idle mode is no longer satisfied between the state measurement value and the state prediction value issued by the prediction module, after receiving information that a certain amount of the resources is available, after receiving information that the reliability of the state measurement value exceeds a predetermined threshold, at the time issued by the scheduling algorithm, when at least one of the above occurs, it is adapted to switch back from the idle mode to the training mode, for the user equipment according to claim 1 or 2.

8. Based on the state estimation value issued by the state estimation module and, in some cases, also based on the requirements from the state control module, the configuration parameters selected by the resource control module are suitable for controlling at least one of the frequency of basic state measurement, the transmission frequency of basic state measurement values, the modulation and coding method, and at least one of the transmission powers implemented between the user equipments, for the user equipment according to claim 1 or 2.

9. A process of implementing control in a set of user equipments, wherein each user equipment updates its own state based on the state measurement values and state prediction values obtained for a subset of the user equipments, at least one of the user equipments includes the following steps, namely, a step of storing at least some of the state measurement values obtained for each user equipment in the subset, or storing statistical values derived from the state measurement values obtained for each user equipment in the subset, Estimating a state prediction value for any one of the user devices in the subset by implementing a prediction model using pre-stored state measurement values or statistical values; Issuing a state estimation value for estimating the state at each respective time of estimation for each state of the user devices in the subset by combining the state measurement values and the state prediction values of the user devices in the subset related to the time of estimation; Selecting configuration parameters for controlling the operation of the resources of the user device based on the state estimation value; Similarly, updating the state of the user device based on the state estimation value; In a process of executing; At least one of the user devices called model improvement user devices operates alternately in a training mode and an idle mode; In the training mode, the prediction model is trained from the state measurement values or statistical values obtained for each user device in the subset, or from the state estimation values issued for each user device in the subset; The process further includes updating the prediction model implemented to estimate the state prediction value according to the prediction model trained in the training mode; The configuration parameters are adjusted such that the reliability of the state measurement values obtained for each user device in the subset is at a higher level than another level of reliability effective during a period dedicated to the idle mode, either preceding or following a period dedicated to the training mode. A process characterized by this.

10. The process according to claim 9, wherein in the idle mode, since the prediction model is pre-trained during the period dedicated to the training mode, the effectiveness of the currently implemented prediction model is evaluated.

11. For any one of the user devices in the subset, each state measurement value stored or used to issue one of the state estimation values is obtained from one or several basic state measurement values. For each user equipment of the subset, the frequency at which the basic state measurement value is obtained is adjusted such that the value of the frequency during the period dedicated to the training mode is higher than another value of the frequency effective during the preceding or subsequent period dedicated to the idle mode. The process according to claim 9 or 10.

12. The model improvement user equipment is the user equipment of claim 1 or 2. The process according to claim 9 or 10.

13. The state of any one of the user equipment in the subset includes the coordinate value of the position of the user equipment in the subset. The process according to claim 9 or 10.

14. The basic state measurement value is provided by at least one positioning system. The basic state measurement value for each user equipment in the subset is wirelessly transmitted to the model improvement user equipment or transmitted inside the model improvement user equipment. The process according to claim 11.

15. The model improvement user equipment switches from the training mode to the idle mode when the coincidence determination criterion of the training mode is satisfied between the state measurement value and the state prediction value inferred using the currently trained prediction model, or after a request to release some resources is received by the model improvement user equipment, or at the time issued by the scheduling algorithm. The process according to claim 9 or 10.

16. The model improvement user equipment has the following conditions, that is, The coincidence determination criterion of the idle mode is no longer satisfied between the state measurement value and the state prediction value used for issuing the state estimation value, After receiving information that a certain amount of the resources are available, After receiving information that the reliability of the state measurement value exceeds a predetermined threshold, At the time issued by the scheduling algorithm, When at least one of them occurs, it switches back from the idle mode to the training mode. The process according to claim 9 or 10.

17. The process according to claim 11, wherein the configuration parameter is selected based on the state estimate value and, optionally, also based on control requirements, in order to control at least one of the frequency of the basic state measurement, the transmission frequency of the basic state measurement value, the modulation and coding scheme, and at least one of the transmission powers implemented between the user devices.

18. A set of user devices adapted to perform control, wherein at least one of the user devices is according to claim 1 or 2.

Images