Methods for collecting data from a group of vehicles

Abstract

The present disclosure provides a method for collecting data from a group of vehicles. A method performed in a vehicle (100) includes: a) wirelessly downloading a data collection target from a host system (200), the data collection target including a data value metric, the data value metric being a function of assigning a data value to a data segment at a given point in time, the data value representing an amount of progress toward the atomic collection target; b) collecting data from a data source (110) in the vehicle (100) over time while the vehicle (100) is traveling; c) recording the collected data in a storage module (142); d) calculating a data value of the recorded data based on the data value metric; e) selecting high-value recorded data segments within a time window from the recorded data based on the calculated data value; and f) wirelessly uploading only the selected high-value recorded data segments to the host system (200).

Description

Technical Field

[0001] The present disclosure relates to the field of data collection by a group of vehicles performing data logging and uploading the logged data to a host collection system. The data logged by the logging vehicles can include raw data captured by sensors, data derived from the raw data (e.g., object detection data), and / or external data (e.g., GPS data, weather data, etc.). In the automotive industry, such data from vehicles is collected for algorithm development and / or validation, such as for training neural networks, particularly in advanced driver assistance systems (ADAS) and in the field of autonomous driving. Background Art

[0002] Known methods for recording data for ADAS algorithm development and validation use a single vehicle or a group of vehicles equipped with the necessary sensor hardware, typically including radar, lidar, and cameras. The vehicle's driver then drives around locations guaranteed to provide valuable and useful data for development and validation. As each vehicle drives, the sensor data is recorded onto a hard drive installed in the vehicle using a vehicle data recording device, such as disclosed in US Pat. No. 6,073,063A. After recording, the recording vehicle's hard drive is returned to a collection site for further distribution and storage.

[0003] The above-mentioned known methods of collecting data from vehicles have many disadvantages, such as long waiting times until data is available, uncertain value of the collected data which may not be as useful as expected, the need to select data after recording, etc.

[0004] In order to address the above-mentioned drawbacks of the known methods, there is a recent trend in the automotive industry to transmit data wirelessly without requiring physical access to the recording carrier.

[0005] Event data recording (EDR) is a technology used to record information related to vehicle collisions or accidents. An EDR is installed in the vehicle and records data during the critical moments of a collision, accident, or other significant event. The amount of recorded event data is minimal. Therefore, EDRs can be easily transmitted wirelessly from the vehicle to a central data collection and storage system.

[0006] US6505106B1 also discloses a group of vehicles, each of which is equipped with an in-vehicle data collection system that sends information to a main server via a cellular network. The main server includes a data repository for storing data collected from the group of vehicles.

[0007] Edge computing is a distributed computing approach that brings computation and data storage closer to where it's needed, improving response times and conserving bandwidth. In the field of autonomous driving, edge computing is considered a promising technology, offloading expensive computation from the vehicle's recording platform to a fixed data center by wirelessly transmitting data between the data center and the vehicle.

[0008] Despite all the developments discussed above, problems still exist in wirelessly transmitting data from a set of recording vehicles to a host data collection system because the amount of data to be transmitted is non-trivial and even in 5G cellular networks, wireless transmission is limited by the available bandwidth.

[0009] Therefore, there is a need to minimize the amount of recorded data that is wirelessly sent from a group of recording vehicles to a host data collection system, for example to allow real-time or near real-time uploading of recorded data and / or supplemental recordings with greater detail via a cellular communication network such as a 5G network. Summary of the Invention

[0010] The present disclosure relates to a method of collecting data from a vehicle and wirelessly uploading at least a portion of the collected data to a host data collection system, the method comprising the following steps performed by a vehicle data logging device in the vehicle:

[0011] a) wirelessly downloading a data collection objective from the host data collection system, the data collection objective including a data value metric, the data value metric being a function that assigns a data value to a data segment corresponding to a given point in time, the data value representing an amount of progress toward achieving the atomic collection objective;

[0012] b) collecting data from a data source in the vehicle over time while the vehicle is traveling;

[0013] c) recording the collected data in a storage module of the vehicle data recording device;

[0014] d) Calculate the data value of recorded data over time based on data value metrics;

[0015] e) selecting at least one high-value recorded data segment to be uploaded within a time window from the recorded data based on the calculated data value;

[0016] f) Wirelessly uploading only the selected at least one high value recorded data segment to the host data collection system.

[0017] In the present disclosure, at the vehicle level, within a storage module, one or more data segments (referred to as recorded data fragments) from the recorded data are selected as high-value data segments (in other words, relevant data segments) that achieve an atomic collection goal (in other words, an individual collection goal). This selection is performed based on the data value calculated using a data value metric downloaded from a host collection system. The selected data segments are defined by the time window of the recording. Only the selected data segments are then sent (uploaded) to the host collection system. This allows for a significant reduction in the amount of data sent from the recording vehicle to the host collection system without affecting the value of the data collected by the host collection system. Saving data storage and processing, for example, in the cloud is expensive in terms of hardware and power resources if the data is not of interest. Making decisions early on about which data is of interest (in other words, high value) and which data is not of interest (in other words, low value) can save significant resources in the long run. The present disclosure allows for such decisions to be made within the recording vehicle based on a predefined list of items that the host collection system considers of interest, corresponding to data value metrics, and a policy that defines how to balance data value according to different metrics.

[0018] In an embodiment, the first method further comprises the following steps: selecting at least one low-value recorded data segment to be deleted within a time window from the recorded data based on the calculated data value; and deleting the at least one low-value recorded data segment from the storage module.

[0019] This allows freeing up storage capacity of the recording module in the recording carrier.

[0020] In an embodiment, the data collection target includes m+1 data value metrics associated with different atomic collection targets, m≥0, and includes a collection strategy defining m+1 corresponding weights for the m+1 data value metrics, and the method further includes the following steps: calculating m+1 groups of data values ​​of recorded data over time based on the m+1 data value metrics respectively; and aggregating the m+1 groups of data values ​​by calculating the weighted sum of the m+1 groups of data values ​​according to the collection strategy to calculate the aggregated data value of the recorded data over time, and wherein the selection of recorded data segments is performed based on the aggregated data value of the recorded data over time.

[0021] A host collection system may be interested in collecting information about things in a certain directory (in other words, about different aspects). The data value metric corresponding to the atomic collection target allows measuring one of the things of interest. A collection target can include different data value metrics corresponding to different atomic (individual) collection targets. A collection strategy defines how to balance the differences between different things of interest in the recorded data. It assigns corresponding weights to different data value metrics (in other words, to different atomic collection targets).

[0022] The first method may further include the steps of wirelessly downloading update information of the data collection target from the host collection system, and updating the data collection target according to the update information.

[0023] As a host collection system's data collection progresses toward atomic (individual) collection goals, the value of the data may change. Typically, if a host collection system has already collected a large amount of data related to a specific thing or a specific scenario, collecting more data related to the same thing or the same scenario is inherently less valuable and less valuable than collecting data about rare things or rare scenarios. Updating data collection goals allows collection strategies and / or data value metrics to adapt to such changes in data value over time.

[0024] In an embodiment, the first method further includes:

[0025] the step of wirelessly receiving a reward value from said host data collection system upon successful uploading of said at least one high-value recorded data segment;

[0026] The step of providing the received reward value as a reward input to the reinforcement learning agent of the vehicle data recording device, wherein the reinforcement learning agent is responsible for deciding whether to upload the recorded data segment and is responsible for selecting the high-value recorded data segment.

[0027] In an embodiment, when a plurality of recorded data segments to be uploaded are selected from the recorded data, the first method further comprises the following steps:

[0028] Calculate the value density for each record data segment to be uploaded;

[0029] The multiple record data segments to be uploaded are placed in a sending priority queue, and the record data segments to be uploaded are sorted based on the value density.

[0030] The first method may further include the following steps:

[0031] determining a plurality of routes that the vehicle can take at a current time;

[0032] For each route, generating a route code that encodes information about the predicted value of the route for a plurality of metrics as a numerical value;

[0033] providing the route encoding and additional environmental information independent of the route to the reinforcement learning agent, the reinforcement learning agent deciding whether to change the current route and selecting one of the determined routes to optimize reward when appropriate;

[0034] recording data from a data source in the vehicle over time as the vehicle travels along a selected route (i.e., a route selected from a plurality of determined routes, or a current route if a decision is made not to change the current route);

[0035] uploading at least a portion of the recorded data to the host data collection system and receiving a reward in return from the host data collection system;

[0036] The reward is provided to the reinforcement learning agent.

[0037] Path planning allows the vehicle to take paths (roads) that are likely to maximize the value of the recorded data, rather than driving randomly. Thus, data is recorded more efficiently with the atomic collection goal in mind.

[0038] The present invention also relates to a method of collecting data from a group of recording vehicles, the method being performed by a host collection system wirelessly connected to the group of recording vehicles, the method comprising the steps of:

[0039] At the data collection management module, providing a data collection goal, the data collection goal including a data value metric, the data value metric being a function that assigns a data value to a data segment corresponding to a given point in time, the data value representing an amount of progress toward achieving the atomic collection goal;

[0040] wirelessly downloading the data collection target to the set of recording vehicles;

[0041] wirelessly receiving recorded data segments from the set of recording carriers; and

[0042] The received recorded data segments are stored in a storage facility.

[0043] In an embodiment, in the step of providing data collection targets, the data collection management module provides m+1 data value metrics associated with different atomic collection targets, m≥0, and a collection strategy defining m+1 corresponding weights of the m+1 data value metrics.

[0044] The second method may further include the following steps:

[0045] at the data collection management module, updating the data collection target in dependence upon the recorded data segments that have been received from the set of recorded carriers, and

[0046] Updated information of the data collection target is wirelessly downloaded to the set of recording vehicles.

[0047] A data value metric may be generated to assign a normalized data value.

[0048] In an embodiment, the method further comprises the steps of calculating a virtual value for a recorded data segment uploaded from a vehicle in the set of recording vehicles based on the data collection target, generating a reward value dependent on the virtual value, and wirelessly downloading the reward value to the vehicle.

[0049] The present disclosure also relates to a vehicle data recording device, the vehicle data recording device comprising:

[0050] at least one interface that receives logged data from a data source in the vehicle and transmits segments of the logged data for wireless upload to a host data collection system;

[0051] a storage module, wherein the storage module stores the recorded data;

[0052] A unit for performing the steps of the aforementioned method.

[0053] The present disclosure also relates to a vehicle including the above-mentioned vehicle data recording device.

[0054] The present disclosure also relates to a host data collection system, comprising:

[0055] a first interface that wirelessly interfaces with a set of recording carriers;

[0056] a second interface that interfaces with a storage facility; and

[0057] A unit for performing the steps of the aforementioned method.

[0058] The present disclosure also relates to a distributed system comprising a set of recording vehicles as defined above, a host data collection system as defined above wirelessly connected to the set of recording vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Other features, objects and advantages of the present disclosure will become more apparent from a reading of the detailed description of non-limiting embodiments made with reference to the accompanying drawings.

[0060] Figure 1 A distributed system for collecting data from a recording vehicle into a host collection system is shown according to an embodiment.

[0061] Figure 2 A functional block diagram of an in-vehicle data recording device according to an embodiment is shown.

[0062] Figure 3 Flowcharts are shown of a first method performed by an in-vehicle data logging device to collect data from a vehicle and wirelessly upload at least a portion of the collected data to a host data collection system, and a second method performed by a host data collection system to collect data from a group of vehicles.

[0063] Figure 4 A flow chart illustrating a process of updating a data collection target according to an embodiment.

[0064] Figure 5 Represents recording data value over time and aggregating data value over time according to three data value metrics according to an embodiment.

[0065] Figure 6 Shown based on Figure 5 Use the aggregated data value to select the record data segments to upload and the record data segments to delete. DETAILED DESCRIPTION

[0066] Figure 1 A distributed system for collecting data from a group or multiple recording vehicles 100 by a host data collection system 200 is shown.

[0067] The recording vehicle 100 may wirelessly communicate with the host collection system 200, for example, via a mobile communication network 300 such as a 5G network.

[0068] The data from each recording vehicle 100 may include different types of data or recorded data, such as raw data, derived data, and / or external data.The data collected from the vehicle 100 is recorded over time, typically while the vehicle is traveling.

[0069] Raw data (also referred to as atomic data or source data) is unprocessed data. Raw data includes, in a non-limiting manner, sensor data from sensors installed in the vehicle 100 (such as radar, lidar, and cameras), GPS data, speed data, yaw rate data, etc.

[0070] Derived data includes data derived from raw data through data processing. Data processing can be performed by hardware devices and / or software devices in the vehicle 100 that execute processing operations and / or algorithms (e.g., machine learning algorithms or any other type of algorithms). Some in-vehicle hardware and software components can perform object detection based on radar, lidar, or camera data (with or without sensor fusion), semantic segmentation on camera data, encoding of signatures describing various aspects of environmental information (in the vehicle environment), calculation of data structures, and / or statistical information output from object detection modules or other algorithms.

[0071] External data includes data from sources external to the vehicle 100. Such data can be obtained by calling an API (for Application Programming Interface) of a third-party data provider via the mobile communication network 300. Non-limiting examples of such data include traffic data (e.g., traffic flow, construction areas, etc.), weather service data, time-related data (e.g., national holidays, weekdays, weekends, etc.), map service data (e.g., road layout information, satellite imagery, mobile network coverage, etc.), current transmission rates through the mobile communication network 300, data from environmental sources (e.g., stores, schools, public transportation stations, parks, etc.), data from routing services, and data from the host collection system 200.

[0072] The host collection system 200 is responsible for providing a data collection target T0, downloading it to a group of recording vehicles 100, and in return, collecting recorded data segments (also referred to as “recorded data segments”) collected by the group of recording vehicles 100.

[0073] The data collection target T0 defines one or more atomic (individual) collection targets c j , where j = 0, ..., m, and includes the atomic collection target c j One or more data value metrics related to Data value measurement is a function that assigns a data value (in other words, a number) to a data segment corresponding to a given time point i (to a data point for time i), the data value representing the achievement of the corresponding atomic collection goal c j In other words, the value of each data is measured Allows measurement towards the atomic collection target c j progress. It gives hints about the value of the data segment. Atomic collection target c j It can be a quantitative target defined by a given target amount or number. Illustrative and non-limiting examples of atomic collection targets are given below:

[0074] - kilometers on specific roads (i.e. highways, cities, ...);

[0075] - Specific recorded time spent in specific scenarios (i.e. X number of hours spent in traffic jams, daytime, twilight, nighttime, weekends, rush hour...);

[0076] - Multiple unique objects of a specific class (i.e., bicycles, trucks, pedestrians, ...);

[0077] - Multiple specific actions (i.e. overtaking, lane changing, etc.);

[0078] - Multiple specific road scenarios (i.e., circuitous routes, specific intersection constellations, passenger intersections, construction areas, tunnels, bridges, ...);

[0079] - Multiple scenarios selected through active learning methods or predictive models;

[0080] - Multiple scenarios selected through outlier detection (in other words: statistically rare detections) and other means of collecting non-representative data points (in other words: single sensor measurements or combinations at a given point in time).

[0081] Optionally, the atomic collection target c can be constrained in some way j To diversify the data collection. For example, a constraint could be to perform data collection across multiple regions (countries, cities, ...). Another constraint could be to limit data collection along routes that have been recorded in the past.

[0082] atomic collect objective c j This can be in the form of code to be executed.

[0083] The data collection target T0 includes m+1 different atomic collection targets c j (form atomic collection target c j Group C: C={c0,…,c m m+1 data value metrics related to In the case of (m≥0 and value index j from 0 to m), the data collection target T0 also includes a collection strategy P, which defines how to collect different atomic collection targets c in the data collection target T0 j Weighted. Strategy P can define how to combine multiple atomic metrics or value metrics. Collect the corresponding weights that strategy P will apply Assigning different data value metrics In other words, the collection strategy P is to collect the target c with m+1 atoms j Related m+1 data value metrics Define m+1 corresponding weights 0≤j≤m. The collection strategy can be expressed as Preferably, the set of weights Make

[0084] As explained previously, each collection value metric is a function that assigns a data value to a recorded data point for time i (in other words: to a data segment for a given time point i) in a record R. A record R is a set of data captured by the recording vehicle 100 and contains raw data provided by sensors in the recording vehicle 100 and / or data from other data sources in the recording vehicle 100 (including derived data and / or external data, as explained later).

[0085] In addition, with the collection strategy P at all time points {t0,...,t n} and the atomic collection target {c0,...,c m The value V(R) (also called "record value") of the record R aggregated on} can be expressed as follows:

[0086]

[0087] Alternatively, the data value metric may be weighted for each policy and the value V(R) may be calculated as an integral over time.

[0088] In certain embodiments, the value metrics are collected Values ​​within a normalized range (i.e., [0, .., 1.]) can be generated. In this case, the data value assigned to the recorded data point at time i is between 0 and 1 (inclusive). In a first illustrative example, the atomic collection target is the number of overtaking actions, and the value metric takes the value 1 when an overtaking action occurs and the value metric takes the value 0 when no overtaking action occurs. In a second illustrative example, the atomic collection target is the number of bicycles, and the value metric gives a value representing the number of bicycles in the scene. The value is normalized between 0 and 1 by a mapping function, based on the maximum number of bicycles, or alternatively based on the average number. Normalization facilitates combining or aggregating data values ​​calculated by different value metrics, as explained later.

[0089] As will be explained in more detail in the description of the method performed on the vehicle side, the collection strategy P is used to temporally aggregate m+1 sets of data values ​​generated by the m+1 data value metrics for the record R, respectively, and thereby generate a set of global or aggregated data values ​​for the record R over time.

[0090] Optionally, the data collection target can be updated by the host collection system 200, as will be described later. As the host collection system collects recorded data from a group of vehicles 100, the value of the recorded data can change over time. Generally, if the system has already collected a large amount of data related to a particular atomic collection target, the value of new data related to the same atomic collection target typically decreases. For example, consider collecting data to develop a machine learning model for a vehicle detector. If the vehicle detector is new, the value of each collected vehicle is significant, as the model needs to be trained on each vehicle to become adept at detecting vehicles. If the model already detects vehicles well, collecting more vehicles will not effectively improve the model. In this case, collecting data on the distribution of collected vehicles, as well as collecting more data on, for example, rare vehicles, would be of interest to aid in developing a better model. For developing a machine learning model, the value of data decreases over time until only slight improvements in the model can be achieved. At this point, a large amount of data is typically required to achieve even marginal improvements. The atomic value metric, and more generally, the atomic value metric for the collection target T0, can be updated based on the amount of data already collected.

[0091] In this disclosure, “k” represents the update index (integer) of the data collection target, k ≥ 0. In other words, by convention, the continuously updated data collection targets are marked as T k (i.e., T0, T1, ...).

[0092] Each recording vehicle 100 has a plurality of data sources (raw data, derived data and / or external data) 110 , a radio transmitter-receiver 120 , a central processing unit 130 and a data recording device 140 .

[0093] The data source 110 includes hardware components and software components that provide raw data and / or derived data, optionally external data. The data source 110 can be of different types, for example:

[0094] - sensors (e.g., radar, lidar, and / or camera) that provide raw data (e.g., radar data, lidar data, and / or camera data),

[0095] - a calculator for processing input data, including raw data, and outputting derived data,

[0096] - a machine learning or algorithmic module that performs one or more tasks (e.g., object detection) based on input data and outputs derived data,

[0097] - Functional modules for receiving external data via the network 300, etc.

[0098] The radio transmitter-receiver 120 has a function of transmitting and receiving radio signals (including information and / or messages and / or data) through the mobile communication network 300 .

[0099] All elements 110 , 120 , 140 of the recording vehicle 100 are connected to a central processing unit 130 which controls their operation.

[0100] The data recording device 140 is responsible for collecting (in other words, receiving) data from a data source in the vehicle 100 and wirelessly uploading (in other words, sending) at least a portion of the collected data to the host data collection system 200 via the mobile communication network 300. It has hardware units and software units (in other words, hardware components and software components) for implementing the first method of collecting data from the recording vehicle 100 and wirelessly uploading at least a portion of the collected data to the host data collection system 200, which will be described later.

[0101] In an embodiment, information exchange between the data logging device 140 and the host data collection system 200 is performed via the in-vehicle radio transmitter-receiver 120. Alternatively, the data logging device 140 may include a radio transmitter-receiver to exchange information directly with the host data collection system 200.

[0102] like Figure 2 As shown, in an embodiment, the in-vehicle data recording device 140 has an input / output interface 141 for interfacing with the vehicle 100 , and a storage module or buffer 142 .

[0103] Furthermore, the in-vehicle data recording device 140 includes hardware and software components (in other words, hardware components and software components) for executing the steps of the first method for collecting recorded data from the vehicle 100, which will be described later and executed by the data recording device 140. In certain embodiments, the hardware and software components include a recorded data management module 143, a reinforcement learning agent component 144, and a central processing unit (CPU) 145. Alternatively, the in-vehicle data recording device 140 may utilize the CPU 130 of the vehicle 100 (instead of the CPU 145).

[0104] The input / output interface 141 exchanges information with the vehicle 100 (e.g., data from a data source within the vehicle, messages received from the host collection system 200 via the transmitter-receiver of the vehicle 100, data uploaded to the host collection system 200 via the transmitter-receiver of the vehicle 100, etc.), as will be explained later. For example, the input / output interface 141 communicates with the central processing unit 130 in the vehicle 100.

[0105] The storage module 142 is responsible for storing the recorded data collected from the vehicle 100 in memory. In an embodiment, the storage module 142 has a hardware portion 142A, such as a buffer, which is a memory for storing (at least temporarily) the recorded data from the vehicle 100. Optionally, the storage module 142 also has a software portion 142B that is responsible for taking certain actions on the recorded data stored in the memory 142A. The software portion 142B runs on the CPU 145. The memory 142A has a given storage capacity and specific utilization attributes that are provided to the reinforcement learning agent component 144. The storage module 142 can access recorded data value estimates associated with the recorded data stored in the memory 142A, as described below.

[0106] The reinforcement learning agent component 144 is a software component that implements the reinforcement learning algorithm and can be run on the central processing unit 145.

[0107] Depend on Figure 2 The task or function of the reinforcement learning agent component 144, schematically represented by block 150 in FIG, is to decide whether to upload or delete recorded data segments and to select recorded data segments from the records R to be uploaded to the host collection system 200 and to be deleted from the storage module 142. In terms of resources and processing power, storing and processing data in the storage facility 400 is expensive if the data is not of interest. The earlier a decision is made about which data is useful and which is not, the more resources and processing power can be saved in the long run. The present disclosure allows for the decision of which data segments to upload and which to delete to be made based on data value metrics and collection strategies that define how to balance data value according to different metrics.

[0108] The reinforcement learning agent component 144 has an interface to the storage module 142 and selects recorded data segments to be uploaded (in other words, sent) to the host collection system 200 and deleted from the storage module 142. In this specification, each data segment selected (for upload or deletion) is referred to as a recorded data segment. It is a data segment corresponding to a time window of the record R. A recorded data segment is defined by a time window within the time range of the record R. In some embodiments, the selection of recorded data segments (for upload or deletion) complies with one or more of the following constraints:

[0109] - Minimum length of a recorded data segment (in other words: minimum duration of the time window of a recorded data segment);

[0110] - the maximum length of a recorded data segment (in other words: the maximum duration of the time window of a recorded data segment);

[0111] - a buffer (in other words: the time before and after the region of interest, such as a tunnel, for example a few seconds, such as 5 seconds);

[0112] - Each recorded data segment must be continuous (in other words: each recorded data segment includes a part of the recorded data that is continuous over time);

[0113] - Different recorded data segments do not overlap.

[0114] Regarding the (time) length of the recorded data segments to be uploaded, it may be desirable not to collect very short information segments in time, such as data segments lasting only 1 second. For example, a recorded data segment must be within a time window of more than x seconds, where x ≥ 10 seconds, in order to be selected for upload.

[0115] In one embodiment, the selection of recorded data segments to upload or delete is performed automatically by the reinforcement learning agent component 144 based on the data collection target T0 downloaded from the host collection system 200, as explained later in the description of the method for collecting data. In another embodiment, the selection can be performed by a reinforcement learning (RL) agent. Alternatively, a fixed algorithm sets thresholds and constraints to locally optimize whether it is best to upload or not upload.

[0116] The goal of the reinforcement learning agent 144 is to maximize an objective function (in other words, a reward function) corresponding to the received data collection goal T0. The reinforcement learning agent 144 attempts to maximize one or more received rewards, which are partially related to the received data collection goal. It can do so by receiving different types of rewards. The following list gives illustrative and non-limiting examples of actions that are rewarded:

[0117] i) successfully uploading a record data fragment (e.g., a large reward relative to the value density of the uploaded record data fragment);

[0118] ii) completing the upload (an additional reward weighted by the total value of the recorded data segment, which provides an incentive to complete the upload rather than switching between valuable segments);

[0119] iii) The process of uploading recorded data fragments (continuous small rewards proportional to the value density of the uploaded data fragments);

[0120] iv) Maintaining good value density (media rewards) for recorded data in the recording buffer 142A in situations where the bandwidth of the mobile network 300 is limited and temporarily prevents the upload of recorded data segments.

[0121] Regarding items i), ii), and iii) above, a distinction can be made between completing an upload and the process of uploading it. If a reward is only received at the end of an upload, it can be difficult to associate the reward with the action taken to upload the data. When a small reward is also given along with the uploaded data, the RL agent receives an immediate reward, indicating that it is on the right track. Finally, we are only interested in fully uploaded recording segments, which is why the model receives a large reward after successfully uploading a recording data segment.

[0122] The value density of a data segment can be defined as the value of the data segment per unit time.

[0123] The reinforcement learning agent component 144 is rewarded only for recorded data segments that have been successfully uploaded. Not uploading anything gives neither reward nor penalty. If the vehicle 100 is traveling in an area with a poor connection to the mobile network 300, the vehicle 100 may not be able to upload much data and therefore may not be able to collect much reward.

[0124] Additionally, and optionally, the reinforcement learning agent component 144 can be penalized, for example, in the following situations:

[0125] - the record buffer may experience buffer overflow (large penalty, but optional);

[0126] - situations such as poor mobile network connection occur (small penalty, but optional);

[0127] - Keeping data segments in the record buffer 142A that are corrupted in any way, or are too short to meet other minimum requirements like minimum contiguous record length.

[0128] Other types of rewards / penalties may be used.

[0129] Rewards and penalties may be provided to the reinforcement learning agent component 144 by the host collection system 200 and / or the log data management module 143 .

[0130] Instead of the reinforcement learning agent component 144, a component that implements an optimization algorithm may be used.

[0131] The recorded data management module 143 is a software module that manages the recorded data stored in the storage module 142 .

[0132] The record data management module 143 has the function of storing the current (valid) data collection target T k Memory 146.

[0133] The first function of the log data management module 143 is to calculate the data value of the log data stored in the storage module 142 over time based on one or more data value metrics and the collection policy (if applicable) for the current data collection target T0. The log data management module 143 has a value calculation component 147 for calculating the data value of the log data. The component 147 can be a software component running on the CPU 145. At the current data collection target (e.g., T0), m+1 data value metrics are defined. and weight Assigning value to data In the case of collection strategy P, the computing component 143 is responsible for using m+1 data value metrics respectively To calculate the value of m+1 groups of data over time for the recorded data (stored in the storage module 142), and then calculate the value of the corresponding weights The weighted sum of the m+1 sets of data values ​​is aggregated according to the collection strategy P to obtain a set of aggregated data values ​​of the recorded data over time. Figure 5 An illustrative example of three sets of data values ​​calculated according to three different metrics and a set of aggregated data values ​​based on policy P for a record R is shown in FIG.

[0134] The second role of log data management module 143 is to take actions such as uploading or deleting the log data in storage module 142 based on the decisions and selections made by reinforcement learning agent component 144 (in other words, based on the output from component 144). Thus, log data management module 143 and reinforcement learning agent component 144 collaborate to manage the log data stored in memory (e.g., buffer) 142A. In one embodiment, log data management module 143 includes two software components to manage the uploading and deletion of log data segments: a send priority queue component 148 and a garbage collector component 149.

[0135] The transmit priority queue component 148 is responsible for controlling the transmission of the recorded data segments (recorded data segments) stored by the storage module 142 for upload to the host collection system 200 via the mobile communication network 300, here via the in-vehicle transmitter-receiver 120. The component 148 can provide at least some of the following features:

[0136] - Data transfer speed via mobile network connection;

[0137] - a sending priority queue with an ordering of the recorded data segments to be uploaded (in other words: sent based on the respective data value density);

[0138] - an estimate of the remaining time to complete uploading the recorded data currently to be uploaded (in other words: the recorded data fragment currently in the sending queue);

[0139] - the sending of a recorded data segment that has already started should not be interrupted by the sending of another segment of higher value density data in the queue (but this is optional);

[0140] During and / or after successfully uploading a recorded data segment to the host collection system 200, the send priority queue component 148 rewards the reinforcement learning agent component 144 in proportion to the total value of the recorded data segment;

[0141] - After uploading the record data segments to the host collection system 200, the send priority queue component 148 may then provide the uploaded record data segments to the garbage collector component 149 to free up capacity in the buffer 142A.

[0142] The garbage collector component 149 is responsible for deleting record data segments (record data segments) from the storage module 142 to free up its storage capacity. Record data segments can be explicitly registered for deletion by:

[0143] - by the reinforcement learning agent component 144 due to a low recorded data value estimate for the recorded data segment;

[0144] - by the send priority queue component 148 after uploading the recorded data segment to the host collection system 200; or

[0145] - by the storage module 142 if the minimum length requirement of the recorded data segment cannot be met; or

[0146] - by the storage module 142 in an emergency situation where overflow of the buffer 142A is imminent (in which case the reinforcement learning agent component 144 will be penalized for poor record buffer management); or

[0147] - When the recorded data is unhealthy (this can happen when the sensor data stream is corrupted).

[0148] like Figure 1 As shown, the host collection system 200 may include a server and has a first interface 210 connected to the mobile communication network 300, the first interface interfaces with a group of wireless recording vehicles 100 through the mobile communication network 300, and a second interface 220 interfaces with the storage facility 400.

[0149] The first interface 210 is connected to the mobile network 300 .

[0150] The second interface 220 is connected to the storage facility 400 .

[0151] The storage facility 400 may be any type of storage system suitable for storing large amounts of data. In an embodiment, the storage facility 400 may have computer resources for storing data, such as in the cloud. It may also be a formatting module responsible for formatting data for storage.

[0152] The storage facility 400 may be external to the host collection system 200 or internal to the host collection system 200 .

[0153] Furthermore, the host collection system 200 includes hardware units and software units (in other words, hardware components and software components) for executing the steps of the second method for collecting data from a group of vehicles, which will be described later and is executed by the host collection system 200. In an embodiment, the hardware units and software units include a data collection management module 230 and a control processing unit (or processor) 240.

[0154] The data collection management module 230 is a software component running on the CPU 240. It has a target provider 231, which is used to provide a data collection target T0 and, in some embodiments, to update the current data collection target.

[0155] The data collection management module 230 may also include a reward generator 232, which is responsible for calculating the virtual value of the recorded data segment uploaded from the recording vehicle 100 based on the data collection target, calculating the reward value based on the virtual value, and wirelessly downloading the reward value to the recording vehicle 100 via the mobile network 300.

[0156] Now refer to Figure 3 A first method of collecting data from a recording vehicle 100 while the vehicle 100 is traveling and wirelessly uploading at least a portion of the collected data to a host data collection system 200, and a second method of collecting data from a group of recording vehicles 100 performed by the host collection system 200, according to a specific embodiment, are described. The first collection method is described below for one of a plurality of recording vehicles 100. All recording vehicles 100 perform a similar first method.

[0157] In the initial step S1, the data collection management module 230 of the host collection system 200 provides an initial data collection target T0. The data collection target T0 includes m+1 data value metrics The data value metric is used to assign data value to recorded data over time to measure the effectiveness of achieving m+1 corresponding atomic (individual) collection goals c j progress in, j ranges from 0 to m. The m+1 groups correspond to the atomic collection targets {c0,…,c m} is marked as C. In the embodiment currently described, the data value metric are normalized and assigned values ​​in the range [0, ..., 1]. In addition, the data collection target T0 includes the corresponding weights Assigning value to data The collection strategy P. m+1 weights The sum of the host collection system 200 is similar to the one containing the data value metric and collection strategy P's files.

[0158] In step S2, the host collection system 200 wirelessly sends an initial data collection target T0 to a group (or multiple) of vehicles via the mobile communication network 300. In an embodiment, the data collection target T0 is sent to each recording vehicle 100 according to a request from the vehicle 100.

[0159] In step S3, in the vehicle 100, the vehicle data recording device 140 downloads (in other words, receives) the data collection target T0 from the host collection system 200 via the mobile network 300. In this embodiment, the data collection target T0 is received via the radio transmitter-receiver 120 of the vehicle 100 and the I / O interface 141 of the recording device 140.

[0160] In step S4, while the vehicle 100 travels along the route, the vehicle data recording device 140 collects (in other words, receives) data over time from a data source in the vehicle 100. The data collected over time forms a record R.

[0161] As data from the data source in the vehicle 100 is received by the vehicle data recording device 140 , the received data is stored in the storage module 142 (here, in the buffer memory 142A) in step S5 .

[0162] In step S6, the value calculation component 147 measures the value of the data. To calculate the data value of the recorded data recorded in the memory 142A over time, j ranges from 0 to m. The result of calculation S6 includes the following: The value of m+1 sets of data recorded over time Figure 5 Figures a), b), and c) show examples of three sets of data values ​​according to metric 1, metric 2, and metric 3. As an illustrative and non-limiting example, metric 1 relates to an atomic collection target c1 consisting of X1 pedestrians, metric 2 relates to an atomic collection target c2 consisting of X2 bicycles, and metric 3 relates to an atomic collection target c3 consisting of X3 motor vehicles.

[0163] In step S7, the calculation component 147 calculates the value of the m+1 group of data The weighted sum of the data is used to aggregate the value of m+1 groups of data according to the collection strategy P To calculate the global data value of recorded data over time, denoted as V C,P (t).

[0164] Based on data value measurement and collection strategy P, the vehicle 100 computes each metric in the metrics to obtain atomic data values, such as Figure 5 As exemplarily shown in a), b) and c) of FIG, the calculated values ​​are combined (aggregated) by weighted sum, and the overall (aggregated) value of the record R at each time point i is obtained, such as Figure 5 As shown in d).

[0165] In steps S3 to S7, the data recording device 140 in the vehicle 100 downloads the collection strategy P and the data value metric c j , for example as code to be executed. Then the data value metric c is run in the vehicle 100 j , the data value is obtained over time according to different metrics in a normalized manner. Values ​​between 0 (not interesting, very low value) and 1 (very interesting, very high value) are assigned over time to the recorded data points for each time point i. All individual data value metrics c are calculated in the vehicle 100 j , and then weighted according to the policy P to obtain the aggregated data value over time, which evaluates the records R within the vehicle 100.

[0166] In the selection step S8 of the first method, the reinforcement learning agent component 144 decides whether to upload or delete the recorded data segment, and selects the recorded data segment for uploading or deletion.

[0167] Based on the data value calculated in steps S6 and S7, preferably based on the aggregated data value V C,P (t), performs the decision and selection of the recorded data segments (upload and / or delete). In this embodiment, each recorded data segment is continuous and defined by a time window between a minimum length and a maximum length. In addition, different selected data segments do not overlap. Figure 6 An example of a high-value recorded data segment (data segment A) to be uploaded and a low-value recorded data segment (data segment B) to be deleted is shown.

[0168] The recorded data segments to be uploaded include high-value recorded data segments within a time window. For example, the value threshold is fixed, and sub-selection of the recorded data buffer for upload is performed based on the fixed value threshold. Alternatively, there is no fixed value threshold, but the most valuable x% of data are selected for upload (e.g., x=20%), regardless of their absolute recorded value. More generally, the segment formation algorithm can be considered a black box that extracts candidate segments for upload. The RL agent can decide to prioritize segments based on segment content, value, and duration.

[0169] In the case where all recorded data are of high value, they are uploaded as a whole. In other words, in this case, the reinforcement learning agent component 144 selects all recorded data R to upload.

[0170] The recorded data segments to be deleted include low-value recorded data segments within the time window. The recorded data segments to be deleted can be selected in a similar manner (as described above). If the storage capacity of the data recording device 140 is sufficient, it can be decided not to delete any recorded data segments (if the data has a sufficiently high value). Low-value data is deleted first because maintaining and processing this data requires money, computing resources, and energy.

[0171] After selection step S8, some data segments of the data record R do not belong to any recorded data fragment to be uploaded or deleted. They can be retained in the recording buffer in advance. Even though these data segments may not be valuable enough for direct upload due to 5G bandwidth limitations, if only low-value data is recorded in the near future of the recorded drive, these data segments may still be candidates for upload in the future. The data retained in the recording buffer after the recorded drive is still valuable and can be uploaded later. Therefore, the RL agent should also be rewarded for keeping valuable data in the recording buffer. The RL agent only needs to prevent the buffer from overflowing. Therefore, low-value fragments need to be deleted. This is encouraged by severely penalizing the RL agent for buffer overflows.

[0172] The reinforcement learning agent module 144 makes the decision (in other words: determines) what to upload and what to delete. In embodiments, other factors may also be considered, such as network availability, traffic, weather, or any other information related to the environment and / or situation of the vehicle 100 that is relevant to making the decision. The vehicles 100 have limited capacity to save recorded data and limited processing and computing power. Furthermore, due to the limited bandwidth available in the mobile network 300, it is not possible to upload very large amounts of recorded data from a group of recording vehicles 100 to the host collection system 200 in real time or near real time. The recording vehicles 100 determine what data to buffer and what data to discard. This decision is made by the reinforcement learning agent component 144. Alternatively, it can be made by another type of algorithm or any method that evaluates the data at various points in time to understand where the time windows of interest are as opportunities to save and upload data.

[0173] Steps S6 through S8 may be performed in real time or near real time as log data is being collected and stored in memory 142A.

[0174] In step S8, the reinforcement learning agent 144 outputs a decision on whether to upload / delete, first information about the recorded data segments to be uploaded, and optionally, second information about the recorded data segments to be deleted. The first information is provided as input to the sending priority queue component 148. The second information is provided as input to the garbage collector component 149.

[0175] Optionally, in step S9, the sending priority queue component 148 generates and manages (in other words, controls) a sending priority queue of the recorded data segments to be uploaded, and in step S10, controls the sending of the recorded data segments to be wirelessly uploaded according to the priority queue. The priority queue is optional. In step S9, the component 148 can calculate the value density of each recorded data segment to be uploaded (in other words, the data value per unit time) and create a sending priority queue by sorting the recorded data segments based on their respective value densities. Therefore, the sorting of the recorded data segments to be uploaded is based on the corresponding data value density of the recorded data segments. In step S10, the recorded data segments to be uploaded are wirelessly sent to the host collection system 200 in the order of the queue via the mobile communication network 300, here via the transmitter-receiver 120 in the vehicle 100.

[0176] In step S11 , under the control of the garbage collector component 149 , the recorded data piece to be deleted is deleted from the memory 142A.

[0177] Furthermore, in step S11 , after each recorded data segment is successfully sent for upload, the uploaded recorded data segment may be deleted from the memory 142A under the control of the garbage collector component 149 .

[0178] In step S20, the host collection system 200 receives the successfully uploaded recorded data segments from the recording vehicle 100. In step S21, the host collection system 200 stores the received recorded data segments in the storage facility 400.

[0179] In step S22, the reward generator 232 calculates a virtual value for each recorded data segment uploaded from the vehicle 100 based on the data collection target T0. The virtual value of the recorded data segment may be the value density of the recorded data segment. Alternatively, the virtual value of the recorded data segment may be calculated using equation (1) applied to the recorded data segment.

[0180] Then, in step S23, reward generator 232 may generate a reward value for the uploaded recorded data segment based on the virtual value of the uploaded recorded data segment. For example, the reward value may include a first reward component proportional to the value density of the uploaded recorded data segment. Furthermore, the reward value may include a second reward component corresponding to successive small rewards for rewarding the process of uploading the recorded data segment.

[0181] In step S24 , the reward value generated in step S23 is wirelessly downloaded from the host collection system 200 to the vehicle 100 via the mobile communication network 300 .

[0182] The vehicle data recorder 140 receives the downloaded reward value (here via the transmitter-receiver 120 of the vehicle 100) and provides the received reward value to the local reinforcement learning agent component 144 as a reward input for learning in step S30. Thus, the reinforcement learning agent component 144 increasingly performs the task of selecting the recorded data segments to be uploaded and uploading them.

[0183] As the vehicle 100 travels, all collected data can be new and improved training data for the reinforcement learning agent component 144. The more the vehicle 100 travels, the more efficient the reinforcement learning agent component 144 becomes. For example, the reinforcement learning agent component 144 can learn the correlation that school buses only occur during school hours, or it can learn the correlation that children are detected at specific times (on the way to or from school).

[0184] Reinforcement learning agent component 144 is initially trained to manage record buffer 142A and effectively decide whether to upload or optionally delete pre-selected recorded data segments, or whether no action should be taken. Training is performed to ensure that data recorder 140 uploads valuable recorded data segments, retains relevant recorded data segments in buffer 142A, prevents buffer overflow, and optionally deletes low-value or defective recorded data. Training can use historical record data similar to the recorded data used in inference mode as input training data, a reward function to determine different types of rewards / penalties as described above, and output training data that includes the different types of rewards / penalties determined by the reward function.

[0185] The collection target T0 may be updated by the host collection system 200 . Figure 4 The process of updating the collection target T0 according to an embodiment is shown. As the host collection system 200 collects collection data selected based on the collection target T0 from a group of vehicles 100, the host collection system 200 makes decisions to update the data collection target T0 based on the recorded data already received from the group of recording vehicles 100. The update of the current data collection target T0 reflects the fact that the inherent value of the recorded data (here, recorded data segments) related to one or more specific atomic collection targets changes over time. In fact, when the data segment related to a given atomic collection target is scarce and the collection demand is high, the value of the data segment can be expected to be high, but as other similar data segments related to the same atomic collection target have been collected by the host collection system 200, the value becomes lower. In the first illustrative example, the initial collection target T0 includes a first atomic collection target c0 consisting of X0 pedestrians and a second atomic collection target c1 consisting of X1 bicycles. If later, the host collection system 200 has received many pedestrians but few bicycles in the recorded data segments (fragments) uploaded from a group of vehicles 100, the host collection system 200 may decide to reduce the weight of the first atomic collection target c0 in the updated collection target T1 And increase the weight of the second atomic collection target c1 In a second illustrative example, the initial collection target T0 includes a first atomic collection target c0 consisting of X vehicles. If, later on, the host collection system 200 receives many vehicles in recorded data segments uploaded from a group of vehicles 100, the host collection system 200 may decide to add constraints to the first atomic collection target c0, such as constraints on the types of vehicles to be collected, to collect only rare vehicles. Alternatively, the data value metric itself can model the aforementioned behavior.

[0186] The updating of the collection target T0 consists in modifying one or more components of the collection target T0 depending on the recorded data segments that have been collected from the recording carrier 100. For example, the collection strategy P can be updated. Alternatively or additionally, one or more atomic data value metrics can be taken from the current collection target T0. and / or one or more new atomic data values ​​may be measured Added to the current collection target T0. Thus, an update may include, in a non-limiting manner, one or more of the following changes:

[0187] - Atomic collection of targets {c0,…,c m Some weights of modifications;

[0188] - Added specific atoms collection to Objective-C j relevant constraints;

[0189] - Add new atomic collection for objective-c m+1 ;

[0190] - Remove atomic collect objective c j .

[0191] To update the current data collection target T0, the host collection system 200 analyzes the recorded data segments that have been received from the set of vehicles 100 to follow the progress toward the current collection target T0. Then, in step S41, the host collection system 200 determines one or more changes to be applied to the current collection target T0 based on the analysis in step S41. In other words, the host collection system 200 determines the updates to be applied to the current collection target T0. For example, the host collection system 200 modifies the atomic collection target {c0, ..., c m Some weights of

[0192] In step S42 , the master collection system 200 downloads update information to a group of vehicles 100 to notify the collection target T0 of the update. For example, the update information includes a new collection strategy or a new weight.

[0193] In step S43, the data recording device 140 wirelessly downloads and receives the update information via the transmitter-receiver 120 in the vehicle 100, and in step S44, updates the data collection target T0 stored in the memory 146 according to the update information. After the update, the new data collection target stored in the memory 146 is T1.

[0194] After updating the collection target, the data recording device 140 continues to perform steps S4 to S11 and S30 to S31 using the updated collection target T1. In steps S20 to S24, the host collection system collects new recorded data from a group of vehicles 100 based on the updated collection target T1 and rewards the vehicles 100.

[0195] The process of updating the data collection target can be performed once or multiple times. When necessary, the collection target can be updated frequently.

[0196] In another embodiment, the vehicle data recording device 140 further includes: Figure 2 Path or route planning functionality 151 is shown. The path (route) planning functionality can be integrated into the reinforcement learning agent component 144.

[0197] The path planning function 151 has two components: the first is a route encoder, and the second is a route selector. The first component (the route encoder) is responsible for determining multiple possible paths or routes and calculating the path or route encoding. The route (path) encoding can be represented by a numerical vector that encodes information about the value of the route with respect to multiple metrics. The second component (the route or path selector) is responsible for deciding whether to change the current route (when appropriate) and, if so, selecting a route (path) from a list of routes associated with the route encoding determined by the route encoder that optimizes (in other words, maximizes) the reward (which also means minimizing the penalty) received from the host data collection system 200 during or after uploading the recorded data. Optionally, the route preselector can preselect a limited number of routes from which the reinforcement learning agent must decide. In embodiments, the route selector performs the task of selecting a route through reinforcement learning. The route selector can consider additional environmental information that is independent of the route. The term "independent" here means independent of static route characteristics, as opposed to dynamic characteristics such as weather. The route selector is rewarded for uploading information. Of great concern may be having 5G frequency band availability to optimize its functionality.

[0198] In an embodiment, the route encoder and route selector are software running on a processor to perform the following steps after downloading the data collection target:

[0199] At the current time t0, multiple routes (paths) that the vehicle 100 can take are determined;

[0200] For each route, generating a route code that encodes information about the predicted value of the route for a plurality of metrics as a numerical value;

[0201] providing the route encoding and additional environmental information independent of the route to the reinforcement learning agent, the reinforcement learning agent operatively deciding whether to change the current route and selecting one of the determined routes to optimize reward when deciding to change the route;

[0202] recording data from a data source in the vehicle over time as the vehicle travels along a selected route (i.e., a route selected from a plurality of determined routes, or a current route if a decision is made not to change the current route);

[0203] uploading at least a portion of the recorded data to the host data collection system 200 and receiving a reward in return from the host data collection system 200;

[0204] The reward is provided to the reinforcement learning agent.

[0205] Path planning allows increasing the amount of high-value recorded data.

Claims

1. A method of collecting data from a vehicle (100) and wirelessly uploading at least a portion of the collected data to a host data collection system (200), the method comprising the following steps performed by a vehicle data recording device (140) in the vehicle (100): Wirelessly downloading (S3) a data collection target (T0) from the host data collection system (200), the data collection target (T0) including a data value metric (v cj ), the data value metric is a function that assigns data value to a data segment corresponding to a given point in time, the data value representing the achievement of an atomic collection goal (c j ) collecting (S4) data from a data source (110) in the vehicle (100) over time while the vehicle (100) is traveling; Recording (S5) the collected data in a storage module (142) of the vehicle data recording device (140); According to the data value metric (v cj ), calculating (S6) the data value of the recorded data over time; The reinforcement learning agent component (144) of the vehicle data recording device (140) selects (S8) at least one high-value recorded data segment to be uploaded within a time window from the recorded data (R) based on the calculated data value; wirelessly uploading (S10) only the selected at least one high-value recorded data segment to the host data collection system (200); receiving (S30) a reward value from the host data collection system (200) after successfully uploading the at least one high-value recorded data segment; and The received reward value is provided (S31) to the reinforcement learning agent component (144) as a reward input.

2. The method according to claim 1, further comprising: selecting (S8) at least one low-value recorded data segment to be deleted within a time window from the recorded data based on the calculated data value; The at least one low-value recorded data segment is deleted (S11) from the storage module (146).

3. The method according to claim 1 or 2, wherein: The data collection target (T0) includes different atomic collection targets ({c0,...,c m }) related m+1 data value metrics ({v c0 ,...,v cm }), m≥0, and includes defining a value metric for the m+1 data ({v c0 ,...,v cm }) of m+1 corresponding weights ({p c0 ,...,p cm }), the method further comprising the following steps: According to the m+1 data value metrics ({v c0 ,...,v cm }) calculating (S6) the value of m+1 sets of data of the recorded data over time; and aggregating (S7) the m+1 sets of data values ​​by calculating a weighted sum of the m+1 sets of data values ​​according to the collection strategy (T0) to calculate the aggregated data value of the recorded data over time, And wherein selecting (S8) recorded data segments is performed based on said aggregated data value of said recorded data over time.

4. The method according to claim 1, further comprising the steps of: Update information of the data collection target (T1) is wirelessly downloaded (S43) from the host data collection system (200), and the data collection target is updated (S44) according to the update information.

5. The method according to claim 1, wherein In the case where a plurality of recorded data segments to be uploaded are selected from the recorded data, the method further comprises the following steps: Calculating (S10) the value density for each recorded data segment to be uploaded; The plurality of record data segments to be uploaded are placed (S10) into a sending priority queue, and the record data segments to be uploaded are sorted based on the value density.

6. The method according to claim 1, further comprising the steps of: At a current time t0, determining a plurality of routes that the vehicle (100) can take; For each route, generating a route code that encodes information about the predicted value of the route for a plurality of metrics as a numerical value; providing the route encoding and additional environmental information independent of the route to the reinforcement learning agent, the reinforcement learning agent deciding whether to change the current route and selecting one of the determined routes to optimize reward when appropriate; recording data from a data source in the vehicle over time as the vehicle travels along a selected route; uploading at least a portion of the recorded data to the host data collection system (200) and receiving a reward in return from the host data collection system (200); The reward is provided to the reinforcement learning agent.

7. A method of collecting data from a set of recording vehicles (100), the method being performed by a host data collection system (200) wirelessly connected to the set of recording vehicles (100), the method comprising the steps of: At the data collection management module (230), a data collection goal (T0) is provided (S1), wherein the data collection goal (T0) includes a data value metric (v cj ), the data value metric is a function that assigns a data value to a data segment corresponding to a given time point (i), the data value representing the achievement of an atomic collection goal (c j ) wirelessly downloading (S2) the data collection target (T0) to the set of recording vehicles (100); wirelessly receiving (S20) recording data segments from the set of recording carriers (100); The received recorded data segments are stored (S21) in a storage facility (400) calculating (S22) virtual values ​​for recorded data segments uploaded from vehicles in the set of recording vehicles (100) based on the data collection target (T0); generating (S23) a reward value in dependence on said virtual value, and The reward value is downloaded (S24) to the vehicle to reward the reinforcement learning agent component (144) of the vehicle that is responsible for selecting high-value recorded data segments.

8. The method according to claim 7, wherein: In the step (S1) of providing a data collection target (T0), the data collection management module (230) provides different atomic collection targets ({c0, ..., c m }) related m+1 data value metrics ({v c0 ,...,v cm }), m≥0, and define the value metric for the m+1 data ({v c0 ,...,v cm }) of m+1 corresponding weights ({p c0 ,...,p cm })'s collection strategy (P).

9. The method according to claim 7 or 8, further comprising the steps of: at the data collection management module (230), updating (S40 to S41) the data collection target (T0) in dependence on the recorded data segments that have been received from the set of recording vehicles (100), and The updated information of the data collection target (T0) is wirelessly downloaded (S42) to the set of recording vehicles (100).

10. The method according to claim 7, wherein: The data value metric is generated to assign a normalized data value.

11. A vehicle data recording device (140), comprising: at least one interface (141) that receives logged data from a data source in the vehicle (100) and transmits segments of the logged data for wireless upload to a host data collection system (200); a storage module (142), the storage module storing the recorded data; Unit for performing the steps of the method according to any one of claims 1 to 6.

12. A vehicle (100), comprising the vehicle data recording device (140) according to claim 11.

13. A host data acquisition system (200), comprising: a first interface (210) that wirelessly interfaces with a set of recording carriers (100); a second interface (220) that interfaces with a storage facility (400); as well as Unit for performing the steps of the method according to any one of claims 7 to 10.

Images