Method and arrangement for determining a current position speed limit in a road vehicle

Abstract

Method and arrangement for determining current location speed limit in a road vehicle. Described herein is a method of determining a current location speed limit in a road vehicle speed limit information system. One or more signals corresponding to respective candidate speed limits for a current location are received. A parameterized heuristics algorithm with an associated cost function is applied to decide which candidate speed limit is applicable, if any. If available for the current location, an estimated true speed limit supplied by a cloud service and an associated confidence in the estimate are received. An online learning or reinforcement learning method is used to continuously fit the parameterization of the heuristics to reproduce the estimated true speed limit supplied by the cloud service with high confidence based on the estimated true speed limit supplied by the cloud service and the associated confidence in the estimate. A speed limit information signal corresponding to the decision of the parameterized heuristics algorithm is output.

Description

Technical Field

[0001] This disclosure generally relates to the verification of speed limit information for road vehicles, and particularly to a method for determining the current position speed limit in a road vehicle speed limit information system, a road vehicle speed limit information system suitable for performing the method, and a computer program including instructions configured to cause the road vehicle speed limit information system to perform the method when executed by processing circuitry. Background Technology

[0002] Speed ​​limit information and speed control functions are existing driver assistance features in automobiles. Euro NCAP and other vehicle safety rating agencies include the presence and performance of such features in their rating procedures. Proper speed limit information in a given driving environment is fundamental to correctly informing the driver or automatically adjusting the set speed used for speed control functions.

[0003] Speed ​​limits at a single location can change over time for various reasons; for example, authorities may change speed limits due to altered risk assessments, or they may be variable speed limits adapted to traffic conditions. Speed ​​limits can be conditional on factors such as a given time of day, current weather, or vehicle type. Supplementary information can limit the applicability of speed limit signs to a certain distance.

[0004] Existing speed limit information systems typically use one or more information sources to infer speed limits applicable to a vehicle in a given driving situation. A typical example of these information sources is a vehicle-mounted onboard camera combined with appropriate image processing software, including visual processing algorithms. Camera-based observations can be combined with an onboard geospatial database, such as a map in a navigation system commonly known as an Electronic Horizon (EH), or an off-board system, such as an information processing service accessible via an internet connection. Different information sources can have very different operating characteristics, which will cause them to supply conflicting information to a decision-making unit that must select a single speed limit value at any given time, for example, to present to the driver.

[0005] For example, onboard cameras may misidentify road signs. In particular, supplementary signs with digital information can sometimes be difficult to interpret completely correctly. Road signs may be completely or partially obscured, covered, damaged, or poorly positioned. Based on brief observations of signs as a vehicle passes them, the location of ambiguous signs may be difficult to correctly associate with the road for algorithmic interpretation. Furthermore, the onboard geospatial database may be outdated after speed limits change, or its information about speed limits may be incorrect for other reasons.

[0006] When using road sign observations from onboard cameras to infer the current effective speed limit in a driving situation, information from other camera observations or additional sensors, such as lane markings, can help infer the occurrence of events other than road sign observations that might affect the validity of previous speed limit information. Examples of such events are lane changes, moving from one road to another, or taking an on- or off-ramp. Based on onboard camera observations and input from the additional sensors, if the system cannot determine the currently effective speed limit with sufficiently high certainty, some speed limit information systems will have to decide which speed limit to display to the driver and when not to display it.

[0007] When multiple onboard information sources, such as the aforementioned cameras and onboard geospatial databases, are combined, they can provide conflicting information about the current speed limit, and the speed limit information system will typically have to choose which source to trust over another. Existing technologies use various conditional heuristics, fine-tuned by engineers, to improve the likelihood of selecting the correct information source in such situations.

[0008] Therefore, in order to implement the aforementioned decisions, speed limit information systems typically implement different parametric heuristic algorithms, which need to be fine-tuned by engineers in labor-intensive testing and verification procedures before the system is put on the market. The more time and resources spent on this tuning, the better the system performance experienced by the customer driving the vehicle.

[0009] As mentioned above, a major challenge in such tuning procedures is preventing the speed limit information system from being tuned to a limited set of test data that may inappropriately reflect the driving routines of all or some customers. Furthermore, it is often prohibitively expensive to provide different tuning for different driver clusters—groups of drivers sharing defined characteristics—where the differences between each market could be the best achievable outcome.

[0010] Supervised learning is a type of machine learning method that uses a set of input values ​​for which the correct output values ​​are known, to train the behavior of a speed-constrained information system, such as that represented by the decision algorithm described above. Therefore, supervised learning is an optimization technique that can be used to support or replace the tuning of decision algorithms for speed-constrained information systems, which are otherwise performed by development engineers, as described above. While supervised learning methods at this stage help automate the optimization of decision algorithms, the process remains labor-intensive because it requires a sufficiently large dataset of labeled training data.

[0011] In recent years, deep artificial neural networks have significantly improved automated image and video classification capabilities, which can help mitigate some of the weaknesses of cameras. However, the execution of such networks requires powerful, or at least specialized, computing hardware. From a cost perspective, such hardware may be feasible for highly autonomous driving platforms. On the other hand, for mass-market advanced driver assistance systems, it is assumed that alternative solutions remain more cost-effective.

[0012] Document US 2017 0 308 759 A1 proposes an attempt to improve the determination of speed limits for vehicles at current locations on roads, disclosing systems, methods, and apparatus for estimating speed limits for vehicles. The disclosed system for estimating speed limits includes one or more perception sensors, an arbitration speed component, an attribute component, an estimator component, and a notification component. One or more perception sensors are configured to generate perception data about the area near the vehicle. The arbitration speed component is configured to determine that a high-confidence or arbitration speed limit is not available. The attribute component is configured to detect one or more environmental attributes based on the perception data. The estimator component is configured to determine the estimated speed limit based on the environmental attributes. The notification component is configured to provide the estimated speed limit to the vehicle's autonomous driving system or driver assistance system.

[0013] Therefore, US20170308759A1 aims to correlate real speed data with environmental properties without taking into account any acquired speed limit data, and in some embodiments even uses acquired speed limit data as real speed data for training.

[0014] However, there is definite room for further improvement of the current position speed limit for road vehicles. Summary of the Invention

[0015] The purpose of this invention is to provide an improved method for determining the current location speed limit in a road vehicle speed limit information system.

[0016] According to a first aspect, a method is provided for determining a speed limit at a current location in a road vehicle speed limit information system, the method comprising: receiving one or more signals corresponding to a candidate speed limit for the current location; applying a parameterized heuristic algorithm with an associated cost function to determine which candidate speed limit is applicable, if any; if applicable for the current location, receiving an estimated true speed limit provided by a cloud service and an associated confidence level in the estimate; using online learning or reinforcement learning methods, such as stochastic gradient descent, incremental stochastic gradient descent, or recursive least squares, to continuously fit the parameterization of the heuristic to reproduce the estimated true speed limit provided by the cloud service with high confidence based on the estimated true speed limit provided by the cloud service and the associated confidence level in the estimate; and outputting a speed limit information signal corresponding to the decision of the parameterized heuristic algorithm.

[0017] Therefore, the above method takes into account reducing the effort required for the initial tuning of the decision heuristic of the road vehicle speed limit information system and thus reducing the associated costs. This, as in the online learning or reinforcement learning used in the above method, will provide continuous improvement in the tuning of the decision heuristic of the road vehicle during operation and provide increased confidence in the registered speed limit values.

[0018] In another embodiment, the method further includes performing online learning or reinforcement learning methods on the associated road vehicle.

[0019] In another embodiment, the method further includes uploading parameterized parameters derived from the fitting heuristic and the cost function values ​​implemented thereon to a cloud service; possibly using online learning or reinforcement learning techniques, the cloud service aggregates the parameter values ​​along with parameters and cost function values ​​received from multiple other vehicles to obtain a parameter set through aggregation; downloads the obtained parameter set back to the road vehicle from the cloud service; and uses the downloaded parameter set to determine which candidate speed limit is applicable, if any, when applying the parameterized heuristic algorithm.

[0020] In yet another embodiment, the method further includes using an absolute or relative threshold in a continuous cycle to improve the cost function value that needs to be surpassed so as to upload or download a new set of parameter values ​​in either direction.

[0021] In an additional embodiment, the method further includes optimizing the clustering of road vehicles and their parameters into a driving environment profile consisting of one or more different clustering criteria.

[0022] In yet another embodiment, the method further includes using an unsupervised machine learning algorithm to perform clustering of road vehicles and their parameters into a driving environment profile.

[0023] In yet another embodiment, the method further includes applying reinforcement learning techniques to clustering.

[0024] In another embodiment, the method further includes using a two-stage learning process that improves the parameterization separation of the camera-only solution to provide a signal corresponding to the appropriate candidate velocity limit for the current location, and parameters for determining the heuristic.

[0025] In yet another embodiment, the method further includes performing online learning or reinforcement learning of a new set of heuristic parameters in a cloud service by having each of the plurality of vehicles compute a cost function for online learning and upload the cost function value, sufficient information for identifying the parameter values ​​of the heuristic on which the cost function value is based, and additional information for clustering where applicable, to a cloud service; and using the information, performing online learning in the cloud service to optimize the parameters of the heuristic and sending the resulting new parameters to the connected road vehicles; and using the received updated parameters in the appropriate road vehicle when it is determined which candidate speed limit, if present, is applicable.

[0026] In another embodiment, the method further includes sending updated heuristic parameters to groups of road vehicles from a cluster during a phase, such that there are one or more control groups using the heuristic parameters from a previous set; in the cloud service, comparing the cost function values ​​of the groups using the updated new heuristic parameters with those of the groups using the old heuristic parameters, wherein the comparison of cost function values ​​from different groups is based on cost function values ​​collected in the same time window; determining whether the updated new heuristic parameters improve the cost function values ​​with statistical significance for the clustering of vehicles, and if so, sending the new parameters to an additional group of that cluster by the cloud service.

[0027] In another embodiment, the method further includes using three or more generations of heuristic parameter sets distributed to different groups for comparison in the cloud service; determining one or more currently optimal heuristic parameter sets; and having the cloud service send these one or more currently optimal heuristic parameter sets to an additional group.

[0028] According to the second aspect, a road vehicle speed limit information system is provided, which has means suitable for performing the steps of the methods described herein.

[0029] In one embodiment, a road vehicle speed limit information system as described above is provided, which also has means for wireless communication with an information processing service.

[0030] According to a third aspect, a computer program including instructions is provided, which, when executed by processing circuitry, are configured to cause the road vehicle speed limit information system described herein to perform the methods described herein.

[0031] According to the fourth aspect, a carrier comprising the computer program described herein is provided, wherein the carrier is one of an electronic signal, an optical signal, a radio signal, or a computer-readable storage medium.

[0032] The above embodiments have the following beneficial effects: they provide continuous improvement in the tuning of the heuristic for determining the speed limit of road vehicles during road vehicle operation. Attached Figure Description

[0033] Below, embodiments described herein will be illustrated in more detail by way of example only, with reference to the accompanying drawings, wherein:

[0034] Figure 1 The proposed method for determining the current location speed limit in a road vehicle speed limit information system is illustrated schematically.

[0035] Figure 2 A road vehicle speed limit information system according to an embodiment herein is illustrated schematically;

[0036] Figure 3 The data processing arrangement of a road vehicle speed limit information system according to embodiments herein is illustrated schematically;

[0037] Figure 4 A speed limit information system having a data processing arrangement arranged in a road vehicle is schematically illustrated according to some embodiments herein;

[0038] Figure 5 A road vehicle with a data processing arrangement located at a remote server is illustrated schematically according to other embodiments herein. Detailed Implementation

[0039] Below are some example embodiments of a method for determining the current location speed limit in a road vehicle speed limit information system 2.

[0040] like Figure 1 As illustrated schematically, the proposed method is based on the understanding that online learning or reinforcement learning methods can be used to support the tuning of decision algorithms for speed-constrained information systems, which are otherwise performed by development engineers, as described above.

[0041] Online learning is a machine learning approach where data becomes available in a sequential order and is used at each step to update our best predictor for future data.

[0042] Reinforcement learning differs from supervised learning in that it does not require presented labeled input / output pairs.

[0043] Therefore, using online learning or reinforcement learning methods to tune the parameterized heuristic algorithm used to determine the speed limit considers a more cost-effective and less time-consuming, labor-intensive solution compared to previously known solutions.

[0044] Furthermore, based on the understanding that the true speed limits for different locations, which have associated confidence levels among those estimates obtained from multiple road vehicles, can be previously uploaded to and stored in one or more databases on one or more remote servers, also referred to herein as cloud services, and that such estimated true speed limits can be combined with online learning or reinforcement learning methods to tune a speed limit determination algorithm to reproduce the estimated true speed limits provided by the cloud services with high confidence based on available candidate speed limits for the current location.

[0045] As used herein, cloud services are any information processing services available to a user on demand from one or more servers of a cloud computing provider via a network connection such as the Internet. Such services include the on-demand availability of computer system resources, particularly tools and applications such as data storage, servers, databases, networking, software, and computing power. These resources can be centralized or distributed across multiple locations, such as one or more data centers, and may also include distributed onboard resources. Cloud services can be delivered via the Internet and are globally accessible from the Internet. Therefore, as used herein, the term "cloud service" is a broad term encompassing all delivery and service models of cloud computing and related solutions.

[0046] Although, as stated above, a database of previously uploaded and stored estimates of true speed limits for different locations with associated confidence levels in those estimates obtained from multiple road vehicles is a prerequisite for the methods described herein, such a database and / or apparatus or method for obtaining the true speed limits of estimates for different locations with associated confidence levels in those estimates and for uploading such information to a cloud service are merely prerequisites for the methods described herein and do not constitute any part thereof.

[0047] Therefore, as Figure 1 The diagram schematically illustrates a method for determining the speed limit at a current location within a road vehicle speed limit information system 2. The method includes receiving one or more signals 3, 4 corresponding to a suitable candidate speed limit for the current location.

[0048] One or more signals 3, 4 corresponding to the respective candidate speed limits can be received from road sign observations, which are acquired from an onboard camera 5 equipped with appropriate image processing software, such as a forward-looking camera. This observation can be combined with speed limit information from an onboard reference location database 6, such as a map, to provide the corresponding candidate speed limits. The reference location database is typically referenced using a Global Navigation Satellite System (GNSS), such as, for example, the satellite-based Global Positioning System (GPS).

[0049] The method also includes applying a parameterized heuristic algorithm 7 with an associated cost function to determine which candidate speed limit is applicable, if any.

[0050] In order to be able to tune a parameterized heuristic algorithm with its associated cost function, the method also includes receiving an estimated true speed limit 9 provided by cloud service 8 and the associated confidence level in that estimate, if available for the current location.

[0051] Tuning is performed using online learning or reinforcement learning methods such as stochastic gradient descent, incremental stochastic gradient descent, or recursive least squares to continuously fit the parameterization of the heuristic based on the estimated true velocity limit 9 of the cloud service 8 supply and the associated confidence in that estimate to reproduce the estimated true velocity limit 9 of the cloud service 8 supply with high confidence.

[0052] The method ends by outputting a speed limit information signal 11 corresponding to the decision of the parameterized heuristic algorithm 7 that is thus tuned.

[0053] Therefore, the above method takes into account reducing the effort required for the initial tuning of the decision heuristic of the road vehicle 1 speed limit information system 2 and thus reducing the associated costs. This, as online learning or reinforcement learning as used in the above method, will provide a tuning 10 for the decision heuristic 7 of the road vehicle 1 that continuously improves during operation, and provides increased confidence in the registered speed limit values ​​3, 4.

[0054] In one embodiment, the online learning or reinforcement learning method can be performed on the associated road vehicle 1.

[0055] In a further embodiment, the method further includes uploading parameterized parameters derived from the fitting heuristic and the cost function values ​​implemented thereon to cloud service 8. Using online learning or reinforcement learning techniques, cloud service 8 aggregates the parameter values ​​along with parameters and cost function values ​​received from multiple other vehicles to obtain a resulting parameter set. The resulting parameter set can then be downloaded back to road vehicle 1 from cloud service 8. When applying the parameterized heuristic algorithm 7, the downloaded parameter set can then be used to determine which candidate speed limit is applicable, if any.

[0056] In some of its embodiments, the method further includes using an absolute or relative threshold in a continuous loop to improve the cost function value that needs to be surpassed so as to upload or download a new set of parameter values ​​in either direction.

[0057] In an additional embodiment, the method further includes optimizing the clustering of road vehicle 1 and its parameters into a driving environment profile consisting of one or more different clustering criteria.

[0058] Clustering criteria can, for example, be applied to geographical areas with similar configurations of road signs and speed limits, based on the frequency of occurrence of different sign categories or the frequency of occurrence of patterns in conflicts between different information sources.

[0059] In another embodiment of this document, the method further includes performing clustering of the road vehicle 1 and its parameters into a driving environment profile using an unsupervised machine learning algorithm.

[0060] Unsupervised machine learning algorithms infer patterns from the dataset without referring to known or labeled results.

[0061] In other embodiments of this document, the method further includes applying reinforcement learning techniques to clustering.

[0062] Reinforcement learning techniques begin to take appropriate actions to maximize rewards in a specific situation. It is used to find the best possible behavior in a given situation. Reinforcement learning differs from supervised learning in that, in supervised learning, the training data has an answer key, and the model is trained using the correct answer itself; in reinforcement learning, there is no answer key, but the reinforcement agent learns by deciding what to do to perform a given task based on its experience.

[0063] In another embodiment, the method further includes using a two-stage learning process that improves the parameterization of the camera-only 5 solution, providing a signal 3 corresponding to the appropriate candidate velocity limit for the current position, and parameters for determining the heuristic 7.

[0064] In another embodiment, the method further includes performing online learning or reinforcement learning of a new set of heuristic parameters in the cloud service 8 by having each of the plurality of vehicles 1 calculate a cost function for online learning and upload the cost function value, sufficient information for identifying the parameter values ​​of the heuristic on which the cost function value is based, and additional information for clustering, where applicable, to the cloud service 8. Using this information, online learning is performed in the cloud service 8 to optimize the parameters of the heuristic, and the resulting new parameters are sent to the connected road vehicles 1. Then, when it is determined which candidate speed limits 3, 4, if present, are applicable, these received updated parameters are used in the corresponding road vehicle 1.

[0065] In another embodiment, the method further includes sending updated heuristic parameters to groups of road vehicles 1 from a cluster during a phase, such that there will be one or more control groups using the previous set of heuristic parameters. In cloud service 8, the cost function values ​​of the groups using the updated new heuristic parameters are compared with those of the groups using the old heuristic parameters. The comparison of cost function values ​​from different groups is based on cost function values ​​collected within the same time window. This concludes by determining whether the updated new heuristic parameters improve the cost function values ​​with statistical significance for the clustering of vehicle 1, and if so, the new parameters are sent by cloud service 8 to an additional group of that cluster.

[0066] In an additional embodiment, the method further includes generating three or more sets of heuristic parameters distributed across different groups for comparison within cloud service 8. One or more sets of heuristic parameters that are currently optimal are determined. These sets of one or more optimal heuristic parameters are then sent by cloud service 8 to the additional groups.

[0067] A further idea is as follows: Figure 2 The road vehicle 1 shown schematically has a speed limit information system 2, which has means suitable for performing the methods described herein.

[0068] Such a device may include one or more interfaces 12 for receiving one or more signals 3, 4 corresponding to a candidate velocity limit for the current location. It may also include one or more interfaces 13 for receiving an estimated true velocity limit 9 provided by the cloud service 8 and the associated confidence level in that estimate, if available for the current location, and one or more interfaces 14 for outputting a signal 11 corresponding to a decision of the parameterized heuristic algorithm 7.

[0069] like Figure 3As shown, the device may also include one or more data processing arrangements 15, each including a processing module 16, which typically includes at least one processing circuitry 17 containing one or more processors and at least one memory 18 storing instructions executable by the processing circuitry 17, such as a computer program 19 (software), including instructions for performing the various steps and processes described herein. It also typically includes an I / O module 20, providing input to the data to be processed and output of the results of such processing.

[0070] In addition, such as Figure 4 As shown, using a suitable communication channel as indicated by arrow 21, the speed limit information system 2 for road vehicles 1 can be configured to communicate with an information processing service 8, such as a cloud service. The suitable communication channel can be based on technologies such as 4G and 5G networks, in-vehicle Wi-Fi, etc.

[0071] Therefore, in some embodiments, a road vehicle 1 speed limit information system 2 as described above is provided, which also has a device 22 for wirelessly communicating with an information processing service 8 such as a cloud service.

[0072] A computer program 19 comprising instructions, which, when executed by processing circuitry 17, is also envisioned to cause the road vehicle 1 speed limit information system 2 described herein to perform the methods described herein. Processing circuitry 17 may include a collection of one or more processors (not shown).

[0073] The data processing arrangement 15, including the processing circuit 17, can be arranged in the road vehicle 1 speed limit information system 2, such as... Figure 4 As shown, or positioned away from road vehicles, such as at a remote server, as Figure 5 As shown in the image.

[0074] In an embodiment where the data processing arrangement 15, including the processing circuitry 17, is located at a remote server 8, the method described herein can be implemented, for example, by continuously streaming data between the data processing arrangement 15, including the processing circuitry 17, and the road vehicle 1. Streaming is the delivery of content in real time as an event occurs, i.e., a continuous stream of data.

[0075] Therefore, system 2 or remote server 8 may include one or more data processing arrangements 15, each including a processing module 16, which typically includes at least one processing circuit 17 containing one or more processors and includes at least one memory 18 storing instructions 19 executable by the processor 17, such as computer programs (software), including instructions for performing the various steps and processes described herein, and also includes an I / O module 20 that provides input to the data to be processed and output of the results of such processing.

[0076] Communication, such as streaming, between road vehicle 1 and data processing arrangement 15 including processing circuitry 17 located at a remote server 8 (e.g., a cloud service), and communication, such as streaming, back to road vehicle 1's speed limit information system 2, may also include a communication network connected to the remote server 8, as shown by arrow 21. This communication network represents one or more mechanisms that road vehicles can use to communicate with the remote server 8. Therefore, the communication network can be one or more of various wireless communication mechanisms, including any desired combination of wireless communication mechanisms such as radio frequency, cellular, satellite, and microwave, and any desired network topology. Exemplary communication networks include wireless communication networks using, for example, Bluetooth, IEEE 802.11, LTE, 5G, etc.

[0077] This document also envisions a carrier (not shown) including the computer program 19 described herein, wherein the carrier is one of an electronic signal, an optical signal, a radio signal, or a computer-readable storage medium.

[0078] The above embodiments have the following beneficial effects: they provide continuous improvement in the tuning of the speed limit determination heuristic 7 of the road vehicle 1 during operation.

[0079] This reduces the effort required for the initial tuning of the decision heuristic 7 in the speed limit information system 2, as online or reinforcement learning systems will continuously improve their tuning as the vehicle 1 operates. Furthermore, by means of the described clustering to different driving environments, it is possible to have individual tuning results for different environments at a fine-grained level, which fit well to the individual environment and provide improved system performance in different environments compared to what could be achieved when treating different environments as the same.

[0080] Continuous tuning is performed using real-world driving and adapts to changing environments. This is nearly impossible when only the previous tuning decision system is used. Furthermore, when reinforcement learning of the tuning parameters is connected to or performed on a cloud service 8, as suggested in this paper, it is also possible to update the decision system 2 of a newly manufactured road vehicle 1 with continuously improving tuning parameters. This also applies to road vehicles without connectivity or an onboard mapping system 6.

[0081] A road vehicle 1 equipped with connectivity to cloud service 8 has limited use for training on road segments where the cloud service's map has recently been updated with high confidence, possibly based on speed limits determined by other vehicles. The real benefit of training the heuristic parameters emerges when vehicle 1 is driving in a similar environment where no recently updated high-confidence map information is available, such as due to fewer vehicles driving there, lack of connectivity provider coverage, etc.

[0082] Those skilled in the art to which these inventions pertain will appreciate the benefits of the teachings presented in the foregoing description and associated drawings, and will be aware of numerous modifications and other embodiments of the invention recounted herein. Therefore, it should be understood that the invention is not limited to the specific embodiments disclosed, and that modifications and other embodiments are intended to be included within the scope of the appended claims. Furthermore, while the foregoing description and associated drawings describe exemplary embodiments in the context of certain exemplary combinations of elements and / or functions, it should be understood that different combinations of elements and / or functions may be provided by alternative embodiments without departing from the scope of the appended claims. In this regard, for example, combinations of elements and / or functions different from those explicitly described above are also conceived to be recounted in some of the appended claims. Where advantages, benefits, or solutions to problems are described herein, it should be understood that such advantages, benefits, and / or solutions may apply to some example embodiments but not necessarily to all example embodiments. Therefore, any advantages, benefits, or solutions described herein should not be considered critical, necessary, or essential for all embodiments or the embodiments claimed herein. Although specific terms are used herein, they are used only in a general and descriptive sense and are not for limiting purposes.

Claims

1. A method for determining the current location speed limit in a road vehicle (1) speed limit information system (2), the method comprising: Receive one or more signals (3, 4) corresponding to the corresponding candidate velocity limit at the current position; A parameterized heuristic algorithm (7) with an associated cost function is applied to determine which candidate velocity constraint (3, 4) is applicable, if any; Its characteristic is that it further includes: If available for the current location, receive the estimated true velocity limit (9) of the cloud service (8) supply and the associated confidence in that estimate; use online learning or reinforcement learning methods to continuously fit the parameterization of the (10) heuristic based on the estimated true velocity limit of the cloud service (8) supply and the associated confidence in that estimate to reproduce the estimated true velocity limit (9) of the cloud service (8) supply with high confidence. Output the speed limit information signal corresponding to the decision of the parameterized heuristic algorithm (11).

2. The method according to claim 1, wherein, The method also includes performing online learning or reinforcement learning methods on the associated road vehicle (1).

3. The method according to claim 2, wherein, The method further includes: The parameterized parameters derived from the fitting (10) heuristic and the cost function values ​​they implement are uploaded to the cloud service (8); the cloud service (8) may use online learning or reinforcement learning techniques to aggregate the parameter values ​​together with the parameters and cost function values ​​received from multiple other vehicles to obtain the resulting parameter set through aggregation; The parameter set obtained from the cloud service (8) is downloaded back to the road vehicle (1); When applying the parameterized heuristic algorithm, the downloaded set of parameters is used to determine which candidate speed limit (3, 4) is applicable, if any.

4. The method according to claim 3, wherein, The method also includes using absolute or relative thresholds in continuous loops to improve the cost function value that needs to be surpassed so as to upload or download a new set of parameter values ​​in either direction.

5. The method according to any one of claims 3 or 4, wherein, The method also includes optimizing the clustering of road vehicles (1) and their parameters into a driving environment profile consisting of one or more different clustering criteria.

6. The method according to claim 5, wherein, The method also includes using unsupervised machine learning algorithms to perform clustering of road vehicles (1) and their parameters into a driving environment profile.

7. The method according to claim 6, wherein, The method also includes applying reinforcement learning techniques to clustering.

8. The method according to any one of claims 1 to 4, wherein, The method also includes using a two-stage learning process that separates the parameterization of the camera-only (5) solution to provide a signal corresponding to the candidate velocity limit (3) for the current position, and to determine the parameters for the heuristic.

9. The method according to any one of claims 3 to 4, wherein, The method further includes performing online learning or reinforcement learning of a new set of heuristic parameters in the cloud service (8) by having each of the multiple vehicles calculate a cost function for online learning and upload the cost function value, sufficient information for identifying the parameter values ​​of the heuristic on which the cost function value is based, and additional information for clustering where applicable, to the cloud service (8), and using the information, performing online learning in the cloud service (8) to optimize the parameters of the heuristic and sending the resulting new parameters to the connected road vehicles (1), and using the received updated parameters in the corresponding road vehicles (1) when it is determined which candidate speed limit (3, 4), if present, is applicable.

10. The method according to claim 9, wherein, The method further includes sending updated heuristic parameters to a group of road vehicles (1) from a cluster during a phase, such that there will be one or more control groups of a previous set using the heuristic parameters; in the cloud service (8), comparing the cost function values ​​of the group using the updated new heuristic parameters with those of the group using the old heuristic parameters, wherein the comparison of cost function values ​​from different groups is based on cost function values ​​collected in the same time window; determining whether the updated new heuristic parameters improve the cost function values ​​with statistical significance for the clustering of vehicles (1), and if so, sending the new parameters to an additional group of that cluster by the cloud service (8).

11. The method according to claim 10, wherein, The method further includes generating three or more sets of heuristic parameters distributed to different groups for comparison in the cloud service (8); determining one or more sets of heuristic parameters that are currently best; and sending these one or more sets of heuristic parameters that are currently best to the additional groups by the cloud service (8).

12. A road vehicle (1) speed limit information system (2) having means adapted to perform the steps of the method according to any one of claims 1 to 11.

13. The road vehicle (1) speed limit information system (2) according to claim 12, further comprising means (22) for wireless communication (21) with information processing service (8).

14. A computer program product comprising instructions which, when executed by a processing circuit (17), are configured to cause a road vehicle (1) speed limit information system (2) according to claim 12 or 13 to perform the method according to any one of claims 1 to 11.

15. A computer-readable storage medium comprising a computer program product according to claim 14.

Images