Method and apparatus for traffic flow control

Abstract

According to a first aspect of the present invention, a method for traffic flow control is presented. The method comprises generating a driving behavior indication for at least one traffic user, in particular a networked motor vehicle, and transmitting the driving behavior indication to the at least one traffic user, in particular to the networked motor vehicle. In this case, the driving behavior indication is generated in such a way that a digital model of a traffic channel is generated by means of a current traffic condition, a traffic flow is simulated on the basis of the digital model, and the driving behavior indication is determined on the basis of the simulation of the traffic flow.

Description

Technical Field

[0001] The invention relates to a method and a device for producing a traffic flow control. The device may be, in particular, an infrastructure system for at least partially automated driving support of guided networked motor vehicles. The invention also relates to a computer program. Background Art

[0002] In vehicle applications, as in other areas of technology, networking plays an increasingly important role. More and more vehicles have the possibility of connecting to other traffic participants, infrastructure components (such as so-called roadside units) or to backend services in the cloud. So-called V2X communication or V2V communication is used here.

[0003] Radio communication standards for this purpose are, for example, direct communication standards based on WLAN (DSRC, IST-G5), or methods based on mobile radio communication in both direct and indirect forms (C-V2X). In particular, the networking of vehicles with infrastructure systems is becoming increasingly important. Such a system can assist the vehicle in its driving tasks, for example by providing environmental information through sensors on the edge of the road that the vehicle's onboard sensor system cannot or can only generate on its own to a limited extent. If, for example, a partially automated or fully automated vehicle can no longer or can only continue to drive due to a limited view of the surroundings, the support can be extended to direct intervention in driving functions. Through this networking of vehicles and infrastructure systems via V2X communication, not only traffic safety but also traffic efficiency can be decisively improved.

[0004] Another possibility for influencing the maneuvers of partially or fully automated vehicles from outside the vehicle is that a plurality of such vehicles directly follow one another and exchange their environmental information and driving tasks, thereby making cooperative behavior of the vehicles possible. This offers advantages primarily in scenarios where connected vehicles interact directly with one another (e.g. changing lanes on a highway), and offers fewer advantages at traffic nodes with large visibility differences where the combined field of view of a plurality of vehicles is not always sufficient to provide an overview of the entire traffic situation.

[0005] From laid-open patent application DE 10 2020 102 329 A1 a method for vehicle control in a traffic situation in which a vehicle merges into a traffic flow, wherein the total energy requirement is minimized, is known. In the method, traffic data of vehicles participating in the traffic situation are detected and at least one control signal is generated for at least one of these vehicles in order to assist the merging. For example, traffic data and / or sensor data of environmental sensor devices of participating vehicles and / or of a vehicle-external traffic observation device are used for this purpose, which traffic observation device can be provided as part of a local infrastructure.

[0006] A traffic flow advisory system is known from the publication DE 10 2010 050 805 A1. Here, the traffic flow of a plurality of road lanes is constructed and forwarded to the vehicle driver. The traffic flow can be expanded with the aid of data transmitted wirelessly from infrastructure sensors. In one embodiment, the system, as an interface between the driver and the machine, provides the driver with advice on which lane is particularly more suitable for his driving behavior based on predetermined criteria. Such indications can include directional visual, acoustic or tactile indications.

[0007] If the determined traffic intensity exceeds the capacity of the road, the traffic becomes denser and starts to back up. Under such conditions, the phenomenon of traffic waves can occur, also known as brake shock waves or "stop and go". A related problem is the detection of "jam ends", in which traffic participants have to brake suddenly and drastically from normal speed.

[0008] The current measure for combating these phenomena is a variable speed limitation. This requires a complex construction and can only be implemented with limited temporal and spatial frequencies.

[0009] Existing solutions for warning of jams involve exchanging driving data between users of navigation software on smartphones. In this case, the data of multiple traffic flows are aggregated and jams are detected from this. However, this aggregation method limits the responsiveness and accuracy of the solution and is not sufficient for better handling of brake shocks and jam end situations. Summary of the invention

[0010] It can therefore be seen as an object of the present invention to provide a reliable method for traffic flow control, in which phenomena such as brake shocks or sudden jams are effectively prevented.

[0011] A further object of the present invention may be to provide a device for traffic flow control, in particular an infrastructure system for driving support of at least partially automated guided, networked motor vehicles, which has a high reliability and by means of which phenomena such as braking shocks or sudden congestion are effectively prevented.

[0012] According to a first aspect of the present invention, a method for traffic flow control is proposed. The method comprises: generating a driving behavior instruction for at least one traffic participant, in particular a networked motor vehicle, and sending the driving behavior instruction to the at least one traffic participant, in particular to the networked motor vehicle. The driving behavior instruction is generated in such a way that a digital model of a traffic corridor is generated with the aid of current traffic conditions, a traffic flow is simulated based on the digital model, and the driving behavior instruction is determined based on the simulation of the traffic flow.

[0013] According to a second aspect of the invention, a traffic control unit is proposed, which is configured to generate driving behavior instructions for at least one traffic participant, in particular a networked motor vehicle, wherein the driving behavior instructions are generated in such a way that a digital model of the traffic corridor is generated with the aid of current traffic conditions, the traffic flow is simulated based on the digital model, and the driving behavior instructions are determined based on the simulation of the traffic flow.

[0014] According to a third aspect of the invention, a device for traffic flow control is proposed, which is in particular designed as an infrastructure system for at least partially automated driving support of guided, networked motor vehicles and is in particular designed to implement the method according to the first aspect. The device comprises at least one communication unit, at least one traffic control unit according to the second aspect, and at least one road traffic sensor, wherein current traffic conditions are detected in particular by means of the road traffic sensor.

[0015] The invention is therefore based on the planning of cooperative maneuvers of the participating vehicles. This not only makes it easier for individual vehicles to handle such traffic situations, but also prevents the occurrence of such braking shocks and jams, in order to benefit all traffic participants.

[0016] For the same reasons, this method offers advantages over simple V2X congestion warnings, since these can neither be received by all road users nor significantly reduce the mentioned phenomena.

[0017] The brake shock wave, in particular in a traffic jam, is a complex nonlinear phenomenon in which the traffic flow as a whole can be regarded as a single oscillating system. Central control of the traffic is important in order to detect the occurrence of oscillations and to prevent them by a unified strategy.

[0018] According to the invention, cluster driving maneuver instructions are used. As a result, in contrast to individual vehicle trajectory recommendations, the load rate of the communication channel remains constant regardless of the number of participating traffic participants. This type of communication places low hardware and software requirements on the participating vehicles and infrastructure systems, both for communication (multicast) and for driver assistance systems.

[0019] The driving behavior instructions according to the invention can also be implemented by non-partially automated or fully automated motor vehicles, so that in addition to existing traffic control options, the method described here is also suitable for mixed traffic conditions. In addition, the superordinate planning enables the planning algorithm to be tailored to the specific situation of the traffic axis for which it is used.

[0020] The generation, transmission and use of driving behavior instructions can be defined in particular in such a way that the infrastructure operator can influence the movement of a plurality of networked motor vehicles, but does not have to be responsible for the movement of the vehicles themselves. This is a decisive advantage of the present invention compared to the prior art, since reliable control of vehicles from the infrastructure and thus responsibility for vehicle movement by the infrastructure is very complex in terms of technology and difficult to achieve in terms of organization and law. The reason for the latter is that often different companies develop and sell infrastructure systems and AD vehicles, which are unable and unwilling to be mutually responsible for the correct behavior of the respective other company's products.

[0021] The expression "networked motor vehicle" includes motor vehicles that have a suitable communication device, by means of which the networked motor vehicle can exchange data with other traffic participants, in particular with infrastructure systems. For this purpose, a wireless data connection is established, via which the networked motor vehicle can send and / or receive data. Preferably, this can be a radio connection, for example a mobile radio connection or a direct wireless connection. This type of communication between a motor vehicle and other traffic participants is also referred to as V2X communication or C2X communication.

[0022] The expression “at least partially automated” includes one or more of the following situations: assisted guidance, partially automated guidance, highly automated guidance, fully automated guidance of the motor vehicle.

[0023] Assisted guidance means that the driver of the motor vehicle continuously guides the motor vehicle either transversely or longitudinally. The corresponding other driving task (i.e. the control of the longitudinal guidance or transverse guidance of the motor vehicle) is automatically performed. This means that during assisted guidance of the motor vehicle, either the transverse guidance or the longitudinal guidance is automatically controlled.

[0024] Partially automated guidance means that the longitudinal and lateral guidance of the motor vehicle is automatically controlled in certain situations (e.g. driving on the motorway, driving in a parking lot, overtaking an object, driving in a lane defined by lane markings) and / or within certain time periods. The driver of the motor vehicle does not have to manually control the longitudinal and lateral guidance of the motor vehicle himself. However, the driver must constantly monitor the automatic control of the longitudinal and lateral guidance in order to be able to intervene manually if necessary. The driver must be ready to take over the full control of the motor vehicle at any time.

[0025] Highly automated guidance means that the longitudinal guidance and lateral guidance of the motor vehicle are automatically controlled within a certain period of time in certain situations (e.g., driving on a highway, driving in a parking lot, overtaking an object, driving in a lane determined by lane markings). The driver of the motor vehicle does not have to manually control the longitudinal guidance and lateral guidance of the motor vehicle himself. The driver does not have to continuously monitor the automatic control of the longitudinal guidance and lateral guidance so that he can intervene manually when necessary. When necessary, a takeover request is automatically output to the driver in order to take over the control of the longitudinal guidance and lateral guidance, in particular in a manner with sufficient time margin. That is, the driver must potentially be able to take over the control of the longitudinal guidance and lateral guidance. The limits of the automatic control of the lateral guidance and longitudinal guidance are automatically identified. In the case of highly automated guidance, it is not possible to automatically reach a state with minimal risk in any initial situation.

[0026] Fully automated guidance means that in certain situations (e.g. driving on a motorway, driving in a parking lot, overtaking an object, driving in a lane defined by lane markings), the longitudinal and transverse guidance of the motor vehicle is automatically controlled. The driver of the motor vehicle does not have to manually control the longitudinal and transverse guidance of the motor vehicle himself. The driver does not have to monitor the automatic control of the longitudinal and transverse guidance so that he can intervene manually when necessary. Before the automatic control of the transverse and longitudinal guidance is terminated, a request is automatically made to the driver to take over the driving task (control of the transverse and longitudinal guidance of the motor vehicle), in particular with a sufficient time margin. If the driver does not take over the driving task, the system automatically returns to the state with the lowest risk. The limits of the automatic control of the transverse and longitudinal guidance are automatically identified. In any situation, it is possible to automatically return to the system state with the lowest risk.

[0027] Driverless control or guidance means that the longitudinal and lateral guidance of the motor vehicle is automatically controlled, independent of the specific application (e.g. driving on the highway, driving in a parking lot, overtaking an object, driving in a lane determined by lane markings). The driver of the motor vehicle does not have to manually control the longitudinal and lateral guidance of the motor vehicle himself. The driver does not have to monitor the automatic control of the longitudinal and lateral guidance so that he can intervene manually when necessary. Thus, for example, the longitudinal and lateral guidance of the vehicle is automatically controlled under all road types, speed ranges and environmental conditions. As a result, all driving tasks of the driver are automatically taken over. Therefore, a driver is no longer required. The motor vehicle can therefore be driven from any starting position to any target position without a driver. Potential problems are automatically solved without the help of the driver.

[0028] Remote control of a motor vehicle means remotely controlling the transverse guidance and longitudinal guidance of the motor vehicle, that is, for example, sending a remote control signal for remotely controlling the transverse guidance and longitudinal guidance to the motor vehicle. The remote control is performed, for example, by means of a remote control device.

[0029] For the digital model of the traffic corridor, in particular data from sensor devices on the infrastructure side can be detected, wherein the data represent the current traffic conditions and are processed together with topological information of the traffic corridor. For example, the data can come from V2X "Cooperative Perception Messages (CPM)" of networked traffic participants. Data from sensor devices on the infrastructure side (e.g. traffic monitoring cameras, radar, lidar, induction coils) are another possibility for data detection. The data can map the observed traffic in time as so-called "object tracking".

[0030] The topological information of the traffic corridor is preferably included in a digital map of the road network. This digital map can be input manually or obtained from a map provider, for example. Alternatively or additionally, the topological information of the traffic corridor can be determined from sensory data, which are received by a plurality of networked motor vehicles, in particular by means of machine learning. For this purpose, for example, common traffic changes and the trajectories of passing vehicles can be aggregated and automatically converted into a topological road model.

[0031] The simulation of traffic flow is preferably based on a real-time capable prediction model with rule-based traffic control. Such a model may, for example, include machine learning (ML) methods, such as convolutional neural networks (CNN) or recurrent neural networks (RNN). Advantageously, the prediction model is able to accurately predict "stop-and-go braking shock waves" and congestion ends so that they can be countered.

[0032] Particularly preferably, the simulation implements a machine learning method, wherein the machine learning model is trained with the aid of historically detected traffic conditions. Alternatively, the entire model can be formed by a CNN, an RNN or something similar. An example of a function that can be trained in this way is an estimate of the probability of a braking shock wave occurring within the next minute based on the current traffic situation.

[0033] Alternatively, the model can be based on physical modeling, for example on numerical simulation. For example, for physical modeling, the road topology can be divided into cells, which can then be used as the basis for the numerical solution of differential equations for the traffic flow model. Here, variables are simulated over time, such as the number of vehicles in each cell or the average speed. Another physical modeling possibility is the simulation of traffic based on agents. Here, the behavior and interaction of all individual traffic participants are modeled (micro-simulation). In the case of this physical modeling, certain model parameters can be set, for example, by multivariable optimization methods. An example of a variable that can be optimized is the average reaction speed of a human driver in the case of braking, front-running traffic.

[0034] A combination of physical modeling and ML methods, so-called "hybrid models", can also be used. Here, certain relationships within the model are trained and then used in the physical simulation model. Relevant functions that can be trained are, for example, the following terms of the transport equation: the terms ensure the propagation of simulated variables (such as traffic density and average speed per cell) in space and time in macro traffic simulations. These terms are typically nonlinear and depend on many factors that may only be partially detected by traditional models, such as the reaction speed of human drivers in braking, front-running traffic under real conditions. Here, these terms can be designed as functions supported by neural networks (also known as RNN, CNN) so that the influence of a large number of (possibly hidden) parameters can be learned. As input, for example, the traffic situation along the traffic corridor and other relevant conditions, such as weather conditions or time of day, can be considered.

[0035] For all of these model types, using real recorded traffic data for setting up and tuning the model can bring advantages.

[0036] In a preferred embodiment of the invention, an optimal driving behavior is determined for a specific section of the traffic corridor using a real-time-capable prediction model and driving behavior instructions are derived therefrom. The driving behavior instructions include, in particular, speed recommendations and / or acceleration and deceleration targets and / or recommendations for a distance to be maintained from the vehicle ahead and / or recommendations for using a specific lane.

[0037] To prevent the occurrence of jam ends and braking shocks, for example, the following simple rules can be applied after the impending event is determined:

[0038] Exemplarily, the following rules may be applicable to the driving behavior indication:

[0039] - You have to slow down in advance (antizipieren) instead of trying to keep a constant distance to the person in front.

[0040] - Acceleration should be kept moderate.

[0041] - The speed target must not significantly exceed the average speed in order to avoid re-braking.

[0042] The extent to which these rules should be applied can be determined in advance from simulations or can be based on empirical values ​​obtained from experience. The challenge here is to find values ​​with a sufficiently wide range of applicability.

[0043] In a preferred embodiment of the present invention, the driving behavior instruction comprises at least a speed recommendation. It has been shown that even small adjustments in speed can effectively prevent jams or braking waves, so that recommendations for speed, in particular maximum speed, are particularly effective in order to achieve an optimized traffic flow. Alternatively or additionally, the speed recommendation may, for example, comprise a speed interval and / or a minimum speed.

[0044] In a preferred embodiment of the invention, the driving behavior instructions are calculated and transmitted specifically for a certain lane. This has the advantage that events such as braking waves or sudden congestion can also be prevented in a lane-specific manner. It is also advantageous for this if the driving behavior instructions include lane change recommendations.

[0045] In a preferred embodiment of the invention, the simulation comprises a self-learning algorithm, wherein the traffic flow control is continuously optimized during operation.

[0046] An alternative to predefined rules for traffic control is a self-learning control algorithm, which is based, for example, on reinforcement learning (RL), "deep Q-learning" (DQL) or the like. In this case, the automated traffic control of the control unit is continuously optimized. The method is divided, for example, into a plurality of phases. After the phase of recording the data, the simulation model can be initialized. Here, the simulation is used as a training environment for, for example, RL methods, in which the model can incorporate control instructions for the traffic control unit in the course of the simulation. Therefore, the simulation is used to generate synthetic data for the initial training of the traffic control. The simulation requires the following traffic flow data as boundary conditions: the traffic flow data can include not only real recorded traffic data, but also synthetically generated scenarios. Here, the effect of the control can be analyzed in the simulation in many scenarios.

[0047] The situations occurring in the simulation are evaluated according to a so-called “reward function.” The “reward function” is defined in such a way that, for example, a speed that is as constant as possible for the traffic is favored and jams and other dangerous situations are reduced.

[0048] By trying out possible actions ("policies") in different state situations, the algorithm learns which sequence of actions leads to better overall results over time. In order to gather new information, the algorithm can deviate to a certain extent from the best control known so far. Typically, the degree of preference for exploration ("exploration-exploitation trade-off") can be input into the RL algorithm. An initial trial phase ("bootstrapping") is expediently carried out using simulation. Thus, unfavorable control sequences that could provide valuable information for learning are not applied to real traffic.

[0049] After having gone through the "offline" training phase and thus being able to achieve a suitable (but not yet optimal) traffic control, it is possible to switch to "live" operation. In this live operation, the control can be further optimized by "online learning". With live operation, the best known control is followed more closely, so that although learning becomes slower, good control is always provided.

[0050] This type of self-learning method allows the control to be adapted to the real local conditions. Therefore, the quality of the model only matters at the beginning, later, the experience from online learning prevails. This is advantageous because in reality the phenomenon to be treated depends on many factors that are not easy to model. For example, traffic rules are not geographically uniform, driving styles are sometimes influenced by culture, special road topologies occur, temporary obstacles to traffic.

[0051] In a preferred embodiment of the invention, the driving behavior instructions are transmitted to a plurality of traffic participants by means of a multicast communication method and / or a broadcast communication method. Multicast communication methods or broadcast communication methods are particularly advantageous in the case of blocked traffic. Thus, the load rate of the radio communication channel can be kept low compared to unicast communication methods, since a single message can control a large number of vehicles without having to address these vehicles individually.

[0052] In order to ensure sufficient specificity of the message in order to efficiently regulate traffic along the corridor, the receivers can be addressed in different ways. One possibility is to adjust the transmission power of the transmitter so that only the vehicles involved can receive the message. Another possibility is to specify a geographical validity area in the message. The validity area can be shown, for example, with the help of a quadrilateral or polygon, the corner points of which are marked with absolute coordinates. The absolute coordinates can correspond, for example, to the UTM or WGS84 standard. In the case of a network of radio communication units, the validity area can be addressed by the nearest radio communication unit in a mesh method. The validity area needs to be checked by the receiver in order to be able to determine the relevance of the message to itself.

[0053] In the message, in particular, instructions about speed control and the distance to the vehicle ahead are transmitted. For this purpose, instructions about the frequency of lane changes, lane selection or giving up the right of way can also be given in the case of merging methods. The target motor vehicle can then incorporate these instructions into its driver assistance system, alternatively, into its trajectory planning for the automated vehicle. The generated driving behavior instructions can be sent to the corresponding motor vehicle together with the environmental information via V2X communication.

[0054] In a preferred embodiment of the present invention, the driving behavior instruction is received by a networked motor vehicle and output to the driver of the motor vehicle, in particular in an audiovisual and / or tactile manner. Thus, the present invention can advantageously also be implemented in mixed traffic of manual motor vehicles and partially automated or fully automated motor vehicles.

[0055] Particularly relevant driver assistance systems that can be used to implement the driving behavior instructions are in particular "Automated Cruise Control (ACC)" or UX measures for supporting the human driver in adhering to a certain speed. These include:

[0056] - A visual warning indication, for example in the cluster display.

[0057] -Acoustic warning indication.

[0058] -Haptic control elements, such as accelerator pedal or steering wheel, ActivePedal or similar.

[0059] Suitable UX devices will appear more and more frequently in motor vehicles manufactured in the future, since they are provided for implementing the EU directive on “Intelligent Speed ​​Assist.” Therefore, an additional use of these UX devices is proposed in the present invention.

[0060] An important feature of the invention is that motor vehicles voluntarily follow the instructions of the infrastructure, as long as these instructions can be implemented functionally and in terms of safety technology. Since the responsibility for safety lies with the motor vehicles, they need to always check the external instructions for plausibility and evaluate the external knowledge relative to their own planning before implementing them. In the case of driver assistance systems in manually controlled vehicles, these driver assistance systems need to be designed in such a way that they do not hinder the driver's judgment ability. In order to ensure good acceptability of the infrastructure instructions on the vehicle side, these infrastructure instructions should only deviate from other traffic situations within an appropriate framework. In particular, the maximum permissible speed must not be exceeded. As a result, the safety concept on the vehicle side remains unchanged and the infrastructure does not need to bear additional safety requirements (in the sense of functional safety). BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Embodiments of the present invention are described in detail with reference to the accompanying drawings.

[0062] Figure 1 An exemplary embodiment of the present invention shows by way of example a system architecture for generating driving behavior instructions and transmitting the driving behavior instructions from an infrastructure system to a networked motor vehicle.

[0063] Figure 2 A flow chart of traffic control is shown according to an embodiment of the present invention.

[0064] Figure 3 A flow chart of a self-learning traffic control is shown according to an alternative embodiment of the present invention.

[0065] Figure 4 In the context of an exemplary embodiment of the present invention, passive recording of traffic data is described.

[0066] Figure 5 In the context of an exemplary embodiment of the present invention, a traffic model is automatically learned from measured traffic data.

[0067] Figure 6 According to an alternative exemplary embodiment of the present invention, a bootstrapping method of a self-learning traffic control unit is described by means of an interactive traffic simulation (offline learning).

[0068] Figure 7 The modeling of a jam end and a braking shock wave is shown by way of example.

[0069] Figure 8 In the context of an exemplary embodiment of the present invention, the operation of a self-learning traffic control is described.

[0070] Fig. 9 An exemplary scenario for backed-up traffic on a highway-like road is shown. DETAILED DESCRIPTION

[0071] In the following description of exemplary embodiments of the present invention, identical elements are denoted by identical reference numerals, wherein a repeated description of these elements is omitted where necessary. The figures merely schematically illustrate the subject matter of the present invention.

[0072] exist Figure 1 1 shows a possible system architecture 100 for generating driving behavior instructions and transmitting the driving behavior instructions from an infrastructure system to networked motor vehicles according to an exemplary embodiment of the present invention. A road traffic sensor 102, such as a camera, detects current traffic and sends data to a first roadside unit 104. The roadside unit 104 also receives CAM / CPM data of networked motor vehicles 112, 114, 116 via V2X communication and is coupled to a traffic control unit 106, which is designed to generate driving behavior instructions for at least one traffic participant, in particular a networked motor vehicle. The driving behavior instructions are generated in such a way that a digital model of a traffic corridor is generated with the help of current traffic conditions, which are detected with the help of the road traffic sensor 102, and a traffic flow is simulated based on the digital model, and a driving behavior instruction is determined based on the simulation of the traffic flow. The driving behavior instructions are transmitted to the roadside unit 104 and, if necessary, to a further roadside unit 108. The roadside units 104, 18 communicate the driving behavior instructions to the networked motor vehicles 112, 114, 116 with the help of a multicast communication method.

[0073] exist Figure 2200 of a traffic control according to an embodiment of the present invention is shown. In a first step 202, passive data detection of current traffic data is performed. In step 204, a prediction model for simulating traffic flow is run with the aid of the detected data. As a result, in step 206, driving behavior instructions for traffic flow control are generated.

[0074] exist Figure 3 300 of a traffic control according to an embodiment of the present invention is shown in FIG. In a first step 302, passive data detection of current traffic data is performed. In step 304, a self-learning model for simulating traffic flow is started with the aid of the detected data. In step 306, a first model for simulating traffic flow is generated with the aid of offline detected training data and a "bootstrapping" method. As a result, driving behavior instructions for traffic flow control are generated in step 308. In parallel, the model is further improved by real-time learning in step 310.

[0075] Figure 4 Passive data acquisition is depicted. Data 402 from infrastructure-side sensor systems and data 404 acquired by various networked road users, for example networked motor vehicles, are combined in object tracking and fusion unit 406 , and the result is made available in database 408 .

[0076] Figure 5 The process of automatically learning the traffic model from the measured traffic data is shown in FIG. Figure 4 ) provides the data. In the first path 502, data aggregation is first performed in step 504, thereby generating a topological map 506 of the traffic corridor involved. In parallel, in step 508, for example, a simulation of the traffic flow is generated on the one hand, and a synthetic traffic scene 510 and simulation parameters 512 are generated on the other hand, for example, by means of regression or ML methods. In step 514, the synthetic traffic scene 510, the topological map 506 and the simulation parameters 512 are used by the prediction model and the simulation engine to obtain an improved traffic model.

[0077] Figure 6The self-development process of a self-learning traffic control unit by means of an interactive traffic simulation (offline learning) is shown. Here, the simulation unit 902 provides a simple first result offline (i.e. not during ongoing operation) and provides the first result via an interface of the self-learning traffic control unit 906. The traffic control unit includes a learning module 908. The provided results are evaluated by means of the learning module 908 via a reward function 910 and the provided results are transferred (übernommen) into a utility function 912. As a result, driving behavior instructions or traffic control strategies ("policies") are generated in the ML module 914, for example by means of deep Q learning, and the driving behavior instructions or traffic control strategies are provided back to the simulation unit 902 via a further interface 916. As a result, the simulation unit 902 can calculate an improved simulation of the traffic flow and better results, which can in turn be transferred to the self-learning traffic control unit 906 for further improvement of the simulation.

[0078] Figure 7 A macroscopic traffic simulation model for jam ends and brake shock waves is depicted. In this model, the road topology 610 of the observed traffic corridor is divided into individual, discrete cells 601, which have, for example, a certain, fixed size. For each cell 601, it is possible to calculate how certain characteristic variables of the traffic flow (e.g., traffic density, number of vehicles per cell, average speed, etc.) change over time. Figure 7 610 at a first time 612 and a second time 614 at a later time. In this example, the dark color of the cells represents a low average speed. In the example shown, it can be seen how an area 622 (a continuous area of ​​a cell) with a relatively low average speed is developed against the driving direction 680, so that at a later time 614, further areas 626, 628 with a relatively low average speed are generated. This is a so-called braking shock wave or braking wave. The macroscopic traffic simulation model can predict this type of effect by means of a numerical simulation of the cell 601.

[0079] Figure 8The process of self-learning traffic control in the framework of an embodiment of the present invention is shown, which includes an "online learning" method. In this case, data is detected by means of a sensor device 702 on the infrastructure side, wherein the data represents the current traffic conditions. In addition, V2X data is received from a plurality of traffic participants 704, the V2X data including topological information of the traffic corridor. In a fusion module 708, an environment model is created from the data and topological information of the sensor device 702 on the infrastructure side. The environment model is provided to a self-learning traffic control unit 706 via an interface 714. The self-learning traffic control unit 706 includes a learning module ("Learning Engine") 703, in which different ML methods can be implemented. The learning module 703 recommends one or more actions for traffic control based on an initially learned utility function 724, which is in the form of a traffic control strategy ("Policy") 726. Driving behavior instructions for traffic participants 704 are derived from the traffic control strategy 726 and transmitted via the interface 712 via V2X communication. The sensor systems 702 of the traffic participants 704 and the infrastructure now collect data again, from which a new environment model is created with the aid of the fusion module 708 and which is provided to the self-learning traffic control unit 706. The learning module 703 can now determine with the aid of the reward function 722 whether the utility function 724 and / or further parameters need to be adjusted in order to improve the policy 726. The self-learning traffic control is achieved via the feedback loop thus created.

[0080] Fig. 9A traffic corridor 800 is shown by way of example, in which a traffic flow is controlled according to a possible embodiment of the present invention. The traffic corridor 800 comprises a three-lane carriageway 810, in which traffic participants 823, 834 move on three lanes along a driving direction 818. The traffic corridor 810 is divided into a plurality of sections 822, 824, 826, 828. At the time point shown, traffic is flowing smoothly in cells 822, 824, 826, while traffic is already congested in cell 828. According to the present invention, the formation of a braking shock wave and the resulting "stop-and-go" traffic should be prevented or mitigated. To this end, the traffic flow can be simulated and optimized for each of the sections 822, 824, 826, 828 by dynamically generating specific, corresponding driving behavior instructions for each section 822, 824, 826, 828. The driving behavior instructions can include, for example, suggestions for speed, spacing and / or lane indications. In order to detect the current traffic conditions (e.g. traffic density and / or average speed) within the cell, a plurality of infrastructure units 804, 808 are fixedly arranged along the traffic corridor 810, each of which has a communication unit. In addition, at least one of the infrastructure units 804, 808 has at least one road traffic sensor. The infrastructure units 808, 808 can send data to and / or receive data from traffic participants 832, 834 via V2X communication with the aid of their corresponding communication units. The data sent can be, for example, driving behavior instructions 816 and / or CPMs ("Cooperative Perception Messages") 814. In addition, the infrastructure units 808, 808 can exchange data with the central traffic control unit 806 with the aid of their corresponding communication units. The traffic control unit 806 is configured to generate corresponding driving behavior instructions for sections 822 , 824 , 826 , 828 , wherein the driving behavior instructions are generated in such a way that a digital model of the traffic corridor is generated with the aid of current traffic conditions detected with the aid of at least one road traffic sensor, a traffic flow is simulated based on the digital model, and the driving behavior instructions are determined based on the simulation of the traffic flow.

[0081] In the situation shown, the vehicles in cell 828 are in a traffic jam and are temporarily stopped. Some of these vehicles are designed as networked motor vehicles 834 and report their own positions and detected surrounding objects to infrastructure unit 804 by means of CAM messages and / or CPM messages 814. Infrastructure unit 804 recognizes the position of the end of the traffic jam and the danger of a possible brake shock wave. The vehicles in cells 822 and 824 are heading towards the traffic jam. The vehicle is traveling and cannot yet detect the traffic delay with its own sensors due to the distance or due to obstructed visibility in dense traffic.

[0082] The infrastructure system 812 (comprising the infrastructure units 804, 808 and the traffic control unit 806) can detect the suboptimal speed ratio of the incoming traffic and transmit the corresponding driving behavior instructions 816, which are intended, for example, to gradually increase the speed of the vehicle 832 in the sections 822, 824 (as the congestion end approaches). For each of the sections 822, 824, 826, a driving behavior instruction 816 is generated, which includes, for example, a maximum recommended speed, a preferred lane, a lane change suggestion and / or an instruction for giving up the right of way. If necessary, the suggestions can be further divided according to the vehicle category, for example, the maximum recommended speed for heavy vehicles can be different and can be defined more specifically. Next, the driving behavior instruction 816 can be transmitted by one of the infrastructure units 804, 808 or both infrastructure units, preferably a message 816 is sent for each section 822, 824, 826, 828.

[0083] Some of the vehicles 834 in the section 828 are networked and use ACC devices or UX devices. Here, these suggested driving behavior instructions 816 can be implemented by human drivers or driver assistance systems. Therefore, these vehicles influence the surrounding traffic that may not be networked through their own behavior. Therefore, the entire traffic can be indirectly affected by the infrastructure. Therefore, the transition between incoming traffic and congestion can become smoother, and accordingly, the speed difference is reduced, which can improve safety and driving comfort. In addition, the reduced maximum speed can help to improve the overall average traffic flow.

Claims

1. A method for traffic flow control, wherein: A driving behavior instruction (816) is generated for at least one traffic participant, in particular a networked motor vehicle (112, 114, 116, 832, 834, 835), and the driving behavior instruction is sent to the at least one traffic participant, in particular to the networked motor vehicle (112, 114, 116, 832, 834, 835), characterized in that The driving behavior instructions (816) are generated in such a way that a digital model of the traffic corridor (810) is generated with the aid of current traffic conditions, a traffic flow is simulated based on the digital model, and the driving behavior instructions (816) are determined based on the simulation of the traffic flow.

2. The method according to claim 1, wherein: For the digital model of the traffic corridor (810), data (702) of a sensor device (102) on the infrastructure side are detected, wherein the data (702) represent current traffic conditions and are processed together with topological information (704) of the traffic corridor.

3. The method according to claim 2, wherein: The topological information of the traffic corridors is included in the digital map of the road network.

4. The method according to claim 2, wherein: In particular, by means of machine learning, topology information of the traffic corridor is determined from sensor data (814), which are received from a plurality of networked motor vehicles.

5. The method according to any one of claims 1 to 4, wherein: The simulation is designed as a real-time capable predictive model with rule-based traffic control.

6. The method according to claim 5, wherein: The simulation implements a machine learning method, wherein a machine learning model is trained with the aid of historically detected traffic conditions.

7. The method according to claim 5, wherein: The simulation realizes physical modeling of traffic flow, especially realizes numerical simulation.

8. The method according to claim 5, wherein: The simulation implements a machine learning method in combination with a physical modeling of the traffic flow, in particular with a numerical simulation.

9. The method according to any one of claims 5 to 8, wherein: With the aid of the real-time capable predictive model, an optimal driving behavior is determined for each specific section (112, 114, 116, 832, 834, 835) of the traffic corridor (810) and a driving behavior instruction (816) is derived therefrom, wherein the driving behavior instruction (816) includes in particular a speed recommendation and / or an acceleration and deceleration target and / or a recommendation for a distance to be maintained from a preceding vehicle and / or a recommendation for using a specific lane.

10. A method according to any one of the preceding claims, wherein: The driving behavior instructions (816) include at least a speed recommendation.

11. A method according to any one of the preceding claims, wherein: The driving behavior indication is calculated and transmitted specifically for the determined lane (816).

12. A method according to any one of the preceding claims, wherein: The simulation includes a self-learning algorithm, wherein the traffic flow control is continuously optimized during operation.

13. A method according to any one of the preceding claims, wherein: The driving maneuver instruction is transmitted to a plurality of road users by means of a multicast communication method and / or a broadcast communication method.

14. A method according to any one of the preceding claims, wherein: The driving behavior instruction is received by a networked motor vehicle (112, 114, 116, 832, 834, 835) and output to a driver of the motor vehicle, in particular in an audiovisual and / or haptic manner.

15. A traffic control unit (106, 706, 806, 906) configured to generate a driving behavior instruction (712, 816) for at least one traffic participant, in particular a networked motor vehicle (112, 114, 116, 832, 834, 835), wherein: The driving behavior instructions (712, 816) are generated in such a way that a digital model of a traffic corridor (810) is generated based on current traffic conditions, a traffic flow is simulated based on the digital model, and the driving behavior instructions (712, 816) are determined based on the simulation of the traffic flow.

16. A device for traffic flow control, in particular an infrastructure system (812) for at least partially automated driving support of guided, networked motor vehicles (832, 834, 835), and in particular designed to implement a method according to any one of claims 1 to 11, comprising at least one communication unit, at least one traffic control unit (106, 706, 806, 906) according to claim 15 and at least one road traffic sensor (102), wherein: The traffic control unit (106, 706, 806, 906) is configured to generate a driving behavior instruction (712, 816) for at least one traffic participant, in particular a networked motor vehicle (112, 114, 116, 832, 834, 835), and the communication unit is configured to send the driving behavior instruction (712, 816) to at least one traffic participant, in particular a networked motor vehicle (112, 114, 116, 832, 834, 835), wherein current traffic conditions are detected in particular with the aid of the road traffic sensor (102).

Images