DE102011083677A1 - Method for predicting traffic conditions for e.g. electric car, involves determining future information for traffic conditions of vehicle based on current state of vehicle and historical data - Google Patents

Abstract

The method involves determining future information for traffic conditions of a vehicle (101) based on a current state of a vehicle and historical data (102) by a time window or a simulation process that is performed by a microscopic model or a macroscopic model, where the current state is selected from a group consisting of a position of the vehicle, speed of the vehicle, a road to be driven, a direction and/or an objective. The historical data is transferred to the vehicle over a mobile communication interface. An action is executed based on the traffic conditions. The historical data is selected from a group consisting of traffic density, average speed, speed curve, journey time, traffic light cycle, traffic volume and/or traffic condition. The future information is selected from a group consisting of information from different vehicle, information from a transport infrastructure component, information from a traffic control center, traffic information, information of a digital map, information of a driver assistance system, online data, position of a jam, a future velocity curve, a future acceleration curve, waiting time and queue length at a traffic system, speed strategy when approaching a traffic system and/or safe and efficient driving speed recommendation. The action is selected from a group consisting of an optical feedback, a haptic feedback, intervention into controlling of the vehicle, manufacturing a communication link and/or transmitting a message. An independent claim is also included for a device for predicting traffic conditions for a vehicle.

Description

The invention relates to a method for forecasting a traffic situation for a vehicle and an associated device. Furthermore, a vehicle is proposed with such a device.

A communication from vehicle to vehicle as well as a communication between vehicle and infrastructure (also referred to as C2X communication), traffic flow simulation method as well as methods for the detection of disturbances (eg for traffic jam recognition) are known. Certain traffic flow data may be used to control actions in a vehicle. For example, it is known to evaluate so-called floating-car data or extended floating-car data in a vehicle (see also http://de.wikipedia.org/wiki/Floating_Car_Data ).

Furthermore, sequential Monte Carlo methods (SMC methods) are known ( http://de.wikipedia.org/wiki/Sequenzielle_Monte-Carlo-Methode ), which belongs to the class of stochastic methods for state estimation in a dynamic process (eg in mobile robotics), whose dynamics is known only statistically (significant disturbances) and which can only be partially observed (subdivision into inner, hidden and outer, visible variable). An application example is the exact and continuously updated determination of the location and the speed of an object due to an inaccurate and erroneous measurement of the location (see Tracking). SMC filters are also known as particle filters, sampling importance resampling (SIR), sequential importance sampling (SIS), bootstrap filters, condensation trackers, interacting particle approximations or survival of the fittest.

In this case, it is disadvantageous that no reliable inference can be made about a state of a vehicle which is about to arrive shortly in its respective traffic situation.

The object of the invention is to avoid the above-mentioned disadvantages and in particular to provide an efficient solution for predicting a traffic situation for a vehicle.

This object is achieved according to the features of the independent claims. Further developments of the invention will become apparent from the dependent claims.

In order to achieve the object, a method for forecasting a traffic situation for a vehicle is specified in which, based on a current state of the vehicle and based on historical data, at least one possible future for the traffic situation of the vehicle is determined.

The historical data is, in particular, data without an actuality reference. The historical data may come from different sources (eg, vehicles, infrastructure, etc.) and be stored as information for an area (eg, a road, a stretch, etc.). The historical data can additionally be sorted or segmented according to time and / or area. In particular, with regard to different data profiles can be created, for. As a traffic density over a distance and / or time.

The proposed solution makes it possible to predict one or more possible traffic situations for a time-limited future of a vehicle (also referred to as an EGO vehicle). This may be based on historical data and a current state of the vehicle. In addition, additional data can be taken into account to improve the quality of the forecast. Also, one of the several possible future traffic situations can be selected for the vehicle most suitable traffic situation. This most appropriate traffic situation can be used to detect a specific event and initiate a corresponding action.

• – eine Position des Fahrzeugs;
• – eine Geschwindigkeit des Fahrzeugs;
• – eine zu fahrende Strecke;
• – eine Richtung;
• – ein Ziel.

A development is that the current state of the vehicle comprises at least one of the following parameters:

• A position of the vehicle;
• A speed of the vehicle;
• - a route to be traveled;
• - a direction;
• - a target.

The goal may be z. For example, it may be a (near) intermediate or final destination.

The current state of the vehicle may, for. B. by means of the on-board sensor of the vehicle, for. B. a bus system and / or a control unit, are determined.

• – eine Verkehrsdichte;
• – eine Durchschnittsgeschwindigkeit;
• – einen Geschwindigkeitsverlauf;
• – eine Reisezeit;
• – einen Verkehrszyklus, insbesondere einen Ampelzyklus;
• – eine Verkehrsstärke;
• – eine Verkehrslage.

Another development is that the historical data include at least one of the following parameters:

• - a traffic density;
• - an average speed;
• - a speed gradient;
• - a travel time;
• A traffic cycle, in particular a traffic light cycle;
• - a traffic volume;
• - a traffic situation.

The traffic situation is in particular at least one traffic condition, eg. An averaged state over a predetermined link (eg, sections defined by adjacent TMC locations).

The historical data can be z. B. be collected or provided by a central unit (backend) of a vehicle manufacturer and / or service provider. The historical data may be data from multiple vehicles.

In particular, it is a development that the historical data of different vehicles are aggregated.

So the data z. B. compacted based on cells of a grid, which corresponds to an area of a map, into profiles. Such aggregation can result in time-dependent profile information being provided to a plurality of vehicles.

It is also a development that the historical data is transmitted to the vehicle via a mobile communication interface.

In principle, different communication interfaces are possible for the transmission of (historical and other) data. Examples include mobile radio, Bluetooth, WiFi, radar, etc. Also, a transmission or updating of data at least partially via a physical connection, for. B. a charging cable done.

• – eine Information von mindestens einem anderen Fahrzeug;
• – eine Information von einer Verkehrsinfrastruktur-Komponente;
• – eine Information von einer Verkehrszentrale;
• – eine Verkehrsinformation;
• – eine Information von einer digitalen Karte;
• – eine Information eines Fahrassistenzsystems;
• – Online-Daten.

Furthermore, it is a further development that the at least one possible future of the vehicle is also determined on the basis of at least one of the following information:

• An information of at least one other vehicle;
• An information from a traffic infrastructure component;
• An information from a traffic center;
• A traffic information;
• An information from a digital map;
• An information of a driver assistance system;
• - Online data.

The traffic infrastructure component can be a traffic sign system (eg a traffic light), a railroad crossing, a traffic monitoring or observation unit or the like. The online data can be z. B. be retrievable via the Internet (the access can be made using the above-mentioned mobile communication interface).

In the context of an additional development, the at least one possible future of the vehicle is also determined by means of a time window, wherein the state of the vehicle is recorded for a predefined period of time on the basis of the time window.

In particular, the time window may determine a time period that extends from the present to the past for a predetermined period of time. For example, the time window may have a period of 10 seconds to 300 seconds.

• – eine Position eines Stauendes;
• – ein künftiger Geschwindigkeitsverlauf;
• – ein künftiger Beschleunigungsverlauf;
• – eine Wartezeit und Staulänge vor einer Verkehrsanlage;
• – eine grüne Welle oder eine Geschwindigkeitsstrategie bei Annäherung an eine Verkehrsanlage;
• – eine Geschwindigkeitsempfehlung für sicheres oder effizientes Fahren.

A further development consists in that the possible future for the traffic situation of the vehicle comprises at least one of the following possibilities (possibly with an associated uncertainty or risk assessment):

• A position of a jam end;
• - a future course of speed;
• A future course of acceleration;
• - a waiting time and traffic jam length in front of a traffic facility;
• A green wave or a speed strategy approaching a traffic facility;
• - A speed recommendation for safe or efficient driving.

The possible future for the traffic situations can be forecasts and / or actions.

One embodiment is that the at least one possible future traffic situation for a short-term future of the vehicle is determined by means of a simulation.

The short-term future concerns z. B. a period of up to two minutes. In particular, the simulation may include a plurality of simulation runs, such that a plurality of trajectories may be determined for the EGO vehicle.

The simulation can be a so-called Monte Carlo simulation (particle filter simulation): The particle filter belongs to the class of stochastic methods for state estimation in a dynamic process, which can only be observed incompletely. By means of the particle filter, a current but unknown probability density for a state space can be estimated in order to derive statements about the further traffic flow.

By means of the particle filter, a cloud or a swarm of so-called particles is generated, which are intended to represent points in the state space. Each particle is assigned to a weight. The swarm as a whole represents the probability density. Starting from an initial state, one or more solution curves are assigned to each individual particle by means of a stochastic model of system dynamics. This is a so-called Monte Carlo simulation, because the temporal development of the system is subject to random disturbances. Depending on how the predictions of the measured values derived from the respective solution curve agree with the actual measured values, the weight of the particle can be adjusted, which results in iteratively an improved estimation of the evolution of the probability density in the state space. Highly weighted particles can also be "propagated" according to their weight.

One embodiment is that the simulation is carried out by means of a microscopic model and / or a macroscopic model for a traffic flow.

A next embodiment is that the microscopic model includes provisions for simulating a movement of a vehicle taking into account the movements of the remaining vehicles.

It is also an embodiment that is modeled on the basis of the macroscopic model of the traffic flow as a kind of liquid with predetermined initial conditions and / or boundary conditions.

For example, the acceleration of a virtual vehicle calculated at a given time depends, on the one hand (microscopically), on the distance to the vehicle in front and on the speed of one's own and the vehicle ahead. On the other hand, according to the macroscopic model, the vehicles attempt to achieve a predetermined, spatially dependent speed characteristic estimated from historical data.

A further development is that the simulation determines a future traffic flow in the direction of travel of the vehicle.

Based on the simulation, a future traffic flow in the direction of travel of the vehicle (in the direction of the possible traveled by the vehicle route, also referred to as "downstream") can be determined. In particular, the simulation can have several simulation runs in order to determine the several future possible traffic situations.

Accordingly, a stochastic method for state estimation in a dynamic process is proposed, whereby the dynamic process itself can also be incompletely observed. Using existing observations (in the form of data) and using a suitable process model, an unknown state can be estimated and taken into account for the forecast.

An additional embodiment is that in the simulation of data from at least one other (sample) vehicle mapped to a virtual vehicle and according to the historical data and other possible data virtual vehicles for mapping a corresponding traffic flow generated and their effects on the (EGO -) Vehicle to be determined.

With the help of a stochastic traffic simulation, a representative collection of probable courses can be generated and z. B. in the form of a database the vehicle (EGO vehicle) to determine the trajectory (s) are provided. Using the database, it is possible to determine individual forecasting, forecast blurring, spatial distribution of risks, speed variations and other statistical features.

The approach makes it possible to increase the safety of the user in the EGO vehicle and / or the energy efficiency. Furthermore, it is possible to improve or even enable comfort functions for the user (prediction of a traffic light phase upon arrival at the traffic light, estimation of the waiting time, use of the waiting time by presentation of multimedia content, etc.).

Another embodiment is that several future traffic situations for the vehicle are determined and that traffic situation is selected, which is determined by means of a quality function as the most appropriate traffic situation for the vehicle.

It is also a possibility that based on the most appropriate traffic situation, a predetermined action is performed.

The predetermined action may be performed depending on an application of the vehicle.

• – eine optische Rückmeldung;
• – eine haptische Rückmeldung;
• – eine Eingriff in die Steuerung des Fahrzeugs;
• – ein Herstellen einer Kommunikationsverbindung;
• – ein Übermitteln einer Nachricht.

It is also a further embodiment that the predetermined action comprises at least one of the following possibilities:

• - an optical feedback;
• - a haptic feedback;
• - an intervention in the control of the vehicle;
• - establishing a communication connection;
• - Submitting a message.

Another development is that the possible future for the traffic situation of the vehicle includes a trajectory and / or a speed profile of the vehicle.

The above-mentioned object is also achieved by a device for providing a prognosis of a traffic situation for a vehicle having a processing unit that is set up in such a way that at least one possible future for the traffic situation of the vehicle can be determined on the basis of a current state of the vehicle and historical data ,

The processing unit mentioned here may in particular be embodied as a processor unit and / or an at least partially hard-wired or logical circuit arrangement, which is set up, for example, in such a way that the method can be carried out as described herein. Said processing unit may be or include any type of processor or computer or computer with correspondingly necessary peripherals (memory, input / output interfaces, input / output devices, etc.).

The above explanations regarding the method apply to the device accordingly. The device may be implemented in one component or distributed in several components.

Also, in order to achieve the above object, there is proposed a vehicle having a device as described herein.

The solution presented here further comprises a computer program product which can be loaded directly into a memory of a digital computer, comprising program code parts which are suitable for performing steps of the method described here.

Furthermore, the above problem is solved by means of a computer readable storage medium, e.g. Arbitrary memory, including computer-executable instructions (e.g., in the form of program code) that are suitable for the computer to perform steps of the method described herein.

Embodiments of the invention are illustrated and explained below with reference to the drawings.

Show it:

1 a schematic arrangement for illustrating an efficient prognosis for an EGO vehicle;

2 a schematic diagram of an exemplary implementation of the proposed solution.

In order to create a suitable prediction, preferably different sources of information are taken into account, eg. Historical speed profiles, C2X information (eg, information from other vehicles, infrastructure elements, traffic control centers), onboard sensor (current vehicle and / or other vehicle) information, traffic information, digital map data, driver assistance system data, and / or online data (eg, data available or retrievable via a mobile internet connection).

It is proposed, starting from a current driving state of a currently considered vehicle (also referred to as EGO vehicle) and z. On the basis of (suitable) historical speed profiles to use the available data in the context of repeated traffic flow simulations to initially close to a short-term future vehicle dynamics. The traffic flow simulation takes place, for example, taking into account stochastic elements and supplies, preferably with the aid of a particle filter, an estimated probability distribution of the local traffic situation.

• – eine Position eines Stauendes mit einer zugehörigen Risikoschätzung,
• – ein künftiger Geschwindigkeitsverlauf,
• – ein künftiger Beschleunigungsverlauf,
• – eine Wartezeit und Staulänge vor Verkehrsanlagen, z. B. Ampeln,
• – eine grüne Welle bzw. eine Geschwindigkeitsstrategie bei Annäherung an eine Verkehrsanlage, z. B. Ampel,
• – eine Geschwindigkeitsempfehlung für sicheres und effizientes Fahren.

Based on the probability distribution, among other things, the following information for the EGO vehicle can be provided or improved:

• A position of a tail end with an associated risk estimate,
• - a future speed course,
• - a future acceleration course,
• - A waiting time and traffic jam length before traffic systems, z. Traffic lights,
• A green wave or a speed strategy when approaching a traffic facility, e.g. B. traffic light,
• - A speed recommendation for safe and efficient driving.

This merging of data exploits the fact that traffic does not behave completely randomly but follows certain rules. For example, vehicles do not emerge at a point B out of nowhere, they have to move from point A to there. In addition, traffic patterns in a given road network usually repeat in terms of time of day and / or day of the week. An example of this is the higher traffic density during a rush hour.

These patterns can be derived from already existing (eg previously recorded) historical velocity profiles.

Known information and data may be merged using a stochastic simulation-based approach to derive at least one forecast for the EGO vehicle. Based on the current state of the environment of the EGO vehicle, several possible future motion profiles are determined by means of several simulated traffic passes. The movement profiles can be weighted according to the information about the past, present and future. This can be real future information such as a length of future signal phases that are provided by a traffic light (eg sent) and / or historical data (and derived information) be taken into account. Results can be weighted and it is possible to make statements about the possibility of occurrence or non-occurrence of certain future events. For example, a speed profile can be determined by means of which upcoming traffic events can be handled comfortably, safely and efficiently.

Thus, a road section in front of the EGO vehicle can be taken into account, wherein the simulations should preferably proceed faster than in real time in order to generate as many movement profiles as possible and still allow predictions in real time.

A measure of a quality preferably indicates how well a system performs its task. In the present case, the usefulness of an operating unit (eg, the EGO vehicle or a control unit) can be increased by avoiding situations where utility is reduced. Such a situation may be described as an event that limits the choices of an acting entity in a way that forces them to perform an action that they would not have performed if they had previously known the event. For example, consider an event "road blocked by boulders": the earlier the EGO vehicle recognizes the event, the more time there is to respond. If the event is detected at a very early stage, there are a large number of possible reactions (eg driving around) than with a late detection (eg initiation of emergency braking). Thus, in the case of early recognition, it is possible to choose the reaction which still promises the greatest benefit in terms of an objective function.

When viewed on a scale between 50 m and 200 m, a traffic flow is an uncertain, seemingly noisy process with large fluctuations and fluctuations. It is proposed to discover a system in the process and thus improve the predictability of the traffic even at a relatively fine resolution (granularity). For example, it is possible to use probabilistic methods to predict a near future for the EGO vehicle and derive corresponding actions from it.

1 shows a schematic arrangement for illustrating an efficient prognosis for an EGO vehicle 101 ,

The input data is historical data 102 (Driving data without up-to-dateness originating from the EGO vehicle and / or from other vehicles), data from the EGO vehicle 101 , Data from other vehicles 108 , Traffic information 109 as well as data of a traffic infrastructure 110 used. The data of other vehicles 108 may include data from such vehicles downstream as viewed from the EGO vehicle 101 are traveling, these data being the EGO vehicle 101 can be provided by means of C2C communication (directly or via a central server, for example in the back end of a vehicle manufacturer or service provider).

The historical data 102 can be collected by different vehicles, if necessary structured by cells ("clustered") and stored in a central or distributed database. The historical data 102 can by means of an aggregation 103 be merged to use simulation parameters such. B. derive a traffic density or average speed. The aggregated data becomes a data fusion 104 (Merging the data) provided to a statement about trajectories of the EGO vehicle by means of a simulation and a merit function 101 to enable. Also the data of the EGO vehicle 101 , the other vehicles 108 , the traffic information 109 and the traffic infrastructure data 110 (if available) the data fusion 104 provided.

The data of the EGO vehicle 101 include, for example, a position and a speed of the EGO vehicle 101 , You can via sensors of the EGO vehicle 101 , z. B. by access to a CAN bus in the vehicle, are determined. The position may be a map-fitted and filtered GPS position.

The simulation uses the provided data (the collected knowledge) to generate (virtual) vehicles. In particular, positions and speeds of downstream vehicles may be mapped directly to a particular virtual vehicle in the simulation. This results in the simulation so vehicles that at least for a certain period of a real other vehicle (also referred to as a trial vehicle) correspond, and virtual vehicles. Based on the vehicles in the simulation, the predetermined traffic density is specified according to the traffic parameters.

Thus, the virtual vehicles can be used to propagate traffic events, eg. As auto queues, congestion or compaction based on the simulation to determine.

Once the simulation has been defined, a stochastic description can be generated by means of repeated simulation runs, in which representative patterns are derived from the driver and traffic property models. Thus, a measure of the uncertainty / safety can be determined and a risk assessment or probability analysis can be supported. For example, the speed profiles of the virtual vehicles are adapted to the past profile of the EGO vehicle for each simulation run. The adaptation is performed using a quality function that takes into account the last determined speed and the travel time profile of the EGO vehicle. The best match profile is assumed to be the virtual representation of the actual EGO vehicle. Accordingly, the future behavior of this most suitable virtual vehicle is considered to be the best prediction for the EGO vehicle.

The predictions from the simulation runs are stored in a database 105 saved. Due to the different parameter sets of each run, the database contains 105 thus several possible versions of the near future for the EGO vehicle 101 , An application 106 can the database 105 to determine the likelihood of a particular event taking place. The application 106 can then initiate or perform a predetermined action and / or on a display unit 107 a corresponding message z. B. for the driver of the EGO vehicle 101 output. This allows, for example, driver and situation-adapted information in the EGO vehicle 101 ,

For example, an application 106 "Emergency stop warning" the database 105 poll for speed profiles that include decelerating by more than a preset limit. If there is a statistically significant accumulation of such events in a location, a warning may appear on the display unit 107 be issued. Alternatively or additionally, with an application 106 "Emergency brake" initiated or carried out a corresponding braking operation.

2 shows an exemplary implementation of the proposed solution for the EGO vehicle 101 ,

The application 106 receives vehicle data 201 , ie information about the current state of the EGO vehicle 101 , z. B. via the CAN bus. An external communication interface connects the EGO vehicle 101 using a UMTS router 202 over the internet 203 with a backend server 204 , The backend server 204 collects z. For example, historical data from multiple suitably equipped vehicles stores them and provides the stored information to the vehicles. Such a central storage of the data has the advantage that each vehicle can access historical data of other vehicles in addition to its own historical data. In particular, the data may be from the backend server 204 preprocessed and z. B. be offered in the form of stochastic profiles.

Optionally, the backend server 204 act as a message relay and distribute the current state of each connected vehicle to all vehicles that may be interested in this information.

Further considerations

By networking vehicles with each other (C2C) or with an infrastructure (C2I), traffic safety, efficiency and / or comfort can be increased. Based on different data, eg. As C2C data, C2I data, historical driving profiles (of the current vehicle and / or other vehicles), (extended) floating-car data and other data sources to determine a trajectory and / or a speed profile for the EGO vehicle become. Thus, by means of different sources of information, by means of a corresponding fusion of the available data, short-term forecasts concerning the driving situation of the EGO vehicle (forecast of the trajectory of the EGO vehicle) can be generated.

In this case, the trajectory of the EGO vehicle can not be predicted completely or with a different probability of its occurrence. Thus, the trajectory of the EGO vehicle is part of a dynamic overall system, many vehicles and units of the infrastructure are autonomous units whose behavior itself can only be roughly predestined (with uncertainties).

Preferably, therefore, the prognosis for the EGO vehicle based on a traffic course during a predetermined time interval (also referred to as time window) z. B. with a specific spatial extension. It is possible to make the temporal interval or the spatial extent itself dependent on an ascertainable uncertainty of the prognosis: the greater the (expected) error (uncertainty of the calculation) determined in the prognosis, the smaller the time window. In this way, z. B. a determined error always remain below a predetermined threshold, the time window is adjusted accordingly (with the time window also results from the currently considered speed dependent spatial extent).

It results z. B. an initial or boundary value problem to be solved. Furthermore, a traffic flow model is presented, which can merge existing data and in particular compensates for missing data as far as possible.

In particular, a probabilistic approach is suggested. With the help of a stochastic traffic simulation, a representative collection of probable courses can be generated and z. B. in the form of a database the EGO vehicle to determine the trajectory (s) are provided. Using the database, it is possible to determine individual forecasting, forecast blurring, spatial distribution of risks, speed variations and other statistical features.

The approach makes it possible to increase the safety of the user in the EGO vehicle and / or the energy efficiency. Furthermore, it is possible to improve or even enable comfort functions for the user (prediction of a traffic light phase upon arrival at the traffic light, estimation of the waiting time, use of the waiting time by presentation of multimedia content, etc.).

Objective, motivation

For example, it is assumed that the driver of the EGO vehicle can not freely choose his speed and thus his trajectory. Rather, this is limited or influenced by external factors. Thus, in a traffic flow of consecutively moving vehicles, the driver of the EGO vehicle must constantly react to dynamic influences of the preceding traffic flow, eg. B. adjust its speed. In such a scenario, for the driver of the EGO vehicle, a large number of influences only become immediately apparent (eg with a time horizon of less than 5 seconds), so that he can only react to changes.

If a short-term forecast (with a time horizon of between 5 and 30 seconds) of the traffic flow was available in real-time, the EGO vehicle or driver could anticipate or eliminate many immediate reactions through skilful proactive (or strategic) actions. This would have a beneficial effect on traffic safety and a more harmonious flow of traffic. For example, it would be possible to detect potential dangerous situations in advance and to adjust the speed of the EGO vehicle in good time. Preferably, the information available could be appropriately prepared to extend the driver's perceptual horizon beyond the immediate visual horizon. Also, corresponding actions can be autonomously determined by the EGO vehicle and suggested or performed to the driver (and possibly displayed to the driver). To achieve the extended time horizon, it is advantageous to be able to evaluate a multiplicity of data, wherein preferably the data has high reliability (that is to say a high quality or a low forecast error).

The approach makes it possible to increase safety (for the driver and passengers of the EGO vehicle as well as the other road users), the energy consumption (through predictive driving maneuvers) and the energy management in the EGO vehicle (decision on efficient recuperation with regard to upcoming driving maneuvers) to optimize. Additional comfort applications can also be created and existing comfort applications for drivers and passengers of the EGO vehicle improved.

The data available in the EGO vehicle (self-generated data or data made available by means of external data sources) can be combined in such a way that a prognosis of the (future) trajectory of the EGO vehicle can be determined. This trajectory of the EGO vehicle may be subject to uncertainties, the uncertainties increasing in particular as the forecast horizon increases. Forecasts include in particular probability distributions regarding z. For example, expectancy values of the velocity, and confidence intervals or quantiles (ie time ranges in which a prediction with a certain probability applies).

Based on this consideration, it is a goal to use the available data so that the prognosis of the trajectory of the EGO vehicle is improved. In particular, the trajectory (eg the speed profile) of the EGO vehicle with an associated prognosis uncertainty is to be predicted. Preferably, this is done in real time or near real time.

In an exemplary scenario, a preceding vehicle (also referred to as a sample vehicle) brakes to a stop and then restarts. The EGO vehicle has exemplary access to previous driving information (including the test vehicle) as well as to other data sources. The sample vehicle transmits z. B. data packets with timestamps, positions and speeds to a central office, which forwards this data (after a few seconds latency) in the EGO vehicle.

• – Es wird eine Geschwindigkeitsvorausschau für das EGO-Fahrzeug erzeugt, d. h., eine Geschwindigkeit mit der das EGO-Fahrzeug zu bestimmten Zeitpunkten (etwa in 5, 10, 15, 20 Sekunden) im Verkehrsfluss fahren kann.
• – Es wird ermittelt, ob für das EGO-Fahrzeug eine Notwendigkeit zum Abbremsen besteht. In diesem Sinne kann – eine Position und ein Zeitpunkt für ein Abbremsen mit einer bekannten Unschärfe prognostiziert werden und/oder – bei einer möglichen starken Bremsverzögerung des vorausfahrenden Verkehrs ggf. eine Warnung generiert werden (alternativ oder zusätzlich kann eine vorgegebene Aktion, z. B. ein entsprechender Bremsvorgang eingeleitet werden).

The approach presented here includes, by way of example, the following steps:

• A speed forecast is generated for the EGO vehicle, ie, a speed with which the EGO vehicle can drive in the traffic flow at certain times (approximately in 5, 10, 15, 20 seconds).
• - It is determined whether there is a need for the EGO vehicle to decelerate. In this sense, a position and a time for deceleration with a known blur can be predicted and / or a warning may be generated if there is a possible strong deceleration of the preceding traffic (alternatively or additionally, a predetermined action, eg. a corresponding braking process can be initiated).

forecast

The prognosis of the future trajectory of the EGO vehicle is based z. On estimates of the current state of the upcoming traffic flow (also referred to as traffic flow downstream) as well as models relating to the dynamic development of the current state. Thus, a traffic dynamics has the property that disturbances in the traffic flow upstream move with a certain lawfulness. Such traffic dynamics can be detected or modeled.

Accordingly, a stochastic method for state estimation in a dynamic process is proposed, whereby the dynamic process itself can also be incompletely observed. Using existing observations (in the form of data) and using a suitable process model, an unknown state can be estimated and taken into account for the forecast.

Due to the high dynamics of the individual autonomous participants in the traffic, as stated above, no exact prediction is possible. However, a bandwidth (or blurring) of the possible courses of traffic can be determined and also quantified in the context of the prognosis.

• – eine Notwendigkeit zum Zusammenführen unterschiedlicher Daten(quellen);
• – eine vorgegebene Qualität (Schärfe der Prognose) für eine bestimmte Anwendung (z. B. das Einleiten eines automatisierten Bremsvorgangs);
• – Anforderungen betreffend eine Performance in einer Datenumgebung im Fahrzeug bzw. in einem Backend, das zumindest zeitweise eine Kommunikationsverbindung mit dem Fahrzeug hat.

In particular, all or part of the following aspects are considered:

• - a need to merge different data (sources);
• A predetermined quality (sharpness of the prognosis) for a particular application (eg the initiation of an automated braking process);
• Requirements relating to performance in a data environment in the vehicle or in a backend, which at least temporarily has a communication link with the vehicle.

Thus, preferably, a stochastic method of state estimation is sought in a dynamic process that can only be observed incompletely.

Particulate filter (Sequential Monte Carlo method)

It is proposed to determine the most probable trajectories of the EGO vehicle (run-time forecasts with uncertainty and probability) and, for B. in a database store. Thus, the trajectories can be provided to different applications in the EGO vehicle and further processed accordingly.

At least one application in this way can access the stored trajectories at any time and a suitable strategy, eg. Based on rules for the particular application, determine. The applications may be a dynamic distance control to the vehicle in front (ACC), an energy management of the vehicle, a brake assist or a charge management (for a battery of the vehicle, especially in a hybrid or electric vehicle).

The particle filter belongs to the class of stochastic methods for state estimation in a dynamic process, which can only be observed incompletely. By means of the particle filter, a current but unknown probability density for a state space can be estimated in order to derive statements about the further traffic flow.

By means of the particle filter, a cloud or a swarm of so-called particles is generated, which are intended to represent points in the state space. Each particle is assigned to a weight. The swarm as a whole represents the probability density. Starting from an initial state, one or more solution curves are assigned to each individual particle by means of a stochastic model of system dynamics. This is a so-called Monte Carlo simulation, because the temporal development of the system is subject to random disturbances. Depending on how the predictions of the measured values derived from the respective solution curve agree with the actual measured values, the weight of the particle can be adjusted, which results in iteratively an improved estimation of the evolution of the probability density in the state space. High weight particles can also be "propagated" according to their weight.

• – Es wird die gesamte unbekannte a-posteriori-Wahrscheinlichkeitsdichte geschätzt; der Ansatz ist insbesondere für nicht-Gaußsche (auch bezeichnet als multimodale) Verteilungen anwendbar.
• – Systemdynamik und Messdynamik können nichtlinear sein.
• – Die Simulation der einzelnen Partikel ist parallel durchführbar (somit ist ein parallelisierbarer Lösungsansatz für die Durchführung der Simulation möglich).

The particulate filter method has the following advantages.

• - the entire unknown a posteriori probability density is estimated; the approach is particularly applicable to non-Gaussian (also called multimodal) distributions.
• - System dynamics and dynamics can be nonlinear.
• - The simulation of the individual particles can be carried out in parallel (thus a parallel approach to the simulation is possible).

In the example of traffic simulation, the state space has a high dimension, for example, only a few punctual measured values (the trajectories of the ego vehicle and the signals of the sample vehicle) are available to adapt weights or to generate new particles. It is also possible that the resources for calculations in the high-dimensional state space are limiting factors. To constrain the dimension and enable real-time predictions, C2X data (such as signals from multiple sample vehicles and / or from a traffic infrastructure, such as traffic lights) can be integrated as direct drivers in the simulation.

Simulation of traffic flow: requirements

As mentioned above, it is assumed that the driver of the EGO vehicle can not freely choose his speed and thus his trajectory. Rather, this is limited or influenced by external factors.

Now, it is an objective to qualitatively improve the predicted trajectory (i.e., increase the prediction accuracy) from different data that the EGO vehicle can access. It should be noted that such processing of the data may be performed in the EGO vehicle or in another (possibly central) unit for the EGO vehicle. In the latter case, the result of the calculation can be communicated to the EGO vehicle in a timely manner. Also, any division of the calculations may be divided among different components, one component of which may be the EGO vehicle. This allows z. For example, a distributed calculation of complex forecasts based on different data. In particular, a calculation can be made where resources are available for the calculation. An allocation of resources can also be done dynamically.

• – Daten von Infrastruktur-Komponenten, z. B. Verkehrszeichenanlagen, Ampeln, etc.
• – Daten zu Schaltzeitpunkten von Infrastruktur-Komponenten,
• – Daten vorausfahrender Fahrzeuge (Probe-Fahrzeuge), z. B. Trajektorien oder sonstige Daten,
• – historische Daten des EGO-Fahrzeugs und/oder mindestens eines Probe-Fahrzeugs,
• – historische Daten einer zentralen Datenbank.

The different data can come from different sources of information and be different in nature. Here are some examples:

• - Data of infrastructure components, eg. B. traffic signs systems, traffic lights, etc.
• - data on switching times of infrastructure components,
• - Data of preceding vehicles (sample vehicles), eg. Trajectories or other data,
• Historical data of the EGO vehicle and / or at least one test vehicle,
• - historical data of a central database.

The data can influence the EGO vehicle or the trajectory to be determined more or less indirectly.

In principle, it should be noted that the communication between the above-mentioned components (vehicles, infrastructure components, backend, etc.) can take place via different interfaces. Examples include wireless communication interfaces: mobile radio, Bluetooth, WiFi, radar, etc. Also, communication may be via a line (eg, downloading data via a cable during a charge phase of a battery of the EGO vehicle). In addition, the EGO vehicle can generate and store different data, e.g. B. Sensor data from vehicle sensors or existing in the EGO vehicle cameras.

To model the traffic flow, there are different approaches, such. B. depending on the modeling task. For example, models can be used that can simulate microscopically and model macroscopic influences.

• – Berücksichtigung von historischen Profilen: Insbesondere können Rückfall-Szenarien vorgesehen sein, falls historische Daten nicht in ausreichendem Maß vorhanden sind;
• – Berücksichtigung von Randbedingungen (z. B. Daten über Kreuzungen und Ampelschaltzeitpunkte);
• – Berücksichtigung von Geschwindigkeitsverläufen aus Probe-Fahrzeugen; hierbei sollten insbesondere Latenz-Zeiten, die durch die Übertragung entstehen, tolerierbar sein;
• – Berücksichtigung von Geschwindigkeitsverläufen aus ACC-Daten (ACC: ”adaptive cruise control” = aktive Geschwindigkeitsregelung).

For example, different sources of data have the following requirements:

• - consideration of historical profiles: in particular, fallback scenarios may be foreseen if historical data are not sufficiently available;
• - consideration of boundary conditions (eg data on intersections and traffic light switching times);
• - Consideration of speed profiles from trial vehicles; in particular, latency times resulting from the transmission should be tolerable;
• - Consideration of speed curves from ACC data (ACC: "adaptive cruise control" = active cruise control).

• – Aussage über starke Bremsvorgänge;
• – Aussagen über Wahrscheinlichkeitsverteilungen, etwa für Energieoptimierung und Energiemanagement;
• – Aussagen über Wartezeiten;
• – der Ansatz soll sowohl auf Autobahnen als auch im städtischen Verkehr nutzbar sein.

For example, the following requirements arise for different applications:

• - statement about heavy braking;
• - statements about probability distributions, such as energy optimization and energy management;
• - statements about waiting times;
• - The approach should be usable both on motorways and in urban transport.

Microscopic models of traffic flow

For central and generally accepted tasks of traffic planning, including the dimensioning of infrastructure in road construction, etc., representative processes should generally be generated and simulated. Thus, for example, hypothetical traffic regulations can be tested in terms of their capacity. For this, so-called "microscopic models" are often used. This is understood to mean rules for simulating the movement of each individual vehicle, taking into account the movements of the other vehicles.

For example, a simple vehicle following model presupposes a regulation for determining the acceleration of the next (n + 1) th vehicle according to FIG

The coefficient λ is obtained by calibration with empirical data.

In this vehicle follow-up model, the distance and the difference speed to the car in front (vehicle ahead) are taken into account. In addition, a delay T is assumed, which is intended to express the reaction processes of the driver.

Even more complex microscopic models mainly consider the distance to the vehicle in front and the speed of the EGO vehicle and the person in front. Occasionally, the vehicle ahead of the vehicle in front is taken into account. In this context, the program VISSIM is known ( http://de.wikipedia.org/wiki/VISSIM ), which is often used for simulations of new traffic facilities.

• – Eine Integration von Randbedingungen sollte möglichst direkt in die Simulation erfolgen.
• – Bei Daten über Kreuzungen und Ampelschaltzeitpunkten könnten mikroskopische Fahrzeugfolgemodelle eingesetzt werden.
• – Auch sollten Daten über vorausfahrende Fahrzeuge in der Simulation berücksichtigt werden können. Insofern kann das bekannte mikroskopische Modell um die Berücksichtigung vorgegebener Trajektorien (aus Probe-Fahrzeugen) erweitert werden.
• – Die zu berücksichtigenden Daten umfassen auch historische Geschwindigkeitsprofile, die dazu verwendet werden sollen, den Verlauf der Geschwindigkeit auf einer bereits befahrenen Strecke besser schätzen zu können, als dies lediglich mit einer digitalen Karte möglich wäre. Auch diesbezüglich ist eine Erweiterung bekannter mikroskopischer Modelle erforderlich.

Microscopic models have been adapted and applied in recent years to model the effect of driver assistance systems. For the use of a microscopic model, the following aspects in particular are important:

• - An integration of constraints should be made as directly as possible in the simulation.
• - For data on intersections and traffic light switching times, microscopic vehicle following models could be used.
• - Also data about vehicles ahead should be considered in the simulation. In this respect, the known microscopic model can be extended to include given trajectories (from trial vehicles).
• - The data to be considered also includes historical speed profiles, which should be used to better estimate the course of the speed on an already traveled route than would be possible only with a digital map. Also in this regard, an extension of known microscopic models is required.

Macroscopic models of traffic flow

In order to illustrate the use of a given speed characteristic in the context of a traffic model, a macroscopic model can be used to describe the traffic.

Macroscopic traffic models assume a hydrodynamic traffic flow, ie the traffic flow is modeled as a kind of fluid:

x

eine Position,

t

einen Zeitpunkt,

k(x, t)

eine Verkehrsdichte in Fahrzeugen pro Meter,

v(x, t)

eine Verkehrsflussgeschwindigkeit an der Stelle x zu dem Zeitpunkt t in m/s,

μ

eine Diffusion in dem Verkehrsfluss,

c₀

eine Fortpflanzungsgeschwindigkeit.

They denote:

x

a position,

t

a time

k (x, t)

a traffic density in vehicles per meter,

v (x, t)

a traffic flow velocity at the point x at the time t in m / s,

μ

a diffusion in the traffic flow,

c ₀

a reproductive rate.

The notation means the partial derivative with respect to x, and so forth. The left side of the second equation (2) corresponds to an acceleration (dv / dt) of a hypothetical vehicle "moving" along with the traffic liquid, which is at point x at time t. Thus, the right side of equation (2) is a default for this acceleration.

With regard to hydrodynamics, the terms with derivatives of k express reaction and anticipation processes in the traffic flow which are suitable for solution approaches.

With regard to the speed characteristic, if at an initial time point the traffic velocity v (x, 0) and the traffic density k (x, 0) were spatially constant with v = V, then the solution would be stationary at constant inflow or outflow. For a given traffic density k, the traffic flow would, according to the model, aim at the velocity V (k) if at an initial time v (x, t) and k (x, t) were spatially constant.

betrachtet, das von den folgenden Parametern abhängt:

v_f

einer Wunschgeschwindigkeit,

L_Fz

einer (effektiven) Fahrzeuglänge und

T_Reakt

einer (effektiven) Reaktionszeit.

A differentiable function V (k; v _f ; L _Fz ; T _react ) is referred to as a "speed characteristic" and depends in macroscopic models on the traffic density and possibly other parameters. As an example, a model

which depends on the following parameters:

v _f

a desired speed,

L _Fz

an (effective) vehicle length and

T _react

an (effective) reaction time.

In order to consider a number of N lanes, k can be replaced by k / N in the equation (3).

Macroscopic models can be used either directly or with the help of filtering methods to reconstruct the traffic flow with specified constraints. For this, the above equations are solved in the context of an initial boundary value problem. In this case, suitable boundary conditions are specified data at the track entrance (traffic flow and speed) and at the end of the track (approximately zero acceleration). Initial values of density and speed can also be specified. Macroscopic solutions for the above equations are clearly based on a discretization of the fields traffic density k (x, t) and velocity v (x, t) on a multi-dimensional grid.

For the intended applications, the flow of traffic z. As can be modeled on highways and on urban roads. The latter have a more complex traffic flow because of dynamically heterogeneous influences and boundary conditions (intersections, traffic lights, turnarounds, constrictions, etc.). As these influences are usually not recorded directly, another source of information is advantageous for a good resolution and meaningfulness of foresight. In particular, historical data in the form of historical speed profiles can be used for this purpose. Instead of determining a speed characteristic using the traffic density k as in the model according to equation (3), the acquired historical data can be directly taken into account.

simulation

In particular, it is proposed to expand known approaches of the microscopic technique in such a way that no typical sequence is required, but a prognosis of the current trajectory of the EGO vehicle takes place taking into account the influences shown in the existing data. Here, a model of the traffic flow (short traffic model) is used in the context of a suitable statistical method for the reconstruction and for the prognosis of the procedure.

For example, in the simulation, data, such. For example, initial data, sample vehicle data, historical data (profiles), and constraints (eg, including traffic volume) are read in and used to perform a simulated traffic flow.

For this purpose, the simulation can generate individual virtual vehicles and model the movement of each individual vehicle. The street itself can z. B. as a grid with support points (eg., With a distance of one meter in the longitudinal direction) are shown. It is also possible to model transverse movements such as lane changes with the aid of a two-dimensional representation.

The particle-lattice structure of the simulation allows for both microscopic phenomena (such as low TTC values) and macroscopic information (such as speed profiles, curvatures of the road, changes in allowable speed) to be taken into account. Thus, the acceleration of each virtual vehicle calculated at a given time depends on the one hand (microscopically) on the distance to the vehicle in front as well as on the speed of the driver's vehicle and the vehicle ahead, etc. On the other hand, as in macroscopic models, the vehicles try to achieve a predetermined, spatially dependent speed characteristic estimated from historical data.

TTC values denote a time until a vehicle theoretically approaches and collides with a preceding vehicle (TTC: Time to Collision) if both travel at an unchanged speed. The TTC value is equal to the distance divided by the difference speed.

Due to the dispersion of driver and vehicle features, there is some degree of blur even with very detailed microscopic modeling.

Therefore, an exact prediction of the individual vehicle accelerations is not possible. However, a realistic statement about the statistical distribution of the processes is possible.

initial conditions

• – Eine maximale Komfortbeschleunigung wird anhand einer Normalverteilung bestimmt (beispielhafte Werte: Mittelwert etwa 4 m/s², Standardabweichung etwa 1 m/s²);
• – Eine Gleichgewichtszeitlücke wird anhand einer logarithmischen Normalverteilung bestimmt mit einem beispielhaften Medianwert von etwa 1,2 s;
• – Eine Mindestzeitlücke des Regelbereichs wird in etwa zu 50% der Gleichgewichtszeitlücke mit einer Zufallskomponente bestimmt.

First, for example, virtual vehicles are generated to which attributes are assigned:

• - A maximum comfort acceleration is determined by means of a normal distribution (exemplary values: mean value about 4 m / s ² , standard deviation about 1 m / s ² );
• An equilibrium time gap is determined on the basis of a logarithmic normal distribution with an exemplary median value of approximately 1.2 s;
• - A minimum time gap of the control range is determined in approximately 50% of the equilibrium time gap with a random component.

In an exemplary scenario, a simulation begins with the receipt of a first signal from a preceding sample vehicle. The sample vehicle has determined a data series (its trajectory) with timestamps, positions, and speeds for a given time interval and sends them to the EGO vehicle. This data is impressed on a virtual vehicle, i. H. for one of the simulation virtual vehicles. This virtual probe vehicle will also be treated differently during the simulation than the other virtual vehicles (for which no data series was obtained). Thus, the initial position and speed of the virtual test vehicle are set.

Next, virtual vehicles are generated downstream of the sample vehicle. For this purpose, a condition "bound" (true or false) is first estimated at the current position on the basis of the historical profile. According to this state estimation, a time gap is determined on the basis of the logarithmic normal distribution, the parameters of this distribution depending on the state estimation. Thus, a position for another virtual vehicle downstream with respect to the sample vehicle z. B. defined according to the SMC method for stochastic simulation in known distributions. The assigned speed for this virtual vehicle is governed by the given historical speed profile (but is preferably at least as great as the speed of the sample vehicle). Such assignment occurs until the downstream end of the route.

The end of the route can z. B. be an intermediate goal or an upstream point where z. B. a turn would be possible.

Subsequently, a time gap is determined in a similar manner upstream of the sample vehicle. The assigned speed upstream of the test vehicle depends here on the minimum of the speed of the sample vehicle and its speed profile. The generation of additional virtual vehicles is generated upstream in the direction of the EGO vehicle (until the beginning of the route is reached).

boundary conditions

Different types of boundary conditions are possible for different applications. For example, the distance may be considered in a range from x = 0 to x = L. For a replica of the traffic on this route, the data would in principle be required at both locations 0 and L, since the vehicles downstream beyond x = L also influence the behavior of the vehicles within the route (for example: traffic jam).

Upstream boundary conditions can be considered as follows: The particle-grid construction allows for immediate representation of vehicles entering the route. They "enter" the grid at a time that can be determined according to traffic demand using a vehicle generator at the edge of the line. For this purpose, time gaps from the model distribution (modified exponential distribution) can be determined at random. After the vehicles (particles of the model) pass the position x = 0, they move according to the model equations.

Downstream conditions can be considered as follows: Auxiliary lines can be used to implement boundary conditions at the end of the simulated segment downstream of the simulated path (x = L). They are propagated from there according to so-called "free boundary conditions", d. that is, at the end of the simulated path at x = L, the zero acceleration (that is, at a constant speed) motion continues.

Use historical speed profiles

For the simulation, historical data, such. B. in the form of a distribution or a profile used. For example, historical data may be grouped according to day, time, weather condition, season, holiday or general events (football final). Also combinations of the above criteria can be used for grouping. Furthermore, historical data can be used according to a LIFO principle (the last recorded data is used first). Furthermore, very old data could be deleted. Also, the historical data itself can be suitably modeled so that a compact knowledge base can be provided by the historical data. Furthermore, it is possible that different data may be collected and taken into account as historical data for future simulations.

• – Eine historische Geschwindigkeit in der Gitterzelle.
• – Ein boolescher Wert, der angibt, ob ein Verkehr in der Gitterzelle gebunden ist.
• – Ein boolescher Wert, der angibt, ob es sich bei der Gitterzelle um ein stromabwärtiges Streckenende handelt.
• – Ein boolescher Wert, der angibt, ob sich in der Gitterzelle ein Stau befindet.
• – Eine Wartezeit in der Gitterzelle bei existierendem Stau
• – Eine kumulierte Reisezeit bis zu der Gitterzelle.

For example, for each grid cell of the simulation, at least a portion of the following data may be provided:

• - A historical speed in the grid cell.
• - A Boolean value that indicates whether traffic is bound in the grid cell.
• A Boolean value that indicates whether the grid cell is a downstream end of the line.
• - A Boolean value that indicates whether there is a jam in the grid cell.
• - A waiting time in the grid cell with existing traffic jam
• - An accumulated travel time to the grid cell.

The assignment of bound or unbound traffic can take place by means of a hysteresis: If downstream of a critical value of the speed is fallen below, bound downstream traffic flow is assumed downstream until at one point a second (higher) value of the speed is exceeded. The unbound traffic is again continued upstream until the lower value is again fallen below. This creates a stable representation with hysteresis.

The bound traffic is a traffic in which the driver or the EGO vehicle can not choose the speed itself, but follows a vehicle in front. The EGO vehicle is thus part of a linked chain of road users.

acceleration request

xc

Position des EGO-Fahrzeugs;

xp

Position des vorausfahrenden Fahrzeugs;

d_gap

Entfernung zwischen EGO-Fahrzeug und vorausfahrendem Fahrzeug;

v_carnow

Geschwindigkeit des EGO-Fahrzeugs;

v_target

Geschwindigkeit des Zielfahrzeugs (hier des vorausfahrenden Fahrzeugs).

An acceleration request of the EGO vehicle for a unit time (discrete system based on time units as the smallest unit considered) can be calculated based on the following values:

xc

Position of the EGO vehicle;

xp

Position of the preceding vehicle;

d _gap

Distance between the EGO vehicle and the vehicle ahead;

v _carnow

Speed of the EGO vehicle;

v _target

Speed of the target vehicle (here the vehicle ahead).

The position here preferably refers to the front of the vehicle. The difference d _gap between the vehicles is given here d _gap = xp - xc.

• – Ein Modus ”ungültige Werte” wird bestimmt, falls die Geschwindigkeit negativ ist oder der Abstand null beträgt. Dies führt z. B. zu einem Abbruch bzw. zu einem Neustart der Simulation.
• – Ein Modus ”Panik” erfordert einen Eingriff des Fahrers (keine Regelung mehr möglich). Der Modus ”Panik” tritt ein, wenn Mindestwerte für die Regelung unterschritten werden. Ggf. kann eine Notbremsung automatisch eingeleitet werden.
• – Ein Modus ”Gebunden: Anhalten” tritt ein, falls das vorausfahrende Fahrzeug bereits angehalten hat (z. B. an einer Ampel, etc.). In diesem Fall kann ein automatisiertes Anhalten (Abbremsen) mittels der Regelung veranlasst werden.
• – Ein Modus ”Gebunden: Fahrzeugfolgemodus” tritt ein, falls die Werte im Regelbereich sind. Der Regelalgorithmus kann entsprechend veranlassen, dass das EGO-Fahrzeug dem vorausfahrenden Fahrzeug folgt.
• – Ein Modus ”Gestoppt, Vordermann fährt los” tritt ein, wenn das EGO-Fahrzeug steht und der Vordermann die Fahrt aufgenommen hat. Der Regelalgorithmus kann veranlassen, dass sich das EGO-Fahrzeug dem Vordermann annähert.
• – Ein Modus ”Ungebunden” tritt ein, wenn die Zeitlücke mindestens zweimal der Gleichgewichtszeitlücke ist UND eine ausreichende Zeit bis zu einer möglichen Kollision (TTC-Wert) größer als eine minimale Zeit bis zu einer möglichen Kollision (z. B. in Höhe von 5 s) ist. In Folge kann mittels des Regelalgorithmus eine Relaxation in Richtung des Geschwindigkeitsprofils erfolgen, d. h. die in dem Geschwindigkeitsprofil hinterlegte Geschwindigkeit kann eingestellt werden. Entsprechend kann eine Beschleunigung erfolgen gemäß tdif/tgrid, wobei vdif eine Differenzgeschwindigkeit des EGO-Fahrzeugs zum Geschwindigkeitsprofil an der aktuellen Stelle und tgrid eine Zeitdauer (z. B. 3 s) in der Gitterzelle angibt.
• – Ein Modus ”Ungebunden” (bei einer Option ”freeflag”) tritt dann ein, falls die Zeitlücke zum vorausfahrenden Fahrzeug größer als eine vorgegebene Schranke (z. B. 6 s) ist UND eine ausreichende Zeit bis zu einer möglichen Kollision (TTC-Wert) größer als eine minimale Zeit bis zu einer möglichen Kollision (z. B. in Höhe von 5 s) ist. In diesem Fall kann die Geschwindigkeit des EGO-Fahrzeugs auf die zulässige Geschwindigkeit geregelt werden.

In the acceleration _model , a so-called mode is determined with the aid of the values d _gap , v _carnow , v _target and possibly further parameters:

• - A mode "invalid values" is determined if the speed is negative or the distance is zero. This leads z. B. to abort or to restart the simulation.
• - A "panic" mode requires driver intervention (no regulation possible). The "panic" mode occurs when minimum values for regulation are undershot. Possibly. Emergency braking can be initiated automatically.
• - A "Tied: Stop" mode occurs if the vehicle in front has already stopped (eg at a traffic light, etc.). In this case, an automated stop (deceleration) can be caused by the scheme.
• - A mode "Bound: Vehicle following mode" occurs if the values are within the control range. The control algorithm may accordingly cause the EGO vehicle to follow the preceding vehicle.
• - A mode "Stopped, front man drives off" occurs when the EGO vehicle is stationary and the front man has taken the drive. The control algorithm may cause the EGO vehicle to approach the car in front.
• An "Unbound" mode occurs when the time gap is at least twice the equilibrium time gap AND a sufficient time to a possible collision (TTC value) greater than a minimum time to a possible collision (eg, at the level of 5 s is. In consequence, by means of the control algorithm a relaxation in the direction of the velocity profile can take place, ie the speed stored in the velocity profile can be set. Accordingly, acceleration may occur according to tdif / tgrid, where vdif indicates a differential velocity of the EGO vehicle to the velocity profile at the current location and tgrid a period of time (eg, 3 seconds) in the grid cell.
• A "Unbound" mode (with a "freeflag" option) then occurs if the time gap to the vehicle in front is greater than a predetermined barrier (eg 6 s) AND there is sufficient time until a possible collision (TTC) occurs. Value) is greater than a minimum time to a possible collision (eg, at the rate of 5 seconds). In this case, the speed of the EGO vehicle can be controlled to the permissible speed.

control algorithm

In the vehicle following mode, it is assumed that each EGO vehicle aims for a state of equilibrium, which may need to be readjusted by external influences. In equilibrium, each EGO vehicle would have the desired distance and speed as the preceding vehicle. By way of example, the control according to a damped harmonic oscillator can take place with an equilibrium time gap (length of the vehicle is taken into account), a spring constant, and a damping (subcritical to critical).

In addition to the equilibrium time gap each vehicle can be assigned a minimum time gap, which defines the maximum deflection and the lower limit of the control range. The effective spring constant results from the maximum comfort acceleration or the maximum rash.

Unpredictable, sudden events may cause the dynamic controlled variables to occasionally be outside the control range. In this case, for example, if the minimum time gap is undershot, a panic model can take control of the desired acceleration. For example, in the panic model, a maximum deceleration associated with the driver is assumed.

Description of the simulation runs

The equations of motion can be solved stepwise in given time units (also referred to as time slices Δt, eg of 0.1 s). The new position is calculated from the speed multiplied by δt. The new speed is calculated from the acceleration specification multiplied by δt. The acceleration default normally equals the desired acceleration, whereby it is preferably ensured that no limit values are undershot and / or exceeded.

• – Das erste (stromabwärtige) Fahrzeug in der Simulation fährt mit konstanter Geschwindigkeit weiter, bis es die Endstrecke erreicht und das Gitter verlässt.
• – Ein Probe-Fahrzeug erhält diejenige Trajektorie, die sich durch die von dem Probe-Fahrzeug übermittelten Daten ergibt (solange das Signal von dem Probe-Fahrzeug vorhanden ist). Liegt für das Probe-Fahrzeug kein C2C-Signal mehr vor, so wird es wie jedes andere virtuelle Fahrzeug simuliert.
• – An der ”Einfahrt” zur Strecke werden Fahrzeuge nach einem Nachfragemodell (bei einer vorgegebener Verkehrsstärke) erzeugt. Ihnen werden entsprechend dem historischen Profil anfängliche Geschwindigkeiten zugeordnet. Gleichzeitig wird darauf geachtet, dass kein Stau am Anfang der Strecke entsteht.

The following exceptional situations can be treated separately:

• The first (downstream) vehicle in the simulation continues to travel at a constant speed until it reaches the final distance and leaves the grid.
• A sample vehicle receives the trajectory resulting from the data transmitted by the sample vehicle (as long as the signal from the sample vehicle is present). If the C2C signal is no longer present for the sample vehicle, it will be simulated like any other virtual vehicle.
• At the "entrance" to the track, vehicles are generated according to a demand model (at a given traffic volume). You will be assigned initial speeds based on historical profile. At the same time, care is taken that no traffic jams occur at the beginning of the route.

The simulation runs until all vehicles of the vehicle memory have left the track or until the maximum simulation time (eg 300 seconds) is exceeded.

All trajectories of the simulated vehicles (with time stamp) as well as the identification of the test vehicle are preferably stored.

Concept for traffic light data and queues

In principle, it is possible to consider queues without speed, which were stored in the historical speed profile, in the simulation.

It should be noted that it is unrealistic to assume that a queue experienced by a single historic vehicle is constantly being experienced by all other vehicles in a simulation. This would sooner or later lead to a virtual standstill. Rather, it is realistic, from the totality of historical data, a model for the temporal probability distribution of all potential causes of queues (traffic lights and intersections) z. B. to develop in the backend.

Forecast of the trajectory: quality function

To evaluate both historical profiles and individual trajectories from the simulation runs, a merit function is used. For this purpose, an error measure for the predicted travel time during the last X meters (eg with X = 100) can be determined. The error measure is calculated using the squared value of the difference between the estimated travel time minus the recorded travel time, divided by an estimate of the expected variance of that size. The quality is then (analogous to the Gaussian function) an exponential error measure related to this squared value.

For example, a quality function for determining the expected variance may be determined as follows: σ / μ ≈ 0.15 · (480 / T) ^1/2 .

T

die Verkehrsgeschwindigkeit in einer Zelle (z. B. in Sekunden),

σ

die Standardabweichung und

μ

den Mittelwert,

wobei in diesem Fall μ = T gilt.Denote this

T

the traffic speed in a cell (eg in seconds),

σ

the standard deviation and

μ

the mean,

in which case μ = T.

Optionally, a speed difference (predicted versus measured) divided by the variance of the speed can also be determined as an error measure.

Selection of a historical profile

Even before a simulation run, the previous trajectory of the EGO vehicle can be compared with the existing historical profiles. From this, a fitness value can be assigned to each existing profile with the aid of the quality function.

For this purpose, the information "cumulative travel time to the grid cell" of the historical data can be used. Historical profiles are randomly chosen for simulation with a probability according to their quality.

If no historical profile is suitable at all, a surrogate profile can be generated from the previous course of the travel of the EGO vehicle, which essentially uses the permissible speed as a profile.

Identification of probable EGO vehicle trajectories

Based on the simulation, probable trajectories of the EGO vehicle can be determined by weighting them according to their fitness ("fitness"), selecting those of sufficient quality (when a given threshold is reached or exceeded). The time series of the positions, speeds, and accelerations of the candidates can be stored in a database of "predictions". Each prediction is assigned a weight according to the quality, and the resulting database is made available to all applications.

• (1) Synchronisierung der Simulation zum Zeitpunkt des letzten Signals von dem Probe-Fahrzeug (wegen Verzögerungen der Übertragung (Latenzen) ist dies nicht zwangsläufig der aktuelle Zeitpunkt).
• (2) Lokalisierung des EGO-Fahrzeugs (Zuordnung auf dem Simulationsgitter) zu diesem Zeitpunkt.
• (3) Lokalisierung der Position des dem EGO-Fahrzeug nächsten virtuellen Fahrzeugs und aller weiteren virtuellen Fahrzeuge des Simulationslaufs, die sich innerhalb einer vorgegebenen Toleranz zu dem EGO-Fahrzeug befinden.
• (4) Konstruktion der Zeitreihen: a) Alle vorhandenen simulierten virtuellen Zeitpunkte werden in einen Puffer für die Prognosezeitreihe (synchronisiert mit dem EGO-Fahrzeug) geschrieben. b) Falls die EGO-Fahrzeug-Trajektorie auch Zeitscheiben vor dem ersten Simulationszeitpunkt des jeweiligen virtuellen Fahrzeugs enthält, werden die fehlenden Werte der virtuellen Fahrzeuge mit dem für die Simulation verwendeten historischen Profil aufgefüllt. c) Falls die simulierte Zeit nicht für den Prognosehorizont ausreicht, wird der virtuelle EGO-Fahrzeug-Kandidat verworfen.
• (5) Der Fitnesswert aller virtuellen EGO-Fahrzeug-Kandidaten wird bestimmt; derjenige EGO-Fahrzeug-Kandidat wird verworfen, dessen Fitness-Wert einen vorgegebenen Wert nicht erreicht.
• (6) Es erfolgt ein Abbruch bei einer Überschreitung einer vorgegebenen Rechenzeit oder bei Erreichung einer ausreichenden Zahl von geeigneten EGO-Fahrzeug-Kandidaten.
• (7) Für die geeigneten EGO-Fahrzeug-Kandidaten werden prognostizierte Verläufe in der Datenbank gespeichert.

An example algorithm has the following sequence:

• (1) Synchronizing the simulation at the time of the last signal from the sample vehicle (because of delays in transmission (latencies) this is not necessarily the current time).
• (2) Localization of the EGO vehicle (assignment on the simulation grid) at this time.
• (3) Locate the position of the virtual vehicle next to the EGO vehicle and all other virtual vehicles of the simulation run that are within a given tolerance to the EGO vehicle.
• (4) Construction of the time series: a) All existing simulated virtual times are written to a buffer for the forecast time series (synchronized with the EGO vehicle). b) If the EGO vehicle trajectory also contains time slices before the first simulation time of the respective virtual vehicle, the missing values of the virtual vehicles are filled up with the historical profile used for the simulation. c) If the simulated time is not sufficient for the forecast horizon, the EGO virtual vehicle candidate is discarded.
• (5) The fitness value of all virtual EGO vehicle candidates is determined; that EGO vehicle candidate is discarded whose fitness value does not reach a predetermined value.
• (6) An abort occurs if a given calculation time is exceeded or if a sufficient number of suitable EGO vehicle candidates are reached.
• (7) For the appropriate EGO vehicle candidates, predicted gradients are stored in the database.

Although the invention has been illustrated and explained in detail by the embodiments, the invention is not limited thereto. Accordingly, variations may be derived therefrom by those skilled in the art without departing from the scope of the invention.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited non-patent literature

• http://en.wikipedia.org/wiki/Floating_Car_Data [0002]
• http://en.wikipedia.org/wiki/Sequenzielle_Monte-Carlo-Method [0003]
• http://en.wikipedia.org/wiki/VISSIM [0112]

Claims (20)

Method for forecasting a traffic situation for a vehicle, - Is determined on the basis of a current state of the vehicle and based on historical data at least one possible future for the traffic situation of the vehicle. The method of claim 1, wherein the current state of the vehicle comprises at least one of the following parameters: A position of the vehicle; A speed of the vehicle; - a route to be traveled; - a direction; - a target. Method according to one of the preceding claims, in which the historical data comprise at least one of the following parameters: - a traffic density; - an average speed; - a speed gradient; - a travel time; A traffic cycle, in particular a traffic light cycle; - a traffic volume; - a traffic situation. Method according to one of the preceding claims, in which the historical data are aggregated by different vehicles. Method according to one of the preceding claims, in which the historical data is transmitted to the vehicle via a mobile communication interface. Method according to one of the preceding claims, in which the at least one possible future of the vehicle is also determined by means of at least one of the following information: An information of at least one other vehicle; An information from a traffic infrastructure component; An information from a traffic center; A traffic information; An information from a digital map; An information of a driver assistance system; - Online data. Method according to one of the preceding claims, in which the at least one possible future of the vehicle is also determined by means of a time window, the time frame being used to record the condition of the vehicle for a predefined period of time. Method according to one of the preceding claims, in which the possible future for the traffic situation of the vehicle comprises at least one of the following possibilities: A position of a jam end; - a future course of speed; A future course of acceleration; - a waiting time and traffic jam length in front of a traffic facility; A green wave or a speed strategy approaching a traffic facility; - A speed recommendation for safe or efficient driving. Method according to one of the preceding claims, in which the at least one possible future traffic situation for a short-term future of the vehicle is determined by means of a simulation. Method according to Claim 9, in which the simulation is carried out by means of a microscopic model and / or a macroscopic model for a traffic flow. The method of claim 10, wherein the microscopic model comprises rules for simulating a movement of a vehicle taking into account the movements of the remaining vehicles. Method according to one of claims 10 or 11, wherein based on the macroscopic model of the traffic flow is modeled as a kind of liquid with predetermined initial conditions and / or boundary conditions. Method according to one of claims 9 to 12, wherein the simulation of a future traffic flow in the direction of travel of the vehicle is determined. Method according to one of claims 9 to 13, wherein in the simulation of data from at least one other vehicle mapped to a virtual vehicle and generated according to the historical data and other possible data virtual vehicles for mapping a corresponding traffic flow and determines their impact on the vehicle become. Method according to one of the preceding claims, in which a plurality of future traffic situations for the vehicle are determined and that traffic situation is selected which is determined by means of a quality function as the most suitable traffic situation for the vehicle. The method of claim 15, wherein based on the most appropriate traffic situation, a predetermined action is performed. The method of claim 16, wherein the predetermined action comprises at least one of the following: - an optical feedback; - a haptic feedback; - an intervention in the control of the vehicle; - establishing a communication connection; - Submitting a message. Method according to one of the preceding claims, in which the possible future for the traffic situation of the vehicle comprises a trajectory and / or a speed profile of the vehicle. Device for providing a forecast of a traffic situation for a vehicle with a processing unit that is set up such that - Based on a current state of the vehicle and based on historical data at least one possible future for the traffic situation of the vehicle can be determined. Vehicle with a device according to claim 19.

Images