EP4405928A1 - Verfahren und vorrichtungen für automatische, kooperative manöver - Google Patents
Verfahren und vorrichtungen für automatische, kooperative manöverInfo
- Publication number
- EP4405928A1 EP4405928A1 EP22790491.9A EP22790491A EP4405928A1 EP 4405928 A1 EP4405928 A1 EP 4405928A1 EP 22790491 A EP22790491 A EP 22790491A EP 4405928 A1 EP4405928 A1 EP 4405928A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- maneuver
- cooperation
- maneuvers
- machines
- machine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/161—Decentralised systems, e.g. inter-vehicle communication
- G08G1/163—Decentralised systems, e.g. inter-vehicle communication involving continuous checking
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/167—Driving aids for lane monitoring, lane changing, e.g. blind spot detection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/22—Platooning, i.e. convoy of communicating vehicles
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/40—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P]
- H04W4/44—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P] for communication between vehicles and infrastructures, e.g. vehicle-to-cloud [V2C] or vehicle-to-home [V2H]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/40—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P]
- H04W4/46—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P] for vehicle-to-vehicle communication [V2V]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
Definitions
- the present disclosure relates to methods and apparatus for providing cooperative maneuvers involving a plurality of machines, particularly vehicles.
- Cooperative maneuvers refer to maneuvers involving more than one machine in a shared environment.
- a suitable selection of a cooperation maneuver can, for example, optimally control a traffic flow and/or increase the safety of a plurality of machines that share a working space.
- There are different approaches and protocols for performing cooperative maneuvers There are no solutions, especially for automatically generated cooperation maneuvers. Improvements in these areas are desirable.
- One object is to provide methods and devices for collaborative maneuvers involving multiple machines in a common workspace.
- Some embodiments of the disclosure solve the specific problem of generating and providing a cooperative maneuver to overtake one or more vehicles. This can be done using machine learning to provide safe and performant maneuvers under real-time conditions. For this purpose, a classifier is first trained for a predetermined maneuver success. This separates successful from unsuccessful maneuver candidates. In addition, a regression model is trained on an extensive data set of randomized baseline situations. The regression model can be used to distinguish between good and bad cooperation maneuvers within the set of successful cooperation maneuvers and finally to identify an optimal cooperation maneuver. By means of appropriately weighted quality functions, an optimization for different goals such as avoiding jerks or the length of the overtaking maneuver can be achieved.
- a first aspect relates to a method comprising the steps:
- a machine can be a vehicle, in particular a passenger car or a truck.
- a machine can also be a controlled flying object, e.g. an airplane, an airship, a balloon or a drone.
- a machine can be a machine tool, in particular a robot, e.g. an industrial robot with serial and/or parallel kinematics.
- a machine can also be a construction machine, agricultural machine and/or a (driverless) transport system.
- the method can also be carried out for a representation of a machine instead of for a machine. This will be described in more detail later. This can be done in particular for simulation purposes, so that technical knowledge can be gained for a number of cooperating machines that are to be physically implemented.
- a cooperative maneuver includes maneuvers for at least two machines involved in the cooperative maneuver.
- the maneuvers should be specified in such a way that the machines can successfully complete the cooperative maneuver.
- a cooperative maneuver can, in principle, contain maneuvers which—if they were executed—can lead to damage or impairment of one machine by the other machine. These can, for example, be differentiated from unsuccessful and successful cooperation maneuvers by means of a corresponding evaluation.
- Obtaining shared information about the surroundings can in particular include measuring and/or simulating. Additionally or alternatively, obtaining shared information about the area can include receiving and/or fetching this information.
- An environment from which the information about the environment originates can include, for example, a shared workspace for a number of robots. Additionally or alternatively, an environment may include a road on which multiple vehicles are moving. An environment can also be a railway line. An environment may additionally or alternatively include an airspace.
- a predetermined value of the quality criterion can in particular include a predetermined value range, e.g. overtaking distances of less than 100 meters.
- a predetermined value can include an extreme value, e.g., a minimum value or a maximum value.
- the provision can include communicating and/or storing the information on the selected cooperation maneuver.
- the selected cooperation maneuver can be stored in a shared memory space for the machines involved, so that other machines can fetch the information.
- the provisioning can be done for the machine that performed the assessment. Then the cooperation maneuver or at least the maneuver provided for the machine can be provided in order to be implemented in particular by the machine.
- Providing may include communication to one machine or multiple machines. Especially if that machine didn't do the evaluation. If the method is executed on an external device (for example on an infrastructure device), the selected cooperation maneuver can be made available in particular to all machines that the cooperation maneuver includes.
- an external device for example on an infrastructure device
- the method can be carried out on one of the machines involved in the maneuver, for example on a host vehicle.
- the host vehicle may be the vehicle that is to overtake other vehicles.
- the method can also be executed on an infrastructure device or on a machine that is not involved in the cooperation maneuver.
- An embodiment of the first aspect relates to a method, in which the environmental information is at least partially detected by sensors, in particular by one of the two machines.
- the environmental information can also come from a third machine. Additionally or alternatively, the environmental information can also come from a simulation or a modeling. In addition or as an alternative, environmental information can also come from the environment of both machines and can be detected by these in particular using sensors. Additionally or alternatively, environmental information can come from an observer system, for example from an infrastructure information system, especially when the machines are vehicles.
- a piece of environmental information can include, in particular, the number of machines involved in the cooperative maneuver and their states (position, speed, etc.).
- Information about the surroundings can in particular include information about a traffic situation, for example about a street and/or an intersection.
- An embodiment of the first aspect relates to a method wherein a maneuver for a machine includes multiple sub-maneuvers.
- a maneuver can in particular include sub-maneuvers which depend on the type of maneuver. For example, if the maneuver relates to a vehicle overtaking maneuver, then a maneuver may include sub-maneuvers related to accelerating, changing lanes, decelerating, and/or maintaining a current machine state.
- a maneuver can relate, for example, to a longitudinal and transverse change in direction, a change in position and/or a change in altitude.
- the maneuvers of a cooperation maneuver can in particular all include the same number and/or the same type of sub-maneuvers.
- An embodiment of the first aspect relates to a method, wherein the maneuvers of the individual machines are of different lengths in terms of time and/or space.
- the length of the maneuver of a cooperation maneuver can in particular be determined in terms of space and/or time. If the maneuvers are timed, the maneuvers of the individual machines affect a different length of time. Additionally or alternatively, the length can be determined spatially. Then, for example, one machine in a cooperation maneuver can cover a longer distance than another machine. For example, during an overtaking maneuver, the overtaking vehicle generally covers a longer distance than a vehicle that is being overtaken.
- An embodiment of the first aspect relates to a method, in which the evaluation only takes place for some of the maneuvers comprised by the cooperation maneuver.
- an evaluation can be directed to only one maneuver of a cooperative maneuver. This can be done in particular when the cooperative maneuver involves only two machines. However, even with more than two machines, a single, in particular predetermined, maneuver can be used for the evaluation, e.g. the maneuver for an overtaking vehicle. Alternatively, the evaluation can take place on the basis of several, in particular all, maneuvers of a cooperation maneuver.
- An embodiment of the first aspect relates to a method, in which the evaluation is carried out using a trained algorithm, in particular using a neural network.
- a trained algorithm can in particular include machine learning. Training can take place in particular on the basis of supervised learning or on the basis of reinforcement learning.
- the algorithm can be based on previously machine-generated maneuvers and/or cooperation maneuvers. Additionally or alternatively, training can take place through a maneuver that has currently taken place and/or a cooperation maneuver. The same can apply to testing a trained algorithm.
- One embodiment of the first aspect relates to a method, in which the evaluation is carried out using a binary classification, in which successful cooperation maneuvers in particular are distinguished from unsuccessful cooperation maneuvers.
- the binary classification can be carried out in particular by a trained algorithm.
- the trained algorithm can be based in particular on machine learning, for example the binary classification can be carried out by an appropriately trained neural network.
- the binary classification makes it possible in particular to differentiate between a successful cooperation maneuver and an unsuccessful cooperation maneuver, so that only successful cooperation maneuvers have to be processed further. A processing time of the method can be shortened as a result. This is particularly useful when the method is run under real-time conditions.
- a classification can also differentiate whether a safety distance is guaranteed, whether predetermined rules, for example traffic rules, are observed and/or whether a cooperative maneuver is efficiently composed of sub-maneuvers, for example such that no redundant maneuvers, eg duplicate maneuvers, are included.
- a classification can be carried out as to whether dependent sub-manoeuvres have been successfully resolved, it being possible for the dependent sub-manoeuvres to be defined in particular between different maneuvers of a cooperation maneuver.
- An embodiment of the first aspect relates to a method, in which the evaluation takes place by means of a regression, by means of which a quality of the cooperation maneuver is determined in particular.
- the regression can be carried out in particular by a trained algorithm.
- the trained algorithm can be based in particular on machine learning, for example the binary classification can be carried out by an appropriately trained neural network.
- the regression can take place in particular according to the binary classification.
- the regression can only be performed for successful maneuvers.
- no cooperation maneuvers are evaluated with regard to their quality, which are not implemented anyway because they are not successful.
- a processing time of the method can be shortened as a result. This is particularly useful if the method is to be generated and provided under real-time conditions, e.g. for an overtaking maneuver.
- An embodiment of the first aspect relates to a method, in which the quality criterion is composed of a number of evaluation variables and in which the individual evaluation variables are weighted in particular.
- An evaluation variable can be a position-based variable, for example a position, a distance covered, a speed, an acceleration and/or a jerk. For example, an overtaking maneuver can be evaluated according to how much distance is required for it. This is because, particularly in the case of oncoming traffic, the distance in the lane can be a safety-relevant evaluation variable for oncoming vehicles.
- An evaluation variable can in particular be the duration of a cooperation maneuver. According to the above justification, a duration can also be a safety-relevant evaluation variable.
- An evaluation variable composed of several evaluation variables can be composed, for example, of a longest distance of a maneuver of a cooperation maneuver and of a jerk that is exerted on the machine by the maneuver.
- these assessment variables can be weighted. Depending on the selection of the weightings, different types of cooperation maneuvers can be specifically preferred.
- One embodiment of the first aspect relates to a method, in which the cooperative maneuver is provided to at least one additional machine.
- Area information maneuvers includes. This can be the case, for example, if some machines have a high probability of not affecting the cooperative maneuver.
- the information about the cooperation maneuver can nevertheless be communicated to these machines. This can be particularly advantageous when these machines have additional information that was not included in the environmental information for the method, but which is nonetheless relevant to the cooperative maneuver. Then these machines can send the additional information to the machines of the cooperative maneuver and the cooperative maneuver can be adjusted accordingly if necessary.
- the provision can take place by communicating with one or more machines and/or by storing the information on a shared memory, for example a server, from which the machines can obtain the information.
- An embodiment of the first aspect relates to a method, wherein one of the two machines is a host machine and wherein the provisioning takes place from a host machine to at least the second machine.
- the host machine may provide the cooperative maneuver to all machines of the cooperative maneuver. Additionally or alternatively, the cooperation maneuver can also be provided by the host machine for machines not included in the cooperation maneuver, in particular if these could influence the cooperation maneuver.
- An embodiment of the first aspect relates to a method, comprising the step:
- a confirmation can be made by all machines for which the selected cooperation maneuver is provided.
- a confirmation from all machines involved in the cooperation maneuver can be a prerequisite for the cooperation maneuver to be able to be carried out.
- a confirmation can in particular include an approval or a rejection of the cooperation maneuver.
- a refusal can in particular include a counter-proposal for the cooperation maneuver or at least for the maneuver of the refusing machine. Additionally or alternatively, a refusal can include information on how the cooperation maneuver, in particular the maneuver of the refusing machine, must be adapted so that this machine or for this machine can give its consent.
- a second aspect relates to a machine set up to:
- Embodiments of this aspect can be designed analogously to the first aspect.
- An embodiment of the second aspect relates to a machine, wherein the device is an infrastructure unit.
- the device may be a machine, a host machine, involved in the cooperative maneuver.
- a third aspect relates to a method comprising the steps:
- FIG. 4 shows a method in pseudocode according to an embodiment of FIG.
- a corresponding device for carrying out or producing the method, or for a corresponding system which comprises one or more devices, and vice versa.
- a corresponding device may include a feature to perform the method step described, even if that feature is not explicitly described or illustrated.
- a corresponding method can include a step that performs the described functionality or can be used to produce a corresponding structure, even if such steps are not explicitly described or illustrated are.
- a system can also be provided with corresponding device features or with features in order to carry out a specific method step.
- the host vehicle (HV) in the present exemplary embodiment is the vehicle approaching a slower remote vehicle (RV) in the same lane (RV1) and is also the one proposing an overtaking maneuver as indicated by the solid arrows displayed. It can possibly take other vehicles into account, e.g. B. a vehicle (RV2) in front of the slower vehicle (RV1) or a vehicle in a different lane (RV3). All vehicles periodically exchange their status (position, speed, acceleration) by sending Cooperative Awareness Messages (CAMs), which form the basis of the environment model available to each vehicle. Using these CAMs, the vehicles can classify distances and speeds relative to other vehicles and, if necessary, trigger a cooperative maneuver.
- CAMs Cooperative Awareness Messages
- the problem to be solved for the HV is to propose and perform a cooperative overtaking maneuver involving RV1 and possibly other vehicles present. This is to avoid collisions and enable a smooth and efficient flow of traffic.
- Supervised learning allows data collection to be separated from assessment.
- Various metrics for merit functions can be implemented using the quality function for training is changed.
- step 210 a sufficiently large and diverse set of training data is generated. For this purpose, random scenarios are created as described later. These define the number of vehicles present as well as their departure lane, initial position and speed. Then, in step 220, the HV proposes a random cooperative maneuver in each scenario, which is then simulated by all participants. Reasonable restrictions are applied, as described later. Finally, success is evaluated and a merit function value determined for each of these commanded maneuvers.
- a binary classifier for predicting the success of a maneuver and a regression model for assessing the quality of a maneuver are trained with this data within the framework of supervised learning.
- the trained models can be applied, for example by simulating new scenarios or by trying out the models in real life.
- the HV proposes the best of several random cooperative maneuvers to execute.
- the safety of a random cooperative maneuver is evaluated using the binary classifier. If the cooperative maneuver is judged to be safe, the regression model is used to assess the quality of the maneuver.
- computing time can be saved as a result. This is because the regression, which requires more computing time, is only carried out for the successful, for example the safe, cooperation maneuvers.
- the best cooperative maneuver is selected and communicated to surrounding vehicles via a wireless communication interface.
- the vehicles then jointly execute the selected cooperation maneuver.
- the performance of the models and the quality of the proposed cooperative maneuvers can be assessed.
- a scenario generator can produce the basis of raw data for learning.
- One goal is to cover the majority of scenarios that can occur in the desired type of cooperative maneuver.
- a scenario generator is a suitable tool to achieve the completeness of the data set so that it represents all conceivable scenarios if possible.
- the simulated scenarios can cover similar issues associated with real data, such as: B. Sensor inaccuracies or noise due to inadequacies in communication technologies. In particular, a non-deterministic channel fading can be added, so that entire messages in particular can be lost. Coupled with the 100ms transmission interval for CAMs, this means that each vehicle is basing its environment model on potentially stale information. This makes the scenarios more realistic.
- the parameters can be varied as shown in Table 1 (Fig. 5). The ranges given reflect a typical situation for an overtaking maneuver on motorways. The actual value for a given random scenario can be found from a uniform distribution:
- x U(x_max, x_min), where x_max represents a maximum value and x_min a minimum value of the distribution.
- the uniform distribution can be discrete.
- the uniform distribution can in particular be continuous.
- v_hv U(v_l_max, v_l_max + v_diff)
- v_l_max is a specified maximum speed for vehicles to be overtaken in lane I.
- n_rv vehicles to be overtaken are generated. Your lane will be randomly selected, with the first vehicle having an equal chance of being placed in the middle or right lane.
- the speed for the vehicle k to be overtaken in lane I can then be determined from:
- v_l_max represents a maximum speed
- v_l_min represents a minimum speed of the vehicle.
- the following conditions can be met to obtain realistic scenarios.
- v_hv U(v_rv,k* + v_diff, v_l,max + v_diff), in case this is required.
- a random RV k* can be chosen on the two right lanes, then v_hv can be recalculated according to formula (4) and in particular the HV can be placed on the same lane as the RV k*.
- a sensible positioning of the vehicles to be overtaken along the road can take place. If the closest RV before the HV (RV1 in Fig. 1) is closer than s_hvrv, a new position within the interval [s_hvrv_min; s_hvrv_max ] can be determined. The interval can represent the distance boundary between the HV and the RV. It can then be ensured that the distance to the preceding RV is at least s_dist for all other RVs on the same lane (RV2 in Fig.
- controlled cooperation maneuvers can be simulated.
- the HV triggers a cooperative maneuver based either on a distance, for example the geometric distance to a slower vehicle in front. Additionally or alternatively, triggering can also occur based on other parameters, such as a time to a collision or a composite parameter that includes a distance, a speed difference and/or a time to a collision. Additionally or alternatively, triggering can also be based on external trigger information.
- the maneuver suggestions can be randomly generated. Exemplary parameter ranges are given in Table 2 ( Figure 6) and are explained below.
- n_hvc maneuver containers consisting of n_hvc maneuver containers.
- the last partial maneuver can in particular be a lane change back to the right lane and/or to an exit lane.
- All n_hvc - 1 maneuver containers in front of it can be filled with random maneuver types - "lane change” (cl), "speed change” (cs) or "maintain current driving status” (cn) - whereby in particular consecutive sub-maneuvers of the same type are avoided.
- the duration of all maneuvers (t_cl , t_cs , t_cn ) can be randomized within the limits given in Table 2.
- the n_rvc,k sub-maneuvers can be generated for all existing vehicles to be overtaken.
- the maximum number of containers for remote vehicles, n_rvc can be chosen lower than for the host vehicle, in particular if it is assumed that vehicles to be overtaken perform fewer actions than the overtaking vehicle during an overtaking process.
- the general logic for sub-maneuver generation may be the same as for the HV, additional restrictions may apply.
- Vehicles to be overtaken may, for example, only be allowed to change lanes to the right, in particular so as not to endanger the overtaking vehicle.
- the target speed v_cs can be defined on the basis of the current lane of the vehicle according to formula (3).
- the host vehicle HV can then evaluate the resulting cooperative maneuver composed of the actions of the HV and the RVs. In particular, it can check whether at least one vehicle will overtake and/or whether vehicles will have an accident and/or perform illegal actions (e.g. fall below one or more safety distances). For example, if overtaking maneuvers are legally only allowed on the left side (as in Germany), cooperative maneuvers in which the HV overtakes a vehicle on the right side can be discarded.
- the HV in the training phase, can generate random maneuvers until the maneuver can be successfully validated. Then it can provide a final proposal to negotiate with the vehicles to be overtaken in this cooperative maneuver. Negotiation can in particular take place in such a way that the proposed cooperation maneuver is approved or rejected by all vehicles that are covered by the cooperation maneuver, ie in particular by all vehicles to be overtaken.
- the set of all executed cooperation maneuvers can be evaluated and later used for model training.
- Collisions generally mean that at least two machines participating in a cooperative maneuver have the same position at the same time. Additionally or alternatively, a collision can be defined by the spatial and temporal safety distances between the machines and/or surrounding structures relevant in the cooperation maneuver being violated. If a cooperative overtaking maneuver is to be evaluated, another criterion can be whether an overtaking maneuver has taken place or not. A successful cooperative maneuver can therefore occur as a result of a binary classifier, e.g. as a collision-free overtaking maneuver.
- Another quality criterion can be the jerk, for example as the third derivative of a position, which a vehicle exerts on a driver during a cooperation maneuver. Since high levels of jerk can be uncomfortable for people in vehicles, a maximum jerk or an integrated jerk over all or parts of the cooperative maneuver can be used. In particular, it can be defined that small changes in the acceleration below a lower threshold j do not cause any costs. If the maximum absolute jerk j_max exceeds this threshold value, linearly increasing costs can be assumed (other formalizations are also possible):
- a path length can be defined as a term of the merit function, namely the distance traveled by the HV during the manoeuvre:
- w_hv d_hv/d*_hv - 1, where d_hv is the longitudinal distance traveled by the HV during the maneuver and d*_hv is the distance it would travel in the same time if slowing down and behind a preceding one vehicle would remain.
- an average distance covered by the vehicles to be overtaken can also be used.
- the distribution of the quality function values for 867,000 successful, randomized, cooperative overtaking maneuvers is shown in FIG.
- the distribution shows that most generated cooperative maneuvers are able to slightly increase the distance traveled by the HV. This is because random maneuvers are validated and the HV thus manages to overtake at least one vehicle.
- formula (8) can be done based on user preferences as shown in Table 3 (Fig. 7). Of course, other user preferences can also be defined.
- Category 1 riders are all about maximizing the distance they travel. For this goal they are willing to accept big jerks.
- drivers are more susceptible to judder and are not willing to sacrifice driving comfort for a longer distance travelled.
- Other optimization goals could be distance traveled versus maneuver time (promoting high traffic throughput) or the maneuver time itself (promoting faster cooperative maneuvers).
- the category-specific quality functions presented here are denoted by w A (1) and w A (2). The resulting distribution for the quality function of both categories is shown in FIG. 3c.
- the binary classification network was trained on the basis of 800000 generated cooperative maneuvers, which were randomly selected from 1067000 cooperative maneuvers. The remaining 25% can be used for testing.
- a 5-fold cross-validation can be used as an example.
- Inputs are the scenarios, the proposed cooperative maneuver and the validation of the maneuver success.
- the scenario description includes the HV and up to n_rv RVs.
- the cooperation maneuvers can be described by the sub-maneuvers involved, ie vehicle IDs, maneuver types and duration. As an example, these 56 features were normalized for the training. The characteristics are listed in Table 4 (Fig. 8). Unused features were initialized with 0, e.g. B. the parameters of the maneuver containers for non-existent RVs.
- a network with 6 x 128 neurons, no dropout layers and a sigmoid activation function can be chosen.
- an Adam optimizer and in particular a stack size of 32 can be used.
- Training performance can decrease exponentially with increasing training epochs in terms of binary cross-entropy loss.
- a sensitivity (“recall") can be considered more critical than a positive predictive value (“precision"). This is because proposing a safe maneuver that is not safe can be more dangerous to passenger safety than rejecting a safe maneuver that is safe.
- the regression models that predict w A (1) and w A (2) can be trained. To do this, 75% of the data samples can be used for training and the remaining 25% for testing. In particular, only successful maneuvers need be used. This reduces the number to 650000 cooperative maneuvers.
- the regression models can be trained on the basis of the same 56 normalized features used for the binary classification, according to Table 4. In particular, relative values can be used for the lane and the speed of the vehicles to be passed, as this advantageously simplifies the extrapolation. In particular, an overtaking maneuver does not need to depend on the absolute lane in which it takes place. However, it can depend in particular on which lanes the vehicles occupy relative to one another.
- the neural networks for the regression can consist of six layers of 512 neurons each for driver category 1 and eight layers of 4 x 512 and 4 x 256 for driver category 2. These dimensions can be found in particular using a grid search. In particular, no dropout layers need to be used.
- a rectified linear unit (ReLU) can be used as an activation function and the mean square error (MSE) can be used as a loss function.
- ReLU rectified linear unit
- MSE mean square error
- the algorithm according to FIG. 4 can now be used for the prediction of cooperation maneuvers.
- the initiating vehicle invokes this algorithm whenever it sees an opportunity for a cooperative maneuver.
- the call can also be based on an external trigger.
- the HV random maneuvers Based on the currently perceived surrounding vehicles, the HV random maneuvers. They can include all surrounding vehicles, include an overtaking maneuver, and avoid superfluous submaneuvers such as back-and-forth lane changes.
- the success of the generated cooperation maneuvers is predicted with the help of the binary classifier.
- the regression model then predicts the target value w_pred for the maneuvers classified as successful.
- the optimal cooperation maneuver can be determined more quickly through this two-stage implementation, since the regression is carried out solely for the successful cooperation maneuvers.
- the HV selects the cooperative maneuver for which the highest w_pred is predicted and proposes it to the surrounding vehicles. They agree and the overtaking maneuver is carried out.
- the surrounding vehicles can also refuse or suggest a modification of the cooperation maneuver or make it a condition for approval. For each new scenario, in particular, a simulation can take place with a driven cooperative maneuver.
- the time taken to generate proposals may increase linearly depending on the number of random maneuver proposals to choose from.
- the generation of a maneuver proposal including scenario generation, success prediction, and w_pred estimation, can take an average of 50ms on an Intel i7-8665U CPU, with most of the time spent estimating w_pred.
- a vehicle can therefore reduce the number of evaluated maneuvers n_gen in order to meet the requirements of a current driving situation.
- special hardware can be used to further reduce the computation time. When using the method, a compromise must be found between increasing the achievable w_pred and the time required to generate the maneuver candidates. An increase in the suggested number n_gen improves in particular the quality of the maneuver suggestions.
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102021124390.4A DE102021124390A1 (de) | 2021-09-21 | 2021-09-21 | Verfahren und Vorrichtungen für automatische, kooperative Manöver |
| PCT/EP2022/076181 WO2023046728A1 (de) | 2021-09-21 | 2022-09-21 | Verfahren und vorrichtungen für automatische, kooperative manöver |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4405928A1 true EP4405928A1 (de) | 2024-07-31 |
Family
ID=83898543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP22790491.9A Pending EP4405928A1 (de) | 2021-09-21 | 2022-09-21 | Verfahren und vorrichtungen für automatische, kooperative manöver |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US20240355210A1 (de) |
| EP (1) | EP4405928A1 (de) |
| CN (1) | CN118202396A (de) |
| DE (1) | DE102021124390A1 (de) |
| WO (1) | WO2023046728A1 (de) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116055629B (zh) * | 2022-05-27 | 2023-10-20 | 荣耀终端有限公司 | 一种识别终端状态的方法、电子设备、存储介质和芯片 |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170016177A (ko) * | 2015-08-03 | 2017-02-13 | 엘지전자 주식회사 | 차량 및 그 제어방법 |
| WO2018160724A1 (en) * | 2017-02-28 | 2018-09-07 | Wayfarer, Inc. | Transportation system |
| US10188021B2 (en) * | 2017-06-19 | 2019-01-29 | Cnh Industrial America Llc | Path planning system for autonomous off-road vehicles |
| DE102018109885A1 (de) * | 2018-04-24 | 2018-12-20 | Continental Teves Ag & Co. Ohg | Verfahren und Vorrichtung zum kooperativen Abstimmen von zukünftigen Fahrmanövern eines Fahrzeugs mit Fremdmanövern zumindest eines Fremdfahrzeugs |
| US11657251B2 (en) * | 2018-11-12 | 2023-05-23 | Honda Motor Co., Ltd. | System and method for multi-agent reinforcement learning with periodic parameter sharing |
| US11480976B2 (en) * | 2018-12-07 | 2022-10-25 | Qualcomm Incorporated | RSU based vehicle maneuver scheduling |
| US11249480B2 (en) * | 2019-08-08 | 2022-02-15 | Toyota Motor North America, Inc. | Autonomous vehicle positioning system |
| US11460847B2 (en) * | 2020-03-27 | 2022-10-04 | Intel Corporation | Controller for an autonomous vehicle, and network component |
| DE102020004553A1 (de) | 2020-07-27 | 2020-10-22 | Daimler Ag | Verfahren zur Steuerung automatisiert fahrender Fahrzeuge |
| WO2022072764A1 (en) * | 2020-10-01 | 2022-04-07 | Uatc, Llc | Metrics for evaluating autonomous vehicle performance |
-
2021
- 2021-09-21 DE DE102021124390.4A patent/DE102021124390A1/de active Pending
-
2022
- 2022-09-21 US US18/689,371 patent/US20240355210A1/en active Pending
- 2022-09-21 WO PCT/EP2022/076181 patent/WO2023046728A1/de not_active Ceased
- 2022-09-21 CN CN202280060823.6A patent/CN118202396A/zh active Pending
- 2022-09-21 EP EP22790491.9A patent/EP4405928A1/de active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| DE102021124390A1 (de) | 2023-03-23 |
| CN118202396A (zh) | 2024-06-14 |
| US20240355210A1 (en) | 2024-10-24 |
| WO2023046728A1 (de) | 2023-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102019121717A1 (de) | Interaktionsbewusste entscheidungsfindung | |
| EP2812891B1 (de) | Verfahren und vorrichtung zur gemeinschaftsbasierten navigation | |
| EP3776515B1 (de) | Verfahren und vorrichtung zur abstimmung von fahrmanövern zwischen einem fahrzeug und mindestens einem alius-fahrzeug | |
| EP4193135B1 (de) | Computerimplementiertes verfahren zum bereitstellen eines test-verlaufs zu testender verkehrsszenarien | |
| WO2019206377A1 (de) | Verfahren und vorrichtung zum kooperativen abstimmen von zukünftigen fahrmanövern eines fahrzeugs mit fremdmanövern zumindest eines fremdfahrzeugs | |
| DE102019103106A1 (de) | Steuerungssystem und Steuerungsverfahren zur interaktionsbasierten Langzeitbestimmung von Trajektorien für Kraftfahrzeuge | |
| DE102015220481B4 (de) | Verfahren und Vorrichtung in einer Verkehrseinheit zum kooperativen Abstimmen von Fahrmanövern von mindestens zwei Kraftfahrzeugen | |
| WO2017137124A1 (de) | Kraftfahrzeug-steuervorrichtung und verfahren zum ermitteln von ausweichtrajektorien für ein kollisionsfreies ausweichmanöver mehrerer kraftfahrzeuge | |
| EP3543985A1 (de) | Simulieren verschiedener verkehrssituationen für ein testfahrzeug | |
| EP4226248A1 (de) | Verfahren und ein system zum testen eines fahrerassistenzsystems für ein fahrzeug | |
| WO2019215222A1 (de) | Verfahren zum betreiben eines kraftfahrzeugs zur verbesserung von arbeitsbedingungen von auswerteeinheiten des kraftfahrzeugs, steuersystem zum durchführen eines derartigen verfahrens sowie kraftfahrzeug mit einem derartigen steuersystem | |
| DE102017200580A1 (de) | Verfahren zur Optimierung einer Manöverplanung für autonom fahrende Fahrzeuge | |
| DE102013202053A1 (de) | Fahrerassistenzeinrichtung für ein Kraftfahrzeug mit mehreren Fahrerassistenzfunktionen und einer zentralen, fahrerassistenzfunktionsübergreifenden Trajektoriengenerierung | |
| WO2021089238A1 (de) | Verfahren zum bereitstellen einer manövernachricht zum koordinieren eines manövers zwischen einem verkehrsteilnehmer und mindestens einem weiteren verkehrsteilnehmer in einem kommunikationsnetzwerk | |
| DE102022134820A1 (de) | Vorrichtung und Verfahren zur Steuerung von Platooning | |
| WO2023046728A1 (de) | Verfahren und vorrichtungen für automatische, kooperative manöver | |
| DE112020006317T5 (de) | Computersystem und verfahren zum trainieren eines verkehrsagenten in einer simulationsumgebung | |
| AT526259A1 (de) | Verfahren zum Trainieren eines künstlichen neuronalen Netzes eines Fahrermodells | |
| WO2019145200A1 (de) | Verfahren zum erzeugen einer fahrstrategie eines fahrzeugs | |
| DE102018211726A1 (de) | Verfahren zum automatischen maschinellen Trainieren eines elektronischen Fahrzeugführungssystems, sowie Kraftfahrzeug | |
| DE102020209403A1 (de) | Verfahren zum dynamischen Anpassen eines Fahrzeugverbands aus mehreren Fahrzeugen, sowie Computerprogrammprodukt und elektronisches Verwaltungssystem | |
| DE102019215141B4 (de) | Verfahren zum Prognostizieren einer zukünftigen Verkehrssituation in einer Umgebung eines Kraftfahrzeugs durch Bestimmen mehrerer in sich konsistenter Gesamtszenarios für unterschiedliche Verkehrsteilnehmer; Kraftfahrzeug | |
| WO2020177957A1 (de) | Möglichkeit zur fahrerbewertung | |
| DE102020204992B4 (de) | Verfahren und Vorrichtung zum Austauschen von Informationen zwischen zumindest einer Fahrzeugkommunikationseinheit und einem Netzwerk | |
| DE102022207041A1 (de) | Verfahren zum Planen von Bewegungstrajektorien durch ein Modell für maschinelles Lernen |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20240306 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20260402 |