Method, device, system, neural network and computer program for operating a vehicle

Abstract

The invention relates to a method (100, 200), device, a neural network (DNN) in particular a Generative Pre-trained Transformer (GPT), and a computer program for operating (106) a vehicle (EgoF). The procedure (100, 200) comprises: training (201) and / or operating (203) a neural network (DNN, GPT) to operate (106) a vehicle (EgoF), wherein the training (201), in particular retraining (202) and / or operating (203) is carried out depending on data based on driving situation data that are characteristic of a driving situation that at least potentially affects the vehicle (EgoF), in order to generate evaluation information (104) that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation.

Description

[0001] The invention relates to a method for operating a vehicle, in particular comprising executing a measure in the vehicle. The invention further relates to a (corresponding) device, a neural network, in particular a Generative Pre-trained Transformer (GPT), and a computer program for operating a vehicle.

[0002] Methods for partially and fully automated driving and parking are known from the state of the art. In particular, such methods are based on (typically programmed, predefined) objective criteria or rules. Such methods allow, for example, maneuver planning, trajectory planning, or maneuver execution based on vehicle sensor data as well as such criteria and rules. Known methods are based on specific technical rules or criteria and / or material-related metrics. This also applies to the corresponding behavior of such vehicles in road traffic. Some aspects and questions of ethics, morality, and law (e.g., with regard to highly automated driving or parking) are not adequately addressed by the state of the art. These are typically complex and subjective, which is why they cannot be programmed into a specific set of unambiguous rules.

[0003] The object of the invention is to provide an improved method, an improved device, an improved neural network, in particular a generative pre-trained transformer, and an improved computer program for operating a vehicle. In particular, those that make it possible to advantageously, particularly quickly, flexibly, and precisely consider a driving situation, the consequences of a driving situation in the vehicle's surroundings, and / or to react to a (possibly only possible) driving situation of the vehicle.

[0004] This is not about (unconditional) compliance with laws, regulations, or homologation law regarding the vehicle as a product. These are not called into question by the submitted document.

[0005] Rather, the aim of this document is to take into account the (traffic) legal and / or moral aspects better (than in current practice) and / or beyond the extent prescribed by laws and / or standards.

[0006] The problem is solved by each of the independent claims. Advantageous embodiments are described, among other things, in the dependent claims. It should be noted that additional features of a patent claim dependent on an independent patent claim can form a separate invention, independent of the combination of all features of the independent patent claim, without the features of the independent patent claim or only in combination with a subset of the features of the independent patent claim, which invention can be made the subject of an independent claim, a divisional application, or a subsequent application. This applies equally to technical teachings described in the description, which can form an invention independent of the features of the independent patent claims.

[0007] According to a first aspect of the invention, a method for operating one (or more) vehicles is described. The (first) method comprises: determining the driving situation data that are characteristic of a driving situation that at least potentially affects the vehicle; and operating at least one neural network depending on the data based on the driving situation data, wherein evaluation information is generated that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation; and (thereafter, depending on a result, in particular a partial result or preliminary result) operating the vehicle depending on the evaluation information.

[0008] The legal and / or moral assessment of the driving situation can refer to (already existing, possible, and / or preferably: predicted) consequences of the driving situation (e.g. for the vehicle, users of the vehicle and / or one or more road users).

[0009] Preferably, one or more of the legal aspects described in this document may refer to (or be selectively selected from) one or more aspects of traffic law or regulatory law. Compared to general law (e.g., product law relating to vehicles as a product), this has certain special features and does not represent a traditional area of ​​law in the legal sense.

[0010] In particular, one or more legal assessments are carried out in the sense of: - Traffic civil law, in particular liability law (compensation and damages following a traffic accident); and / or - traffic violation law; and / or - traffic criminal law; and / or - Traffic administrative law; and / or - Insurance law.

[0011] In particular, the legal assessment (also: assessment of the legal aspects) can be carried out essentially in accordance with regulatory matters (in Germany or corresponding regulatory matters of the respective countries) or the respective current successor regulations of this type, concerning: - Road Traffic Act (StVG); and / or - Road Traffic Regulations (StVO); and / or - Road Traffic Licensing Regulations (StVZO); and / or - Driving Licence Regulation (FeV) and / or - European Directive 2015 / 413.

[0012] Alternatively or additionally, the legal aspects described in this document may be case law and / or interpretation of (traffic) law and / or corresponding case law.

[0013] In particular, the neural network can be trained and / or operated to interpret traffic law, in particular case law, on the driving situation (e.g. specific ones that at least potentially affect the vehicle).

[0014] These may involve legal aspects that (in contrast to the unquestioned observance of laws or traffic regulations) are complex and / or unknown to the borrower or cannot be adequately interpreted by them, especially during a critical situation. This can bring significant benefits, especially to users (borrowers, end users, vehicle operators).

[0015] For example, the assessment of the driving situation refers to a legal and / or moral situation, situation, consequence corresponding to the driving situation or its consequences, in particular due to a driving situation that has already occurred, is taking place or is expected, predicted, in particular a (e.g. possible, expected, predicted) initial scenario of the driving situation or its consequences, in particular concerning legal and / or moral consequences (also to be understood as: situation, situation, consequence).

[0016] The consequences of the driving situation can include a (e.g., legally and / or morally) risky situation, a near miss, an accident, an interaction between road users (e.g., influencing mutual movement, e.g., yielding right of way, pushing or harassing, displacing, impolite driving, impairment of safety and / or comfort), material and / or moral (also: intellectual, emotional) damage (e.g., courteous or impolite driving). The driving situation can be a driving situation that cannot, is not likely to, or has not led to any material damage.

[0017] In particular, the method can (also) generate evaluation information or use it to operate the vehicle, which indicates sufficient legal and / or moral safety (e.g., according to a predetermined criterion) and / or legal and / or moral concern that exceeds a certain level. For example, evaluation information indicating sufficient safety or improved legal and / or moral consequences can trigger and / or unlock a functionality, in particular one or more performance features of a functionality.

[0018] The operation of the vehicle, in particular a measure implemented in the process, may also involve a (further) improvement of the legal and / or moral consequences (situation, position, so to speak, of positive dimensions with regard to the legal and / or moral situation). For example, "the automated vehicle has saved a road user, helped that user, or improved their situation through a measure (motivated by improved consideration of law and / or morality)."

[0019] The evaluation information may identify or take into account qualitative and / or quantitative measures and / or coded information, symbolically represented information, graphical information and / or textual information, one or more evaluations (generated by the neural network).

[0020] The legal and / or moral consequences differ in particular from or go beyond the (purely) material consequences or (any clear, unambiguous) violations of rules of conduct or traffic regulations.

[0021] For example, the measures of legal and / or moral consequences are subjective or influenced by history, (respective) culture, (respective) jurisprudence, precedents, (respective) morals, ethics (e.g., of certain societies, cultural circles, social classes, etc.) and / or cannot be clearly derived rationally and / or logically. The invention also includes the recognition that these (significantly more than purely logical or material-related criteria) are crucial for the market launch and success of modern vehicles.

[0022] In particular, the method comprises generating the evaluation information which characterizes (e.g. qualitative and / or quantitative) measures of the legal and / or moral evaluation (of the driving situation, consequences, etc.) in an explicit form (e.g. as a parameter, an array of parameters, coded parameters, table, etc.) and / or as a difference (e.g. measures of the difference as a difference, factor, coefficient) or a comparison to other (possible) consequences (e.g. direct material damage, property damage, personal injury for one or more participants in the driving situation).

[0023] For example, at least one neural network is trained, retrained and / or operated to generate the measures of the legal and / or moral evaluation in the evaluation information in an explicit form and / or as a difference or comparison to other consequences (e.g., to immediate material damage, property damage, personal injury).

[0024] In the context of this document, the term "characterize" should also be understood as "represent" or "characterize." This means that "characterizing" can particularly mean "representing" or "characterizing" (e.g., for qualitative and / or quantitative measures).

[0025] The evaluation information may in particular: - a (e.g. potential) violation of law or morality (e.g. to a certain extent); and / or - a (e.g. sufficient or insufficient) conformity with the law and / or morality; and / or - an improved legal and / or moral situation, condition, consequence; take into account or indicate.

[0026] This may refer to the (e.g. specific, respective) driving situation and / or to one or more road users involved or at least potentially involved in the driving situation (who, e.g. could play a role or are expected to play a role depending on the development of the driving situation).

[0027] In particular, at least one step of the procedure may take into account: - adjustable criteria (e.g. by the user or operator of the vehicle) or one or more thresholds relating to legal and / or moral aspects; and / or - the area of ​​operation of the vehicle, e.g. country, province, place where the vehicle is currently being operated or where the driving situation occurs;

[0028] For example, the importance of the legal and / or moral aspects for the vehicle or the sensitivity of the vehicle (degree of consideration, respective weighting, in particular importance) in relation to them can be adjusted.

[0029] In the context of this document, the term assessment is to be understood in particular as a legal and / or moral opinion, assessment, judgment, sentence, total or partial acquittal, degree or proportion of guilt, resulting legal claims, etc. In particular, it is an assessment that differs from or goes beyond a rational assessment or assessment resulting from a rule or set of several rules, in particular an adaptation or clarification of an assessment related to other aspects or consequences of the driving situation, especially in specific cases.

[0030] The term “morality” is (for ease of description) to be understood alternatively or additionally as ethics and / or a subjective opinion (specific to people in general, to certain people, to groups of people) (e.g. evaluation, personal or public opinion and / or an opinion of one or more media, social groups, sections of the population).

[0031] The operation of the at least one neural network may include or be the operation of the (at least partially appropriately trained) neural network. The operation of the vehicle may be dependent on the operation of the neural network, in particular the output layer of the neural network.

[0032] The evaluation information can be generated at the output layer of the at least one neural network. In particular, this should also be understood as meaning that the evaluation information is generated or determined depending on the data generated or provided by the output layer.

[0033] Preferably, at least one neural network is or is trained, in particular retrained, depending on first data which characterise one or more driving situations from the past (e.g. a large number of precedents) predominantly or entirely objectively, in particular in the form of respective driving situation data and / or protocols, witness statements, data based on sensors; and / or depending on second data which characterise one or more at least partially, predominantly or entirely subjective legal and / or moral assessments, in particular one or more points of view, opinions, media reports, judgments of those involved in the driving situation (e.g. road users relevant to the driving situation) and those not involved in the driving situation (e.g. concerning legal and / or moral analysis) on the one or more (preferably the same or similar) driving situations.

[0034] Preferably, the operation of the neural network comprises a generative process. This process can be initiated (also: stimulated or initiated) depending on data based on the driving situation data. For example, the evaluation information (e.g., specific or case-specific for the vehicle, driving situation, user, etc.) is generated when the neural network is operated (e.g., stimulated) with data dependent on (specific) driving situation data.

[0035] The neural network described in this document is preferably a neural network (primarily, to an increased extent) set up and / or trained, in particular retrained accordingly (primarily thereon), to evaluate the legal and / or moral aspects, in particular the consequences of driving situations (according to the characteristics described in this document).

[0036] By means of such an orientation or specialization specific to the task of the invention, undesirable effects or influences (typical for neural networks) can be reduced and / or a rapid generation of the evaluation information, in particular the real-time capability of the method or device, can be significantly improved.

[0037] The term "neural network" refers, in particular, to a deep neural network. Alternatively or additionally, the term "neural network" also refers to another form of artificial intelligence (in particular, one comprising a neural network or a comparable structure). In particular, the training and / or operation of one or more neural networks is understood to mean the analogous application of other types of machine learning.

[0038] For example, the at least one neural network, the training and / or operation thereof can comprise or be an analogous application of one or more of the following conditions or combinations thereof: fully or partially supervised learning, fully or partially unsupervised learning, at least partially reinforcement learning, multitask learning, classification and regression trees, support vector machines, logical and relational learning, probabilistic graphical models, rule learning, instance based learning, latent representation, bio-inspired approaches.

[0039] Particularly preferably, reinforcement learning can be applied in whole or in part. For example, the reinforcement can be performed depending on data (preferably the first data and / or second data, as described in this document), e.g., one or more particularly relevant cases, in particular precedents.

[0040] Preferably, the neural networks or artificial intelligence described in this document (the one or more underlying the models, in particular the generative pre-training transformers or the functionalities) are so-called weak and / or (relatively narrowly) specialized artificial intelligence.

[0041] These (or the corresponding models or neural networks) can be specialized for the execution of the (individual or multiple) functionalities described in this document and / or optimized for the execution of these to a certain extent and / or limited to these (at least predominantly or essentially). This makes the method significantly simpler and / or implementable with limited resources (e.g., using the means of the neural network) and / or more real-time capable and / or (in practical handling) easier to control.

[0042] For example, this can reduce potential risks associated with the use of (e.g., relatively universal or strong) artificial intelligence and / or make the outcome of the procedure more predictable.

[0043] The term "training" refers in particular to machine learning, non-rule-based learning, and / or deep learning, or the execution of machine learning, non-rule-based learning, and / or deep learning. Training can include or be (at least partially, entirely, or predominantly) supervised or (at least partially, entirely, or predominantly) unsupervised training. Training the neural network is particularly (also) to be understood as setting up the neural network.

[0044] The training (or setup) of at least one neural network can include or be end-to-end learning and / or (at least partially, entirely, or predominantly) reinforcement learning. Several, in particular a predominant number, or (essentially) all, intermediate steps necessary for achieving one or more goals can be mapped or integrated into the (at least partially trained) neural network.

[0045] Operating the neural network is or includes, in particular, controlling or stimulating the neural network, in particular the input layer (one or more inputs) of the neural network. Furthermore, changing the operating mode of the at least one neural network can also be considered operating the neural network.

[0046] In particular, the at least one neural network is trained and / or operated depending on a combination of at least two (mutually) corresponding parameters (e.g., data of two different types or significantly different variables described in this document). In this case, one or more multidimensional optima (e.g., with respect to legal and / or moral standards) can be sought, in particular achieved or maintained with a certain degree of approximation.

[0047] Training (especially to be understood as the process of training) can include or be: - determining training data, in particular on the basis of data from previous executions of one or more steps of the procedure (e.g. corresponding to similar characteristics) and / or data relating to legally and / or morally relevant cases from the past; and / or - determining and / or initiating backpropagation, in particular on the basis of data from previous executions of one or more steps of the procedure (e.g. corresponding to similar characteristics) and / or data relating to legally and / or morally relevant cases from the past; and / or - a selection or adjustment of one or more parameters of the training process (e.g., supervision and / or reinforcement), in particular based on data from previous executions of one or more steps of the process (e.g., corresponding to similar characteristics) and / or data relating to legally and / or morally relevant cases from the past. This can, for example, be based on the first data and / or second data described in this document (data based on these data are also to be understood).

[0048] The training and / or operation of the one or more neural networks can be carried out according to one or more predetermined and / or process-determinable (e.g., by a user and / or a dispatcher), adjustable and / or dynamically variable parameters (e.g., as described in the context of this document).

[0049] The term "operating" the at least one neural network can be understood as using the at least one neural network (e.g., the neural network trained according to one or more features described in this document), particularly in active operation. Operating the neural network can mean operating a neural network (e.g., on another computing unit) dependent on or based on the data (training data) of the neural network.

[0050] Depending on the variant of the method, the method comprises or is, for example, the training (pre-training or re-training) of the neural network. In this case, it is to be understood as "the neural network is trained (pre-trained or re-trained)" and / or the operation of the already at least partially trained neural network. In the latter case, it is to be understood as "the neural network has already been trained (pre-trained or re-trained) (e.g., at least to a suitable degree)." The method can therefore comprise or be the first method described in this document and / or the second method.

[0051] For example, the at least one neural network (e.g. to be used during operation) can be understood as features of or data based on the at least one (e.g. at least partially trained) neural network.

[0052] For example, the data is generated during training of the at least one neural network on any computing unit, in particular the computing unit of the backend, in order to then make it operable on one or more other computing units, in particular in (e.g. respective, specific) vehicles, in the central or networked computing unit.

[0053] The operation of the at least one neural network can comprise or be a setting up of the at least one neural network depending on the data from at least one corresponding neural network or one trained for the corresponding use.

[0054] The training, in particular retraining and / or operation of the at least one neural network, can take place inside the vehicle or outside the vehicle (e.g., by transmitting data from or to the vehicle). For example, the neural network can (but does not have to) be located or deployed in the vehicle.

[0055] In particular, at least one neural network is or is trained, in particular retrained and / or operated, to generate evaluation information with regard to a current (e.g. currently started, currently taking place or existing) and / or a (merely) possible, probable, in particular (e.g. for a certain time interval in the near future) predicted driving situation, in particular with regard to the vehicle and / or with regard to certain (e.g. in the surrounding area, relevant to the driving situation, at least potentially affected by the driving situation and / or by a possible interaction with the vehicle) road users.

[0056] Operating the vehicle can include or be executing, controlling, or changing the operating mode of a control unit (particularly also to be understood as a control device) and / or a vehicle functionality, in particular a user functionality and / or system functionality of the vehicle. For example, the vehicle functionality can be a chassis functionality, a comfort function, a functionality for at least partially automated and / or remote-controlled driving, parking, or action of the vehicle (e.g., a driver assistance function). It can also be a functionality effective in the interior or cockpit of the vehicle (e.g., functionality effective relating to physical, particularly mechanical, changes). It can be any fully or partially automated functionality of the vehicle or a functionality in which the user of the vehicle and the vehicle interact.

[0057] Operating the vehicle may involve or be a selection and / or adjustment of one of several functionalities or performance features of one or more of the vehicle's functionalities. These functionalities may be fully or partially automated.

[0058] Operating the vehicle may include or be effecting (e.g., parameterizing, initiating and / or executing, suppressing, deactivating) an action in the vehicle. This may, in particular, include determining or adjusting the parameters of the action.

[0059] For example, depending on the determined evaluation information, in particular on (several, each specific) measures characterizing legal and / or moral consequences, at least one unit of the vehicle (e.g., a control unit, another neural network, functional unit, and / or, directly or indirectly, an output unit and / or an actuator of the vehicle) can be controlled and / or a vehicle functionality can be operated, in particular influenced, controlled (also: triggered), or adapted. This can be done according to a specific (predetermined) dependency (e.g., according to a specific logic, formula, mathematical relationship, etc.).

[0060] Particularly preferably, the evaluation information relates to one or more (possible) variants and / or initial scenarios of a driving situation, in particular several variants or initial scenarios into which a current or predicted driving situation can (still) develop, in particular with a probability measure (e.g. corresponding to a minimum criterion).

[0061] For example, the generation of the evaluation information is carried out with respect to (for) several (e.g. two, three, four, more than four), in particular alternative driving situations, in particular (e.g. still possible) variants or initial scenarios of the driving situation.

[0062] Preferably, the evaluation information relates to non-trivial driving situations and / or driving situations with non-trivial consequences and / or driving situations in which several and / or complex legal and / or moral principles apply.

[0063] For example, these are (one or more) driving situations that: - exceed a certain threshold of complexity, in particular to cover complex driving situations (e.g. involving, in particular interaction, more than two, three, five, seven different road users, in particular vehicles of different types, pedestrians, cyclists, etc.); and / or - have non-trivial, particularly complex (e.g. distributed among several parties) legal and / or moral consequences; and / or . - from a legal and / or moral point of view, simultaneously concern more than two, three, five, seven different traffic rules, codes of conduct, laws, precedents, case law, judgments, moral or ethical principles or precedents.

[0064] For example, the evaluation information can be for several (e.g. possible, in particular predicted with respective probability measures) driving situations or variants of the driving situation (e.g. variants of a further development of the driving situation) or initial scenarios that (each) occur or can occur due to alternative (e.g. not yet known) influencing factors.

[0065] For example, the one or more road users may be a road user (e.g. a road user of a certain class, a pedestrian, child, cyclist, another vehicle, in particular a vehicle of a certain type, etc.) who is in the environment, will participate in the driving situation (possibly, according to expectation or prediction) or will be relevant for it. Alternatively or additionally, it may be a road user who (e.g. from a spatial area that is currently not visible or detectable, e.g. exit, garage, cross street which may be obscured), in particular with a certain degree of probability, may appear and / or become relevant for the driving situation.

[0066] Therefore, the procedure also allows for the assessment and / or consideration of the legal and / or moral consequences if the driving situation, in particular a part (e.g. a causer, participant, influencer) of the driving situation (at least at the current time interval) is not yet or not sufficiently known, has not been recorded by sensors or is not detectably recognizable.

[0067] Particularly preferably, evaluation information can be generated for one or more (e.g., different) reactions of the vehicle, the user of the vehicle, and / or one or more (relevant to the driving situation) road users, e.g., to the driving situation, a road user, or an obstacle (e.g., unexpected or overlooked for a long time). Depending on the processing, in particular comparison (e.g., according to certain criteria) of the multiple evaluation information items, a legally and / or morally better variant of the driving situation and / or the reaction of the vehicle and / or the user of the vehicle can be determined. This can be taken into account in the method. For example, the operation of the vehicle, in particular a measure relating to the vehicle and / or the user of the vehicle, can be dependent on the result of the processing. In this case, for example,a better variant of the driving situation, in particular the initial scenario of a (possible, ongoing or not yet completed) driving situation and / or a better reaction and / or decision of the vehicle (e.g. a functionality of the vehicle) and / or the user of the vehicle.

[0068] For example, the evaluation information can be characteristic (and subsequently taken into account when operating the vehicle): what the legal and / or moral consequences would be if, for example, a scooter rider were to drive out of an exit or a playing child were to run out.

[0069] For example, the neural network is or is set up, trained, retrained and / or operated to generate evaluation information on one or more driving situations, in particular one or more consequences of the driving situations, in such a way that this subjective evaluation (also evaluation, opinion, judgment) of one or more people, in particular a group of people, can be predicted, anticipated and / or simulated (at least approximately or as well as possible).

[0070] Preferably, at least one neural network is set up, trained, in particular retrained, to generate the evaluation data in such a way that the evaluation represented by the evaluation data predicts, anticipates and / or simulates (at least approximately and / or in advance) an evaluation by people, in particular a group of people (e.g. specific people or groups of people characterized by specific characteristics, in particular public opinion of specific media).

[0071] For example, by means of the appropriately trained and / or operated neural network, evaluation information is generated that predicts and / or anticipates and / or simulates a (subjective or subjectively influenced) evaluation of a person, in particular a group of people.

[0072] In particular, the at least one neural network can be set up, trained, in particular retrained, to predict, anticipate and / or simulate the evaluation of a group of people (e.g. a court, a committee, an institution, one or more media, in particular (certain or corresponding to certain characteristics) internet reports, posts, newspapers, television, streaming channels) by taking into account several people (e.g. people with certain characteristics or corresponding to certain characteristics), their respective opinion formation and / or group dynamics (e.g. concerning the interaction of different opinions, dependence on and / or orientation towards one or more opinions of others, etc. when forming an opinion or making a judgment).

[0073] For example, a legal and / or moral assessment of a group of people, in particular comprising different people, parties or interest groups, lobbyists, committees, media, which are particularly characterized by certain features (particularly advantageous), can be predicted, anticipated and / or simulated.

[0074] For example, multiple pieces of evaluation information can be generated corresponding to different (e.g., predicted, anticipated, or simulated) evaluations (subjective evaluations, ratings) of people or groups of people with different characteristics. This information, or information dependent on it, can be processed together, in particular aggregated or processed into a statistical or cumulative value, compared with each other, etc. For example, the vehicle is operated depending on the result of processing multiple (such) evaluations.

[0075] In particular, the neural network can be trained or can be and / or operated to generate evaluation information (e.g. comprising technical, qualitative and / or quantitative measures, objectified or objective) which at least approximately (as accurately as possible) takes into account or characterizes an evaluation (e.g. subjective) which one or more people or groups of people (specific or corresponding to specific characteristics) would form (also: give, submit) or announce regarding the driving situation, consequences of the driving situation and / or concerning the vehicle, and / or users of the vehicle and / or one or more road users.

[0076] The at least one neural network can be trained, in particular retrained, or can be and / or operated to generate evaluation information that takes into account or characterizes an evaluation of the driving situation that at least potentially affects the vehicle: - from a (relatively) neutral moral and / or legal point of view (e.g. with regard to its consequences, as a whole); and / or - from a moral and / or legal point of view of or representing the interests of the user and / or operator of the vehicle; and / or - from a moral and / or legal point of view and / or in representation of the interests of one or more road users (e.g. one or more endangered, injured or endangering persons, damaging persons, etc. or as a respective representative of interests).

[0077] At least one neural network can be trained, in particular retrained and / or operated, to generate specific evaluation information. This can be: - country-specific and / or legal system-specific and / or specific to a particular type of court, commission, body, medium; and / or - for people or groups of people who correspond to one or more specific characteristics, in particular from a social class, educational level and / or attitude.

[0078] For example, the neural network can be trained, in particular retrained and / or operated to predict and / or anticipate and / or simulate a legal and / or moral assessment (e.g. evaluation, opinion, statement, judgment, expert opinion, publication) or action (procedure, prohibition, appreciation, etc.) of a person (e.g. a judge, a moral expert, an ethics expert) or a group of people (e.g. a specific group of people, a group of people comprising specific people, a group of people that includes people with specific characteristics), in particular a court, jury, experts, commission and / or a person or group of people that correspond to specific characteristics, in particular roles, competencies, expertise, mandates.

[0079] For example, people can be members of or groups of people: A (e.g. a specific) court, for example in Germany or Europe: - Traffic Court; - District Court, in particular traffic violation departments; - criminal court; - civil court; - Constitutional Court, in particular the Federal Constitutional Court; - European Court of Justice.

[0080] A (e.g. a specific) court in the USA: - Federal Court of First Instance (US District Courts); - Traffic Court or Municipal Court; - Federal Court of Appeals (US Courts of Appeals); - Supreme Court of the United States (US Supreme Court); - State Court.

[0081] Alternatively or additionally, the assessment information may refer to specific types of courts or institutions that are (primarily) characterized by different variants of decision-making (e.g. known), in particular: - Judicial proceedings, in particular by professional judges (so-called “non-jury proceedings”); - lay judge court; - Arbitration; - jury trial; - Mediation between the parties involved;

[0082] The one or more institutions (e.g. specific institutions) concerning morality, especially ethics: - German Ethics Commission for Automated and Connected Driving; - European Road Safety Authority (EurRAP), in particular at European level; - National Highway Traffic Safety Administration (NHTSA); - Ministry of Industry Public Security, especially for China; - Ministry of Public Security; - other (specific) bodies, commissions, and instances with a significant relevance to ethical and moral issues, guidelines, and judgments. These may include experts (people), academic circles, and specialists from one or more relevant disciplines, industries, legal systems, ethics, and morals.

[0083] Institutions relating to the media or public: - certain media or certain channels; - certain types of media (e.g. newspapers, television, radio, streaming services); - Media with different audiences (e.g. social classes); - Public opinion of (statistical) people of one or more social classes, educational levels and / or attitudes.

[0084] These examples of human groups also include, in particular, the corresponding counterparts or the closest corresponding or comparable authorities in other countries, societies, cultures and / or legal systems.

[0085] In particular, the cases or distinctions described above are taken into account when training and / or operating the at least one neural network. For example, the training of the neural network is or was primarily carried out in a country-specific and / or legal system-specific manner and / or specifically with regard to certain people, groups of people, in particular certain types of institutions, media, and / or specific to the characteristics of one or more people (from the group of people), in particular the social class, educational level, profession, and / or attitude of the one or more people, based on corresponding assessment information.

[0086] For example, a jury, panel, or internet forum comprising a teacher, an engineer, a farmer, an unemployed person, an astronaut, a Democrat, a Republican, an entrepreneur, a blogger, etc.

[0087] For example, the neural network is trained and / or operated (e.g. in operation) at least primarily depending on the corresponding (first) data, e.g. descriptions of the cases (e.g. relevant precedents) and / or (at least predominantly made by humans, corresponding) assessments (e.g. judgments, expert opinions, statements, degree of acceptance, especially of social, media reports).

[0088] Alternatively or additionally, a specific (respectively desired) specificity of the evaluation information can be taken into account during training, retraining, and / or operation of the neural network. For example, the neural network is operated to generate the evaluation information with a specific specificity, in particular entirely or primarily specific to certain characteristics or characteristics corresponding to certain characteristics: vehicles, driving situations, traffic systems, users, legal systems, countries or provinces, people, groups of people, etc.

[0089] For example, depending on the location (also: operation) of the vehicle, in particular in a country, federal state, district and / or area of ​​application of a certain law or legal system in which the (possible) driving situation occurs or may possibly occur, different, in particular adapted, assessment information can be generated.

[0090] For example, evaluation information relating to several alternative driving situations, in particular variants of a (further development of) a driving situation, can be subjected to processing, in particular a comparison (e.g., according to a predetermined mathematical relationship, in particular a formula, coefficients of a predetermined formula, weighting, decision matrix, etc.). The result of the processing can then be taken into account when operating the vehicle.

[0091] For example, this approach can enable consideration of the legal and / or moral aspects of the driving situation, particularly its consequences (e.g., for the respective user). This can be achieved, in particular, by applying a specific weighting to each other and / or to other assessments (e.g., to assess a different type of damage, such as non-material damage).

[0092] The term user of the vehicle can - a passenger, in particular the driver of the vehicle; and / or - an operator (e.g. manufacturer, lessor, transport company, taxi company) of the vehicle; and / or - a user outside the vehicle who, in particular, operates or controls the vehicle at least partially and / or temporarily by remote control and / or teleoperation.

[0093] For example, it may be a user who monitors and / or controls the fully or partially automated and / or at least partially teleoperated driving and / or parking of the vehicle. Alternatively or additionally, the user of the vehicle may be a technical user (e.g., partially, predominantly, or entirely configured with technical means, such as a computer program or artificial intelligence).

[0094] In the context of this document, the term “user” also refers to a male or female user.

[0095] The operation of the vehicle may include or be a measure (e.g. in a particularly difficult or undesirable) driving situation or to avoid a (particularly difficult or undesirable) driving situation, in particular a variant or initial scenario of the driving situation and / or to mitigate the (at least potential, threatening) consequence of a (e.g. predicted) driving situation, in particular one of several variants of a driving situation, in particular one that has not yet taken place or has not yet been completed.

[0096] Alternatively or additionally, the operation of the vehicle may include or be the generation of an output (e.g., graphical, symbolic, textual) for the user, in particular the driver of the vehicle. The output may include or be a decision-making aid for a decision to be made and / or an action to be performed by the user of the vehicle (e.g., in a particularly difficult or undesirable driving situation) or to avoid a (particularly difficult or undesirable) driving situation and / or to mitigate the (at least potential, impending) consequences of a (e.g., possible, expected, predicted) driving situation. For example, potential legal and / or moral consequences can be reduced or averted.

[0097] The operation of the vehicle, in particular a measure relating to the vehicle (e.g. vehicle functionality) and / or relating to the user of the vehicle, can be carried out in the time interval between the determination of the driving situation data or the generation of the evaluation information and a conclusion of the driving situation (e.g. a maneuver, evasive action, lane change, braking, etc.) of the vehicle within the driving situation.

[0098] The evaluation information can be generated at multiple time intervals, in particular points in time, and / or with reference to multiple time intervals, in particular points in time. This information can be generated and updated once or multiple times (e.g., as the driving situation develops). Accordingly, one or more measures can be updated, modified, and / or (e.g., if no longer necessary) not executed, paused, postponed, or canceled.

[0099] For example, the evaluation information may include or be a control signal for operating the vehicle, or the control signal for operating the vehicle may be ascertainable or ascertained, in particular initiated, depending on the evaluation information.

[0100] In particular, the evaluation information can be contained (also: encompassed) as action information (e.g. a control signal for a control unit or vehicle functionality) or within action information for initiating a measure or explicitly or implicitly.

[0101] The driving situation that at least potentially affects the vehicle can be a driving situation that is currently taking place or has been recognized and predicted for a near future of a few fractions of a second or a few seconds, a driving situation that can only take place potentially, in particular with a (certain) degree of probability, and / or a driving situation that has already taken place (e.g. a few seconds or minutes ago), in particular a driving situation that has already taken place.

[0102] Alternatively or additionally, the driving situation that at least potentially affects the vehicle can be a driving situation that can (still) be generated by the vehicle's means (e.g., modified in a certain way) or a new driving situation (e.g., brought about or created by the at least one measure). This can be a driving situation that can be generated (also: brought about, brought about, or created) using the vehicle's functionality for at least partially automated driving and / or parking.

[0103] One or more steps of the method are preferably carried out during operation, in particular during driving or in the context of driving the vehicle. The evaluation information can preferably be generated and / or taken into account during vehicle operation, in particular: - in connection with driving, in particular when the vehicle is operated when the journey is controlled solely or predominantly by the driver; - in connection with driving, vehicle operation during a journey predominantly controlled by a vehicle functionality, in particular a driver assistance system, an active safety system and / or a functionality for at least partially automated driving or parking.

[0104] Multiple evaluation information items, particularly regarding (respective) legal and / or moral consequences (e.g., regarding different variants of the driving situation, different decisions regarding a vehicle functionality, the vehicle user, and / or one or more road users), can be generated and taken into account when operating the vehicle. These can be subjected to a processing step (e.g., according to a predetermined criterion, condition, a mathematical relationship, in particular a formula, coefficients of a specific formula, matrices, characteristic curves, and / or parameter sets, etc.).

[0105] For example, the one or more evaluation information items or the result of processing the respective evaluations are taken into account when operating (e.g. when controlling) an at least partially automated drivable and / or parkable vehicle, in particular the longitudinal and / or lateral guidance of the vehicle and / or the execution of maneuvers.

[0106] For example, the one or more pieces of evaluation information (preferably several pieces of evaluation information relating to different consequences, road users, different legal and / or moral aspects) can be processed as respective input variables for operating the vehicle, in particular a functionality for at least partially automated driving and / or parking.

[0107] In the context of this document, the term "driving situation" can be understood, for example, as a specific situation characterized by a configuration, action, or interaction of road users or by certain driving parameters of road users (particularly at the level of specific objects). In particular, the meaning of the term "driving situation" differs from the frequently colloquially used meaning of the term "traffic situation," which rather corresponds to summary, general, and / or statistical categories such as "free traffic," "heavy traffic," "slow-moving traffic," "traffic jam," "end of traffic jam," etc.

[0108] A driving situation can be described by parameters of the driving situation. One or more parameters of the driving situation can characterize, in particular represent, a specific pattern (also understood as a data pattern), e.g., a pattern characterizing the arrangement and / or speed of objects and / or a pattern of the parameters of the driving situation. The driving situation can also be characterized by a spatial pattern of the so-called open spaces in the vehicle's surroundings or by corresponding parameters.

[0109] Preferably, the at least one driving situation can be characterized by one or more of the following features (or parameters of the driving situation corresponding to these): - a (certain) spatial distribution of road users and / or the movement parameters of road users, in particular an arrangement pattern of road users in the environment of the (real) vehicle; - a (certain) spatial distribution of immobile objects in the vicinity of the vehicle; - a relative position and / or movement parameters to certain types of lane markings, traffic signs, traffic lights (not necessarily to specific traffic lights, etc.); - information about the right of way of the vehicle, in particular vis-à-vis certain road users and / or road users who are actually or at least potentially coming or may come from certain directions, e.g. a crossing road from the right or from the left; - information about an action by a road user in the vicinity of the vehicle, e.g. one which exceeds a limit value, such as honking, flashing lights, pushing, overtaking the vehicle, attempting to overtake and the like.

[0110] Furthermore, the driving situation may be characterized by one or more parameters related to relevant traffic rules, traffic signs, right of way, traffic lights and / or traffic light phases.

[0111] The situation described in this document may have any combination of the characteristics described.

[0112] Preferably, the at least one driving situation may be a driving situation that exceeds certain limit values ​​or a driving situation characterized by parameters that exceed certain limit values. For example, the at least one driving situation may be a driving situation involving an undesirable or dangerous approach to an object or road user, an acceleration value that exceeds a limit value, an undesirable arrangement relative to other road users, etc.

[0113] The driving situation can be a special situation (e.g., a comparatively rare one) or a (at least potentially) dangerous driving situation, e.g., a driving situation for which an increased risk is identified or assumed. For example, the method can include recognizing such a driving situation or a driving situation typically preceding such a driving situation and incorporating the recognized information into the method. For example, one or more steps of the method can be initiated depending on the recognized information, or their execution, execution parameters, mode, etc. can be varied.

[0114] As one or more parameters of the driving situation, a temporal and / or spatial change characteristic, in particular a gradient, such as a temporal and / or spatial gradient of the respective parameter, in particular a parameter value, can (also) be considered or taken into account.

[0115] For example, the driving situation parameter(s) can be taken into account based on or relating to environmental sensor data and / or information transmitted to the vehicle (e.g., from another road user, via a Car-to-Car or Car-to-X system). The environmental sensor data can be data processed in a specific manner from an environment-detecting sensor, in particular a sensor system of at least one vehicle.

[0116] Alternatively or additionally, the process can determine and consider traffic regulation information, identifying one or more traffic regulations (e.g., specific to the location, in particular country, province, or vehicle location). This information can, for example, identify the legal regulations and / or logic.

[0117] Alternatively or additionally, traffic sign information can be determined and considered in the process. This information can indicate the content and / or respective positions of one or more (valid) traffic signs located in the vicinity of the vehicle.

[0118] Preferably, the traffic sign information is information that deviates from and / or goes beyond the content of the traffic sign (as such). For example, the traffic sign information can concern things (e.g., objects, spatial areas, or road users) that acquire a specific significance for the vehicle due to the traffic sign. For example, another road user must be given right of way, or the vehicle itself has right of way over another road user, etc.

[0119] The traffic sign can be (or be selected from) any traffic sign or one assigned to any category according to a standard or agreement (e.g. Vienna Convention, European Standard). The traffic sign can, for example, belong to or be a category of warning sign, prohibition sign, regulatory sign, mandatory sign, directional sign, signpost, border crossing sign, etc. In particular, the traffic sign is a traffic sign whose meaning depends on one or more further (each optional or variable) conditions, e.g. another road user. The traffic sign can, for example, indicate a right-of-way regulation (in its surroundings).

[0120] The traffic sign may optionally comprise a traffic light and / or at least a time-variable display. The traffic sign information can be determined based on sensor data, e.g., camera data from one or more cameras. In particular, the traffic sign information can be determined using a user device carried in the vehicle.

[0121] In the procedure, a (e.g. case-specific) interpretation of (relevant) traffic sign information can be carried out within the framework of the assessment of the legal and / or moral consequences.

[0122] The method has several significant or decisive advantages that are easily understandable for a person skilled in the art. For example, legal and / or moral aspects of a driving situation, particularly its consequences (e.g., usually very complex, locally specific, specific to classes of road users and / or vehicles), and particularly those that differ from rational and / or trivial rules, can be taken into account. This can be done very quickly, particularly (almost) in real time, and / or (e.g., for predicted driving situations) proactively and / or for multiple variants or initial scenarios of the driving situation. This can help a user (e.g., one who does not necessarily have particularly high legal and / or moral competence), especially the driver of the vehicle, who is overwhelmed by a difficult and / or unusual driving situation.In addition, a significant legal, moral and vehicle brand-related advantage (e.g. for the manufacturer, operator, designer) may arise.

[0123] This is not about compliance with laws, regulations, or homologation law with regard to the vehicle as a product. Rather, the (traffic) legal and / or moral aspects arising in certain situations can be taken into account better (than in current practice) or beyond the extent prescribed by laws and / or standards. In particular, the neural network can be trained and / or operated to carry out a complex interpretation of case law, or one that cannot be represented as individual rules, on the driving situation (e.g., specific ones that at least potentially affect the vehicle). The support resulting from the invention can therefore significantly exceed support related solely to compliance with clear rules.

[0124] For example, no (e.g., complex) rules need to be programmed or determined for potentially legally or morally complex driving situations. Overall, compliance with the law and / or morality in traffic can be improved (beyond the prescribed level). Therefore, the invention encompasses several industrially feasible technical solutions to legal and / or moral aspects related to automated and / or remotely controlled driving, parking, and / or operating vehicles.

[0125] According to a further embodiment, the determination of driving situation data is or includes: predicting one or more driving situations and determining the corresponding driving situation data for the predicted driving situation; and / or predicting driving situation data that would characterize one or more driving situations that at least potentially affect the vehicle.

[0126] In this case, the one or more (respective) driving situation data (also: sets of driving situation data for a respective driving situation) can each correspond to one or more variants of a possible development of a current or predicted driving situation.

[0127] The prediction of (relevant) driving situation data that would characterize one or more driving situations that at least potentially affect the vehicle can be based on one or more past driving situations and the corresponding driving situation data. In this case, the driving situation data (e.g., the sensor data that would be determined during these situations) can be predicted (more or less directly) (possibly without recording a driving situation as such).

[0128] The evaluation information for one or more (e.g., potentially expected or generable) driving situations can be used for predictive operation, in particular for initiating a predictive measure. In this case, a certain (sufficient) time (e.g., a few seconds) can be gained during which the most precise evaluation information can be generated and / or the measure can be executed (still to change the consequences of the driving situation). Particularly considering the state of development of neural networks at the time of this application, the time advantage of approximately a few seconds can provide decisive advantages.

[0129] According to a further embodiment, the method comprises: determining, in particular predicting, driving situation data for a plurality of driving situations that at least potentially affect the vehicle, in particular for a plurality of variants or initial scenarios of a current or predicted driving situation; and (thereafter), generating the plurality of evaluation information items corresponding to the different driving situations; and: processing the plurality of evaluation information items. The result of the processing can then be taken into account in the method, in particular in one or more steps of the (first and / or second) method.

[0130] For example, processing can identify a measure to resolve a dilemma. Specifically, a dilemma between: - a legal consequence and a moral consequence; - a legal consequence and a material consequence (e.g. a material asset); - a moral consequence and a material consequence (e.g. a tangible asset); - a legal and moral consequence on the one hand and a consequence related to the material (e.g. a material asset) on the other hand.

[0131] This may involve a dilemma arising within the context of the driving situation or as a result of the driving situation. For example, a measure (e.g., regarding braking, steering, especially evasive maneuvering, or a combination thereof) that results in (moderate) property damage can only be initiated if a consequence (e.g., legally entirely or relatively uncritical) but morally or ethically significant is imminent.

[0132] The (several) variants of the driving situation can include or be variants of a further development (also: change) of a driving situation (e.g. within a few seconds).

[0133] For example, the processing can include or be a comparison of the (respective) legal and / or moral consequences (here also: aspects) with the driving situations corresponding to several driving situation data. For example, respective measures of the legal and / or moral consequences can be evaluated from assessment information (e.g. with certain limit values, variables, and / or compared with each other) and / or according to one or more specific criteria. The vehicle can then be operated depending on the result of the processing. The several driving situations, in particular variants or initial scenarios of the driving situation, can be alternative driving situations (to each other). These can, for example, differ qualitatively, in particular fundamentally and / or (significantly) qualitatively. An example of qualitatively different driving situations, in particular, can include or be, for example, braking and evasive maneuvering (e.g., in the event of an obstacle) by the vehicle and / or another road user (relevant to the driving situation). A qualitative difference between driving situations can, for example, be in a range of values ​​of speed, in particular longitudinal speed and / or lateral speed), acceleration (e.g., longitudinal acceleration and / or lateral acceleration), radius of a curve or the path curve or trajectory of the vehicle and / or another road user.

[0134] For example, legal and / or moral consequences of driving situations are compared that differ qualitatively (e.g. due to the type of maneuver of the vehicle and / or another road user) and / or quantitatively (e.g. in steps of less than 5%, 10%, 20%, 30%, 50%). For example, the generation and processing of at least 2, 4, 8, 16, 32 such (e.g. potentially possible) driving situations (e.g. as variants that could at least potentially, in particular probably, arise from a given initial situation). Subsequently, such a driving situation, in particular the development of the driving situation or variant (e.g. with similar other criteria), can be preferred, selected and / or aimed for which (at least presumably or according to the evaluation information) has better legal and / or moral consequences.

[0135] Preferably, the respective legal and / or moral consequences of several driving situations corresponding to the several driving situation data: - affect different road users, in particular the consequences are of different nature and / or different severity; and / or - affect the vehicle or the user of the vehicle differently, in particular have consequences of a different nature and / or different extent.

[0136] These can each be taken into account in the process (individually, together or as a result of applying a processing as an intermediate step).

[0137] According to a further embodiment, the respective evaluation information takes into account, in particular identifies, the legal and / or moral consequences of several driving situations corresponding to the several (respective) driving situation data. This can be carried out (e.g. in each case or in a differentiated manner) with regard to the vehicle; and / or the user of the vehicle; and / or one or more different road users in the vicinity of the vehicle. This can be done in such a way that the respective results of respective actions (in particular also to be understood as actions) of one or more participants in the driving situation are taken into account, in particular identified, in particular as actions of one or more (specific) different types and / or extents, in particular in each case or in a differentiated manner.

[0138] For example, the evaluation information can comprise a data structure, such as a table or array. The respective sequences can be identified separately or in specific groups (in a differentiated manner). Such evaluation information can be particularly well-suited (with or without a processing step) for operating the vehicle.

[0139] This can then be taken into account when operating the vehicle, particularly when effecting (e.g., initiating and / or executing) at least partially automated longitudinal guidance, lateral guidance, and / or maneuvers. This results in an advantage that is easily understood by those skilled in the art.

[0140] For example, a driving situation, a variant of the development of a driving situation, a starting scenario, or a generable driving situation (e.g., taking possible interventions into account) can be selected or brought about as a better driving situation with regard to legal and / or moral consequences. This can preferably be carried out in advance, before the start or completion of the driving situation, or even before it can be recognized by the vehicle's means.

[0141] In other words, the invention can create the possibility of selecting from several (initially only possible, potential, expectable) future scenarios (“futures”) with regard to a driving situation (so to speak in advance) or of bringing about these.

[0142] According to a further embodiment, this comprises at least one neural network: - a Generative Pre-trained Transformer (GPT); and / or - a generative, in particular statistical model, in particular a model trained on a specific type of vehicle, the specific vehicle, a specific user and / or operator of the vehicle; or is based on such a model and / or is controlled by such a model.

[0143] In particular, said model may comprise or be a language model or a model with language support, in particular a so-called large language model.

[0144] According to a further embodiment, at least one neural network is trained, in particular retrained, depending on data based on: - legal information, in particular concerning traffic law, laws, interpretations of laws, regulations, court rulings; and / or - information relating to morality, in particular ethics, in particular specialist literature, decisions or opinions of commissions, media contributions, in particular concerning moral theses and / or principles and / or precedents; and / or - information related to subjective opinions, in particular media contributions, particularly concerning the formation of opinions by several people and / or evaluations.

[0145] For example, training (e.g., determining training data and / or supervising, etc.) may be carried out depending on such legal information.

[0146] The following publicly known sources are mentioned merely as examples of (suitable) legal information (and to better classify the features of the invention): - Peter Hentschel (founder), Peter König, Peter Dauer (editor): Road Traffic Law (= Beck's Short Commentaries. Volume 5). 44th edition. CH Beck, Munich 2015, ISBN 978-3-406-69610-7. - Ferner (ed.): Handbook of Road Traffic Law. 2nd edition. Nomos Verlag, 2006, ISBN 3-8329-1281-9. - Strehl: Road Traffic Law. 26th edition. Heinrich Vogel Verlag, 2006, ISBN 3-574-27311-8. - Spreng: The New Road Traffic Law. Beck Legal Advisor at dtv, ISBN 3-423-50633-4. - Burmann et al.: Road Traffic Law. Commentary. 21st edition. CH Beck, 2010, ISBN 978-3-406-59421-2. - Rainer Heß, Michael Burmann: The development of road traffic law in 2011 NJW 15 / 2012, 1042. (following the previous article [...] in 2010 in NJW 16 / 2011, 1124).

[0147] In particular, this should be understood as meaning that the neural network is trained, in particular retrained, and / or operated primarily (also) based on such data (e.g., transformed into appropriate tokens), in particular a suitable selection from this data. For example, this data is considered in the process (e.g., during training or retraining) with priority or with a significantly higher weighting than other data.

[0148] For example, at least primarily dependent, in particular on the basis of the data described above: - determining training data; and / or - detecting and / or initiating backpropagation; and / or - a selection or adjustment of one or more parameters of the training process; and / or - Supervising or reinforcement; carried out.

[0149] Particularly preferably, supervision and / or reinforcement is carried out depending on the data described above (e.g., controlled with regard to qualitative and / or quantitative parameters). For example, the above-mentioned data are given priority or are given a significantly (significantly) higher weighting than any other data.

[0150] Alternatively or additionally, it should be understood that the data described above are given a comparatively high weighting when training, retraining and / or operating the neural network.

[0151] The information described above may, for example, include or be text data (e.g. corresponding to a natural or coded language), coded data, in particular with or without conversion into tokens.

[0152] According to a further embodiment, the operation of the vehicle comprises triggering, changing and / or at least partially and / or temporarily suppressing a triggering of a functionality of the vehicle.

[0153] This may involve temporarily activating, modifying, and / or at least partially suppressing the activation of a vehicle function, particularly only if a legal and / or moral consequence otherwise meets a certain criterion. This can be carried out (only or primarily) for the respective driving situation or for the respective specific case and / or deviating from an (otherwise applicable) rule.

[0154] For example, the triggering of a passive safety system, in particular a restraint system (e.g. airbags, a pyrotechnic device) and / or an active safety system (e.g. a device for, in particular, anticipatory pedestrian protection, etc.) in the vehicle (e.g. in borderline cases where this is not absolutely necessary) can be controlled (e.g. activated, deactivated, briefly suspended, changed) depending on whether the user or driver of the vehicle would (presumably) have a comparatively high degree of culpability or no or low degree of culpability for the driving situation, in particular the consequence of the driving situation. The (in each case possible) legal and / or moral consequences of the triggering, non-triggering, or a specific triggering can be taken into account.

[0155] According to a further embodiment, at least one neural network is trained, in particular retrained, depending on: - initial data that predominantly or entirely objectively characterise one or more driving situations from the past, in particular in the form of protocols, witness statements, driving situation data, data based on sensors; and - second data which characterise one or more at least partially, predominantly or entirely subjective legal and / or moral assessments, in particular one or more points of view, opinions, claims, media reports, judgments on the one or more driving situations from the past.

[0156] According to a further embodiment, the method comprises transforming the driving situation data into a character set, in particular tokens, in particular tokens comprising text information or voice information, wherein the character set characterizes the driving situation, in particular the course and / or one or more consequences of the driving situation; and training, in particular retraining and / or operating the at least one neural network depending on the character set.

[0157] The invention (thus also) includes the idea of ​​transforming driving situation data determined as a function of one or more vehicle sensors, in particular characterized by one or more corresponding features and / or data formats known to those skilled in the art, into a character set, in particular a character sequence, which is (particularly well) suited for operating the at least one neural network. The character sequence can in turn be characterized by corresponding features and / or data formats of tokens (preferably linguistic tokens) known to those skilled in the art. This (also) makes it possible to use a neural network, in particular based on a comparatively universal model, e.g., a foundation model (e.g., particularly effectively). Thus, evaluation information (within seconds or fractions of seconds) relating to a (e.g., immediately imminent) driving situation can be generated and taken into account.

[0158] For example, the transformation may include or be a processing step, particularly an interpretive one. During the transformation, the driving situation that at least potentially affects the vehicle may be characterized, in particular described or represented, in the form of a (e.g., formalized) character set.

[0159] For example, the character set or character sequence identifies the course and / or one or more consequences of the (e.g., currently ongoing, possible, expected, or predicted) driving situation. In particular, the character set at least partially identifies one or more (at the time of transformation only possible), expected, or predicted courses, variants, initial scenarios, and / or one or more (respective) possible, expected, or predicted consequences of the driving situation or the courses, variants, or initial scenarios.

[0160] Preferably, the character set can be a character sequence (particularly also to be understood as a character stream) of the corresponding tokens (e.g., elements of text information and / or speech information representing). For example, the character sequence can characterize the current and / or predicted course (also: dynamics, gradual changes) of the driving situation. This allows the method to be executed particularly (computationally) efficiently and / or (almost) in real time.

[0161] The transformation can be performed during or in conjunction with the determination, in particular the prediction, of the driving situation data, and / or for time intervals or sections of the driving situation in the near future (e.g., sections of fractions of a second or several seconds). The time intervals or sections of the driving situation (e.g., sections of fractions of a second or several seconds) can thus be transformed step by step (e.g., for different variants or predicted initial scenarios of the driving situation).

[0162] For example, the transformation into a character set for a driving situation takes place, in particular in the form of the character sequence before the beginning, during the course (e.g. during individual time intervals, e.g. seconds or phases of a maneuver) of the (currently occurring or predicted) driving situation.

[0163] For example, the character set, in particular the character sequence (preferably in sections), is provided to the input layer of the at least one neural network or read in by it. In this case, the operation of the at least one neural network can be carried out particularly efficiently, quickly, and / or almost in real time depending on the (thus transformed) driving situation data or the generation of the evaluation information. In particular, this (only) makes it possible to expediently operate a neural network trained on the basis of text data, in particular one comprising a language model, with the driving situation data (depending on the data based on the driving situation data).

[0164] By using the character set, in particular the character sequence or the text information and / or speech information, a neural network trained or specialized depending on text information or speech information, in particular a generative pre-trained transformer and / or large language model, can (also) be used.

[0165] The invention can help in practical implementation by accessing (very efficiently and economically) such models, individual modules, or building blocks that primarily specialize in speech or text information (e.g., originating from other industries). In particular, the at least one neural network can comprise one or more features of a chatbot, in particular a chat GPT, or be operated depending on it.

[0166] For example, the driving situation data available as image data, point clouds, vector objects, of an environment model, in particular a data-based model (e.g., representing the vehicle's surroundings), e.g., based on the data fusion of various data from different sensors, can be transformed into one or more character sets, in particular character sequences. This can be performed step by step, e.g., in sections of a few fractions of a second or seconds. This allows the process flow to be further accelerated and optimized.

[0167] The transformation can be performed using the same or another (trained) neural network. This can include or be a generation of the transformed information (the character set or character sequence). The neural network (e.g., certain layers of the neural network) and / or another neural network, or the interconnection of these neural networks, can be configured to be trained, retrained, and / or operated according to the features described in this document.

[0168] According to a further embodiment, at least one neural network is or is trained, in particular retrained, to generate evaluation information that predicts, in particular anticipates and / or simulates an evaluation that a person, in particular a group of people, would have (in particular also to be understood as: announce, express, communicate, publish) with regard to one or more driving situations, in particular to one or more possible, expected or predicted consequences of the driving situation.

[0169] For example, a legal and / or moral assessment can be predicted, in particular anticipated or simulated, which a person with certain characteristics or corresponding to certain characteristics (e.g. a judge, expert in morals or ethics), in particular a group of people with certain characteristics or corresponding to certain characteristics (e.g. a court, group of experts, representatives of certain societies and / or cultural circles) would have (in particular also to be understood as forming, generating and / or announcing) in relation to the driving situation (which may not yet have taken place or been completed).

[0170] The evaluation information can take into account, in particular be indicative of, an evaluation (in particular an opinion, assessment, judgment, sentence, full or partial acquittal, degree or proportion of guilt, degree of acceptance, etc.) that a person, in particular a group of people, would have in relation to the driving situation, in particular to a (possible, expected or predicted) course and / or one or more (possible, expected or predicted) consequences of the driving situation.

[0171] For example, evaluation information is generated in advance for one or more variants or initial scenarios of the driving situation. This can occur at a time when the corresponding operation of the vehicle, in particular a measure or a change in the decision (to be made by humans and / or the vehicle's functionality), can (still) provide a significant advantage, particularly with regard to one or more consequences of the driving situation.

[0172] Preferably, at least one neural network is trained, in particular retrained, and / or operated to operate the vehicle (by means of the evaluation information described in this document) in such a way that (in respective driving situations) a legal and / or moral situation, situation, in particular legal and / or moral acceptance (e.g. in the respective society, legal systems, certain groups of people, public) results which meets a specific criterion and / or is improved (optimized to a certain extent). In this case, one or more acceptance criteria can be stored and / or taken into account as a specific and / or (e.g. by the user and / or operator of the vehicle) determinable function dependent on location, user, legal system, society, settings.

[0173] According to a second aspect, the invention comprises a (second) method for operating a vehicle. The (second) method comprises: setting up, training, in particular retraining, a neural network for operating a vehicle based on driving situation data that are characteristic of a driving situation that at least potentially affects the vehicle, to generate evaluation information that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation.

[0174] For example, at least one (e.g., pre-trained) neural network (e.g., based on a corresponding foundation model, particularly within the framework of the computer program product) is provided and retrained (e.g., depending on the data of the specific vehicle or specific characteristics corresponding to the vehicle, user, location, and / or driving situation, etc.). The method may include determining the data and / or executing the training, particularly retraining the neural network (e.g., specific to the vehicle, user, etc.) (e.g., taking into account the settings of the specific user and / or operator).

[0175] The features of the (first) method relating to operating the vehicle or operating the neural network according to the first aspect and features of the (second) method for training, in particular retraining, according to the second aspect can be combined with one another in a variety of ways. In particular, a resulting method comprises the features of the first method and features of the second method. Therefore, the features described in this document can each refer mutatis mutandis to the first method and / or the second method. In the context of this document, the term "method" is to be understood in particular as the first method, the second method, or a combination of a first method and a second method.

[0176] According to a further aspect, the invention comprises a neural network for operating a vehicle. The neural network is trained (i.e., trained at least to a certain degree) and / or retrainable and / or operable based on driving situation data that are characteristic of a driving situation that at least potentially affects the vehicle. Evaluation information is generated (by means of the neural network, at the output layer of the neural network) that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation.

[0177] In the context of this document, the term neural network is to be understood in particular as data of the neural network, in particular certain layers or building blocks of the neural network serving a specific purpose or as data for setting up or updating the neural network (ie a part, certain layers or building blocks of the neural network).

[0178] In particular, the neural network is configured for training and / or operation in and / or for the vehicle. In particular, it is configured, in particular specialized, trained, and especially retrained, primarily for the legal and / or moral assessment of driving situations (e.g., through a suitable architecture). Features of the invention can (also) achieve improved speed, real-time capability, and / or reliability of the resulting assessment information (and thus also of the operation of the vehicle).

[0179] In particular, the training and / or operation of the neural network can be performed in a computing unit located outside the vehicle (e.g., by means of the device, remotely, depending on the data determined by the device and / or for the vehicle). In this case, the (first and / or second) device can be configured to determine and / or receive the corresponding data.

[0180] According to a further aspect, the invention comprises a (first) device. This can be a device for operating the vehicle. The device is designed to determine the driving situation data characterizing a driving situation that at least potentially affects a vehicle; and is designed: to operate at least one neural network depending on data based on the driving situation data, wherein evaluation information is generated that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation; and is further designed: to operate the vehicle depending on the evaluation information.

[0181] According to a further aspect of the invention, a (second) device is described. This can (likewise) be a device for operating the vehicle or be integrated with or connected thereto. The (second) device is designed to set up, train, in particular retrain, a neural network for operating a vehicle as a function of data based on driving situation data that are characteristic of a driving situation that at least potentially affects the vehicle. The neural network will be set up, trained, or will be, in particular retrained, and / or operable to generate evaluation information that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation.

[0182] The features of the (first) device relating to operating the vehicle or operating the neural network according to the first aspect and features of the (second) device for training, in particular retraining, according to the second aspect can be combined with one another in a variety of ways. In particular, the (resulting) device comprises the features of the first device and features of the second device. Therefore, the features described in this document can each refer mutatis mutandis to the first device and / or the second device. The term device is to be understood in particular as the first device, the second device, or a combination of a first device and a second device.

[0183] The setting up, training, in particular retraining and / or operation of the neural network can be carried out by means of the computer program product described in this document.

[0184] According to a further aspect of the invention, a vehicle is described. The vehicle is designed: to determine the driving situation data characterizing a driving situation at least potentially concerning a vehicle; and is designed: to train, in particular retraining and / or to operate at least one neural network. In this case, the neural network is or has been trained, in particular retrained and / or is operable to generate evaluation information that takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation. Furthermore, the vehicle can be designed: to operate the vehicle, in particular a functionality of the vehicle depending on the evaluation information.

[0185] The vehicle may comprise the first device and / or the second device and / or be designed to train, in particular retrain and / or operate, a neural network according to one or more features described in the present document (e.g. inside the vehicle and / or outside the vehicle, by data transmission or remote control, depending on the data from the vehicle).

[0186] The vehicle is preferably a motor vehicle, in particular a road vehicle (e.g., a car, truck, van), or a two-wheeler (e.g., a motorcycle, scooter, pedelec, bicycle). The vehicle can be fully, predominantly, or partially (e.g., manually) controllable (e.g., steerable) by a user of the vehicle.

[0187] Alternatively or additionally, the vehicle can be a vehicle that is at least partially automated and / or remotely controlled and / or that can park, maneuver, exit, or operate at least partially automatically. For example, it is a vehicle that can be driven and / or parked with an automation level of approximately 2, 3, 4, or 5.

[0188] In particular, it may be an unmanned and / or at least partially and / or at least temporarily remotely controllable vehicle (e.g., a vehicle that can be controlled, driven, parked, maneuvered, or acted upon by a remote control). Certain functionalities of the vehicle (e.g., at least partially automated driving and / or parking) may be influenced, in particular controlled, by the user and / or a remote dispatcher.

[0189] In particular, the vehicle is an aircraft (e.g. airplane, drone, aerodynamic guided body, cruise missile or cruise missal), a watercraft (e.g. ship, yacht, boat, boat drone, a hydrodynamic guided body or torpedo).

[0190] It can be an emergency vehicle, in particular a military vehicle (e.g., a tank, an infantry fighting vehicle, etc.) and / or a rescue vehicle (e.g., a fire engine, a police vehicle, an emergency medical vehicle, or at least partially automated and / or remote-controlled drones, perhaps with corresponding functionality). The method can be particularly advantageous for such vehicles (e.g., due to critical situations, decision-making pressure, dynamics, etc.). For example, the invention can help reduce negative and / or unintended effects or collateral consequences of actions (automated and / or performed by humans).

[0191] The driving situation can (e.g., in the cases described above) include or be a combat situation or an operational situation. In this case, other (in the vicinity or relevant) road users can be, for example, emergency personnel, soldiers, civilians, agents, etc., or their vehicles. These can be recognized as such in the process and / or taken into account (e.g., according to their respective roles or parameters).

[0192] The one or more road users described in this document (relevant to the driving situation) may (each) be typical users of a road traffic, in particular urban road traffic, or comparable to the (ego) vehicle (e.g. according to the described characteristics) or other, different vehicles of one or more of the types described above.

[0193] According to a further aspect of the invention, a computer program product (e.g., comprising one or more computer programs and / or specific data, in particular data of the neural network, data for setting up the neural network) is described. The computer program product, when executed on at least one computing unit of the (first and / or second) device, is configured to execute one or more steps of the (first and / or second) method according to one or more features described in the present document.

[0194] In particular, the computer program product comprises the data of the at least one neural network and / or data for setting up, in particular updating, and / or for training (e.g. execution of the training), in particular retraining and / or operating the neural network.

[0195] The computer program product can be designed as an update of a previous computer program, which, for example, includes the parts of the computer program or the corresponding program code as part of a functional extension, for example, as part of a so-called "remote software update." The computer program product comprises, in particular, a medium readable by the data processing device on which the program code is stored, or at least an encrypted file. The computer program product can also include or be an authorized access right to stored data of the computer program product.

[0196] Exemplary embodiments of the invention are explained in more detail below, without limitation of generality, with reference to the figures. They show: Fig. 1 illustrates examples of the method 100 for operating a vehicle EgoF. The focus is on the active operation of the vehicle EgoF. Fig. Figure 2 illustrates examples of the method 200 for operating the vehicle EgoF. The focus is on training 201, retraining 202, and / or operating 203 the at least one neural network DNN, GPT.

[0197] The solid arrows indicate a preferred, suggested sequence of steps. Furthermore, the arrows can also mean that the step pointed to by the arrow is executed depending on the result of the step from which the arrow begins. The dot-dash arrows each indicate an influence during an operating cycle, in particular a control of the execution of a step pointed to by the arrow, depending on the result of one or more steps (from a previous operating cycle) from which the arrow(s) originate. The method can, in particular, comprise only part of the illustrated sequence and / or also additional steps. Fig. 3 illustrates the vehicle EgoF and the device A for operating a vehicle EgoF or the device A for training 201, retraining 202 and / or operating 203 the neural network DNN, GPT. Fig. 4 illustrates an exemplary (current or predicted) driving situation relevant for the vehicle EgoF. Fig. Figure 5 illustrates a driving situation. In particular, it illustrates a possible, expected, or predicted (further) development, particularly the dynamics of the driving situation.

[0198] The (first) method 100 is started in a step (not explicitly shown). Data based on previous operating cycles is read in and considered.

[0199] The acquisition and / or prediction of the driving situation data 101 takes place depending on the data on the basis of one or more sensors of the vehicle EgoF, in particular KS (= camera or a camera system), RV (= front radar), LID (= LIDAR, in particular solid state lidar), RVL (= front left radar), RVR (= front right radar), RHR (= rear right radar, RHL (= rear left radar), of the vehicle EgoF.

[0200] Alternatively or additionally, data based on the data of a chassis sensor, steering wheel sensor, pedal sensor, and / or an interior sensor (e.g. for detecting or predicting an intention AB1, AB2, AB3 and / or decision E1, E2, E3 of the user N) are determined and taken into account.

[0201] Alternatively or additionally, the sensor data may also include data (e.g. object recognition data, classification data) based on these or similar sensors (e.g. a model of the environment UFM).

[0202] For example, previous driving situations (e.g., from the recent past) and / or a previous course of the driving situation (e.g., as a sequence, order, dynamics of individual states or changes) can be recorded and taken into account when determining and / or predicting one or more driving situations that at least potentially affect the vehicle EgoF. Additionally, in step 102, the driving situation data for the current driving situation can be determined and / or predicted for one or more driving situations that are possible in the near future (e.g., expected with at least certain probability measures) (e.g., variants of the further development of the current driving situation or one of these subsequent driving situations).

[0203] The driving situation data may identify a time-limited, adapted or (at least primarily) related to a section of the road, such as an intersection, a roundabout or an entrance or exit ramp, section of a driving situation (e.g. 1, 2, 5, 10, 20, 30 seconds).

[0204] Sensor data can be processed (e.g. aggregated among each other and / or brought into the form of a data model of the vehicle's environment EgoF) in order to determine the driving situation data.

[0205] The driving situation data can then be determined and / or predicted as an environment model, in particular on the basis of data from one or more sensors, in particular the result of the sensor fusion and / or a modeling of the environment of the vehicle EgoF.

[0206] For example, the (current and / or predicted) driving situation is classified (e.g., into at least 4, 8, 16, 32, 64, 128 classes) and / or the pattern (e.g., as similarity and / or dissimilarity in at least 4, 8, 16, 32, 64, 128 patterns) of the driving situation is determined. The result of the classification or pattern recognition can then (possibly after a preprocessing step) be used as input, in particular as a prompt or stimulus for the generative pre-trained transformer GPT to generate (possibly after further inputs) the evaluation information characterizing the respective legal and / or moral evaluation of the driving situation.

[0207] In a step 102, which can be carried out alternatively or in addition to step 101, a driving situation (which at least potentially concerns the vehicle (EgoF)) is predicted for the near future (e.g. for a few fractions of a second, or a few seconds).

[0208] The driving situation data is (also) generated for the predicted driving situation, in particular for several variants, especially initial scenarios of the driving situation. In this embodiment, the legal and / or moral assessment is based on a predicted future driving situation, i.e., a driving situation that will occur with a certain probability.

[0209] This makes it possible to assess the driving situation in advance and / or to react to the (likely) legal and / or moral consequences, e.g. to take a measure (e.g. a countermeasure) before these have occurred.

[0210] For example, the operation of the vehicle EgoF can be carried out depending on the evaluation information, in particular in such a way that from a legal and / or moral point of view, an improved, desirable outcome of the (current or possibly, more or less certainly imminent) driving situation is sought, in particular brought about.

[0211] In a further embodiment, the sensor information comprises an image, point cloud, in particular an image sequence or point cloud sequence that characterizes a part of the environment of the vehicle EgoF (e.g., representing it two-dimensionally or three-dimensionally).

[0212] In particular, a two-dimensional or three-dimensional or stereoscopic image is captured by a camera KS (e.g. to be understood as one camera or as a camera system, e.g. operated with images from several cameras) of the vehicle EgoF, which captures the surroundings or the parameters of the current driving situation in the visible spectrum and / or in the infrared spectrum.

[0213] Alternatively or additionally, a three-dimensional image of the surroundings of the vehicle EgoF can be captured, particularly in the form of a point cloud. A time-of-flight camera 108, a radar system, or a lidar of the vehicle EgoF can be used for this purpose. Furthermore, an image sequence consisting of consecutive images depicting the surroundings of the vehicle EgoF can be captured. In such an embodiment, the image data of the image sequence characterizes.

[0214] In a further embodiment, communication data is received from at least one other road user and / or from the traffic infrastructure. The communication data is processed to generate at least part of the driving situation data. Communication can occur, in particular, via Dedicated Short Range Communication (DSRC), for example, Car-2-Car or Car-2-X. For example, an intention can be received from the other road user, such as whether a vehicle ahead is braking. The traffic infrastructure can provide, for example, a context relating to the driving situation.

[0215] In step 104, the driving situation data are processed using a tokenizer, whereby a character set, in particular a character sequence, in particular a set or sequence of tokens corresponding to the driving situation can be generated. The elements of the character set or token can be understood as an information element represented in a suitable manner.

[0216] During transformation 104, a character set, preferably a sequence, in particular a stream of tokens, is generated from a sequence, in particular a stream of driving situation data by means of a corresponding tokenizer, which can then (very advantageously) be provided for operating 203 the at least one neural network DNN, in particular the generative pre-trained transformer GPT, or can be read in by it.

[0217] Preferably, the neural network DNN, GPT can comprise a so-called LLM (= Large Language Model), be or be set up or retrained 202 on the basis of such a model and / or be controlled by such a model.

[0218] Preferably, driving situation data (e.g. image data, image sequence data, vector data, point clouds, coordinate data, etc.) can be transformed into a character string comprising text characters (e.g. corresponding to or based on natural language).

[0219] In a further embodiment, the image data is processed using the generative pre-trained transformer GPT or another generative pre-trained transformer operating as an image-to-text transformer to generate or transform at least a portion of the driving situation data.

[0220] Based on the data from the vehicle sensors (e.g. image data, infrared data, LIDAR data, RADAR data), a character set (also, character string, character sequence) is generated as a token and / or as a text in (natural, formalized, coded) language. The character set can characterize the (e.g. semantic) content (also: content, description of the driving situation, the course of the driving situation, the initial scenarios of the driving situation, their consequences) and / or dynamics (e.g. change over time) of the driving situation data (e.g. the images, image sequences).

[0221] In a corresponding manner, the driving situation data that characterize one or more predicted driving situations can (also) be processed.

[0222] For example, one or more driving situations can be predicted and the corresponding driving situation data characterizing them can be determined or (e.g. in the same step of the method) driving situation data for a driving situation in the near future can be predicted.

[0223] The data transformed (in this way) (the character set, in particular the character string) can then be provided for operation 203 (e.g., controlling, stimulating, prompting) of the at least one neural network DNN, in particular of the generative pre-trained transformer GPT, in particular as input data, control signal, input (prompt, stimulus) and / or read in by it.

[0224] In step 104 (corresponding to step 203), the evaluation information is generated. For example, the evaluation information can indicate a probable distribution (e.g., shares or proportions) of the legal and / or moral guilt or liability between the vehicle EgoF and one or more other road users (e.g., involved or potentially, probably, predicted to be involved).

[0225] The assessment information may identify these (various, specific legal and / or moral aspects) from the perspective of or on behalf of the user and / or operator of the vehicle and / or one or more (potentially or likely affected) road users, in particular in a differentiated form.

[0226] For example, the evaluation information may identify or take into account qualitative and / or quantitative measures (e.g. more than 2, 4, 8, 16, 32, 64 gradations) and / or coded information, symbolically represented information, (possibly descriptive) text information (e.g. with reference to concrete reasons, traffic rules, judgments, etc.), an evaluation (generated by the neural network).

[0227] A distinction can be made between legal, moral, ethical aspects and / or predicated opinions (also: attitudes) of a person or group of people.

[0228] One or more steps of the method 100, 200 can be applied primarily with regard to driving situations, consequences, or cases that have a comparatively high complexity (exceeding a certain threshold) and / or legal and / or moral aspects that cannot be described by clearly formulatable rules.

[0229] In simple, manageable cases (e.g., through the application of rules), complex or comparatively lengthy processing using DNN or GPT neural networks can be dispensed with. Instead, the steps of the procedure, e.g., generating and / or considering the evaluation information, are applied to driving situations, consequences, or cases that tend to or potentially overwhelm a vehicle functionality or the user of the vehicle (e.g., for the rented vehicle as the user of the vehicle).

[0230] In particular, the legal and / or moral consequences may involve costs or liability and / or (legal and / or moral) guilt, an ethical problem in one or more variants, initial scenarios or their respective possible consequences.

[0231] For example, the assessment information may be indicative of a predicted (also to be understood as simulated and / or anticipated) legal and / or moral opinion or information derived from an opinion.

[0232] An expert opinion or information dependent on the expert opinion may include or be a derivation, justification, consideration and / or conclusion of an expert opinion or recommendation on this basis.

[0233] The evaluation information, which can particularly characterize a part of such an assessment, can be stored in a coded and / or encrypted form and / or together with the relevant data. For example, the evaluation information can be stored together with or associated with a corresponding time value, position value, characteristics, and / or identification of one or more participants in the driving situation.

[0234] Alternatively or additionally, an expert opinion can be generated, particularly in a text-like form, particularly written or representable in text form, particularly for use after the driving situation, in particular characterising (e.g. pre-estimated) moral and / or legal consequences of the driving situation.

[0235] Such an expert opinion, in particular a derivation, justification, assessment, conclusion and / or recommendation (e.g. understandable for a layperson), can be generated in relation to the driving situation as a whole (e.g. from a neutral moral and / or legal point of view) and / or in relation to the ego vehicle (e.g. from a moral and / or legal point of view or representation of the interests of the user or operator of the vehicle) and / or in relation to one or more road users relevant to the traffic situation (e.g. from the moral and / or legal point of view of one or more road users, in particular those at risk, those injured or those posing a threat, those causing damage, etc.).

[0236] The evaluation information or the information output to the user as text, symbol, or speech, depending on the evaluation information, may, for example, (also) refer to a recommendation for action and / or instructions regarding (possible) consequences of the driving situation or consequences of the driving situation that have already occurred. The evaluation information and, if applicable, the corresponding output and / or measure may refer to one or more (e.g., specific) consequences of the driving situation (e.g., overall, for the vehicle, and / or one or more specific road users), take these into account, and in particular, identify them.

[0237] In step 105, (respective) evaluations are processed with respect to several driving situations that at least potentially concern the vehicle EgoF, including in particular: variants predicted with respective probability measures, initial scenarios and / or consequences of the driving situations.

[0238] These can be several evaluations that are alternative to one another (e.g. concerning alternative variants or initial scenarios) and / or evaluations from the perspective of the vehicle EgoF, user N of the vehicle EgoF and / or different road users VT1, VT2, VT3, VT4, VT5, VT6.

[0239] In one example, step 105 may include or be a comparison or weighting (e.g., variants according to a predetermined mathematical relationship). In this case, the operation 106 of the vehicle EgoF may be dependent on the corresponding comparison of the respective measures of the legal and / or moral consequences (e.g., with certain standard values, threshold values, and / or with each other). For example, depending on the result of processing 105, in particular the comparison, a decision may be made for one of several possible variants and / or initial scenarios of the driving situation and / or an influence to be exerted on the driving situation (e.g., a measure). In the selection, the respective legal and / or moral consequences

[0240] In step 106, the operation of the vehicle EgoF is carried out (also: controlled, influenced) depending on the evaluation information (also to be understood as information derived from the evaluation data).

[0241] For example, a decision is made between (e.g., mutually alternative) driving situations, variants, or the initial scenario of a driving situation and / or an influence on the driving situation (e.g., the further course of the driving situation) depending on the respective evaluation information. By operating 106 the vehicle, a better or improved variant can be selected and sought, in particular, achieved.

[0242] The operation 106 of the vehicle EgoF can include striving (also: selecting, causing), avoiding (also: avoiding) and / or influencing the course of the driving situation.

[0243] Alternatively or additionally, depending on the evaluation information, at least one output unit of the EgoF vehicle can be controlled, whereby the user of the EgoF vehicle is provided with user information, in particular driver information, dependent on the evaluation of one or more driving situations, in particular alternative variants or initial scenarios of the driving situations and their respective consequences. The output can be provided in particular visually (e.g., as a symbol), via a display unit of the EgoF vehicle, or audibly, for example, as (symbolic or warning) audible messages and / or a generated voice output.

[0244] In a further embodiment, the control of functional unit 119 based on the evaluation information can be at least partially modified, suppressed, interrupted, and / or aborted by a corresponding input from the user of the vehicle EgoF. For example, the vehicle EgoF can be braked and / or evaded to avoid negative (exceeding a certain threshold) legal and / or moral consequences (e.g., to a certain extent or direction, parameter).

[0245] In a further example that can be combined with all examples described in this document, the method can comprise changing, suppressing, interrupting and / or canceling the consideration of one or more (e.g. certain) legal and / or moral consequences and / or a variant of the operation of the vehicle EgoF, in particular the operation of a vehicle functionality, in particular one or more performance features, by the user, in particular with a particularly confirming, controlling, prescribing operating action, and / or non-activation of a rejecting operating action.

[0246] The user N can decide (e.g. in the specific case, on a case-by-case basis) to take control of the corresponding vehicle functionality of the vehicle EgoF (e.g. if he is sure, wants to act differently or a malfunction or incorrect assessment is suspected) or to give it up.

[0247] In a further example, the operation 106 of the vehicle EgoF is or includes operating, in particular controlling, an at least partially assisted, at least partially automated or autonomous and / or at least partially and / or at least temporarily remote-controlled driving and / or parking of the vehicle EgoF. The term "parking" is to be understood in particular as maneuvering, parking, and reversing out of a parking space within the scope of this document.

[0248] The Fig. Figure 4 shows, among other things, an example driving situation. It illustrates a dilemma. It involves the vehicle EgoF and several road users VT1, VT2, VT3, VT4, VT5, VT6. Of course, it could also be a much more complex case. In particular, the resolution of non-trivial or complex cases would be impossible to describe or specify with practical rules.

[0249] In the present case, (at least) three pieces of evaluation information can be generated 104. These include (at least) the evaluation of the legal and / or moral consequences that would (presumably) result from different operating actions BA1, BA2, BA3 of user B.

[0250] For example, during processing 105, a measure to resolve a dilemma may be determined. Specifically, a dilemma is resolved between: - a legal consequence and a moral consequence; - a legal consequence and a material consequence (e.g. a material asset); - a moral consequence and a material consequence (e.g. a tangible asset); - a legal and moral consequence on the one hand and a consequence related to the material (e.g. a material asset) on the other hand.

[0251] This may involve a dilemma arising within the context of the driving situation or as a result of the driving situation or the movement, action, or behavior of one or more (other) road users VT1, VT2, VT3. For example, a measure (e.g., regarding braking, steering, especially evasive maneuvering, or a combination thereof) that results in (moderate) property damage can only be initiated if a (e.g., legally entirely or relatively uncritical) but morally or ethically significant consequence is imminent.

[0252] The Fig. Figure 5 illustrates a driving situation. In particular, a possible, expected, or predicted (further) development, particularly the dynamics of the driving situation at different points in time, is illustrated. Time t0 roughly corresponds to the current driving situation or the corresponding data, time t1 is in the near future (e.g., from 1, 2, 3, 5 seconds), and time t2 is in the near future (e.g., from 5, 10, 20 seconds). The method can consider driving situation data that characterize a (possible, expected, predicted) driving situation at such points in time or time intervals t1, t2.

[0253] Assisted driving, for example, refers to automation levels 1 and 2, as defined by or based on the Federal Highway Research Institute (BASt), which are generally familiar to experts. Automated driving, for example, refers to automation levels 2, 2+, and 3, as defined by or based on the Federal Highway Research Institute (BASt). Autonomous driving, for example, refers to automation levels 4 and 5, as defined by the BASt.

[0254] The vehicle functionalities of the EgoF vehicle can be driver assistance systems or combinations thereof and / or active safety functionalities or combinations thereof. For example, it can be a functionality comprising or relating to fully or partially automated longitudinal guidance, lateral guidance, maneuver execution, active collision protection and / or pedestrian protection, fully or partially automated evasive maneuvering, and / or user information, in particular driver information.

[0255] The second method 200, which can be freely combined with the first method 100, is described below as an example.

[0256] The method 200 may include training 201, in particular retraining 202 and / or operating 203, the at least one neural network DNN, GPT. This may correspond to the first method 100, so to speak, from the perspective of the neural network DNN, GPT.

[0257] The training 201, in particular retraining 202 and / or operation 203 of the at least one neural network DNN, GPT can be specific, in particular adapted to specific or specific features corresponding to: - EgoF vehicles; - Users of the EgoF vehicle; - countries, societies, legal systems, cultures; and / or - Law and / or morality, especially ethics, that is typical and / or specific to countries, societies, legal systems, cultures, and / or people. This can include, in particular, subjective ideas about law and / or morality.

[0258] Retraining 202 of the neural network DNN, GPT, in particular based on a neural network (e.g., DNN', GPT') that has been set up (developed or pre-trained) for several (multiple) vehicles, is particularly advantageous. This allows for specific evaluations or operation with regard to legal, moral, or ethical aspects for the (respective) vehicle EgoF, user N of the vehicle EgoF, countries or locations, societies, legal systems, cultural circles, and / or groups of people. A specific "conscience" of the vehicle EgoF can be set up, trained 201, retrained 202, and / or operated 203.

[0259] Alternatively or additionally, the at least one neural network DNN, GPT can be retrained 202, in particular adapted, to (personal), in particular adjustable and / or recognized, preferences of the user N, in particular driver and / or operator (e.g. taxi company, transport company, emergency service, etc.) of the vehicle (which can be made adjustable, for example).

[0260] For example, an attitude of the user N and / or operator of the vehicle EgoF regarding a desired weighting of one or more of the following aspects (e.g. in relation to each other) can be taken into account: - one or more moral aspects; and / or - one or more legal aspects; and / or - one or more material aspects.

[0261] This can be taken into account during training 201, retraining 202 and / or operation 203 of the at least one neural network DNN, GPT.

[0262] The one or more settings (described above) can be made adjustable and / or taken into account in particular with regard to the (own) vehicle (EgoF = ego vehicle) and / or other road users VT1, VT2, VT3, VT4, VT5, VT6, in particular road users VT1, VT2, VT3, VT4, VT5, VT6 having certain characteristics and / or road users VT1, VT2, VT3, VT4, VT5, VT6 of certain classes.

[0263] For example, a specific (e.g. increased) weighting of possible legal and / or moral consequences in relation to certain or certain classes and / or characteristics corresponding road users VT1, VT2, VT3, VT4, VT5, VT6 (e.g. children, elderly and / or disabled persons, wheelchair users, motorcyclists, cyclists with or without helmets) and / or animals can be made adjustable and / or taken into account accordingly in the procedure.

[0264] In this case, an at least partially automated functionality of the vehicle EgoF or the at least partially automated drivable and / or parkable vehicle EgoF can be operated, driven, parked or act (at least approximately) according to the criteria and / or weighting between different legal and / or moral and / or material consequences that can be set by the user N and / or operator of the vehicle EgoF.

[0265] The at least one neural network (DNN, GPT) can have, be, or include one or more features of a convolutional neural network (CNN). A CNN is also understood to mean, in particular, a convolutional neural network (CNN), a convolutional neural network (CNN), a convolutional neural network (CNN), or a convolutional neural network (CNN). For example, the at least one neural network (DNN, GPT) can include one or more convolutional layers (also known as convolutional layers).

[0266] During training 201, the kernel of the matrix can be learned. The kernel can be smaller than the input parameters (e.g., with regard to dimensions). This kernel can be used for multiple cases, vehicle types, user types, driving situations, various conditions described in this document, etc.

[0267] The at least one neural network DNN, GPT can (e.g. during pre-training 201) recognize or learn influences, in particular structures of influences, that occur independently of the case, regardless of location, and / or independently of one or more concrete cases (e.g. driving situations, precedents, legal and / or moral consequences, corresponding parameters).

[0268] During operation 203, these can be used, for example, for new cases or circumstances. This can be carried out using a locally meshed layer (e.g., one set up for this purpose), in particular the convolutional layer.

[0269] Furthermore, the information on the influences can be condensed into a matrix or transferred into a more abstract representation. In this case, the strongest features of a matrix can be prioritized or strengthened, while the weaker ones can be discarded. This can be achieved using a locally meshed layer (e.g., a specifically designed pooling layer or subsampling layer). This can lead to a reduction in the amount of data requiring further processing.

[0270] Furthermore, the results from the two aforementioned layers of the aforementioned types can be processed together, in particular, linked appropriately. The incoming and / or outgoing values ​​from both layers can be processed together, in particular, offset. This can be performed in the third layer (or in the third step within the neural network DNN, GPT). This can be an at least predominantly, in particular fully meshed layer of the at least one neural network DNN, GPT.

[0271] Each of these optional features (e.g., of specific layers, the matrix of the topology, etc.) of the one or more neural networks DNN, GPT can be configured, dimensioned and / or optimized (at least to a certain extent) for executing at least a (corresponding) part of the method.

[0272] The method 200 can be operated depending on the data generated in one or more steps of the method 100. In particular, the retraining 202 of the at least one neural network can include DNN, GPT. Steps 101', 102', 103', 104', 105', 106' of the method 100 in one or more further operating cycles can be influenced, in particular optimized, depending on the steps of the method 200.

[0273] The method 100 can be carried out together and / or in an information technology connection with a method for operating the vehicle EgoF, in particular a method for at least partially automated driving and / or parking. This is preferably a separate method 100. In this case, it can be ensured (even better) that an assessment of the (legal and / or moral) consequences is carried out partially, in particular predominantly, independently or separately from the method for automated driving and / or parking of the vehicle EgoF (e.g., the method for automated driving and / or parking of the vehicle EgoF which is also aimed or trained towards other goals). In particular, a different neural network DNN, GPT or partially or predominantly a different sub-layer of the neural network DN, GPT can be provided or operated for the method 100, 200 than for the automated driving and / or parking of the vehicle EgoF. This enables a better result, quality orThe quality of the assessment information can be improved. Furthermore, risks associated with the use of artificial intelligence can be further reduced.

[0274] The invention further relates to a Fig. 3 illustrates a device A for operating 106 a vehicle EgoF. The device A comprises a determination unit configured to determine or predict a driving situation that at least potentially affects the vehicle EgoF and to determine (also: generate) driving situation data characterizing the driving situation.

[0275] Furthermore, the device A can comprise a processing unit ECU configured to process the driving situation data using at least one neural network DNN, GPT, in particular the generative pre-trained transformer GPT, in order to generate evaluation information 104 corresponding to a legal and / or moral evaluation of the driving situation or its consequences. The device A further comprises a control unit ECU configured to operate the vehicle EgoF (e.g., to influence, to control) in particular at least one functional unit of the vehicle EgoF, depending on the evaluation information.

[0276] The device A can have features and / or advantages corresponding to the method 100, 200. In particular, the device 100 can be further developed with the features of the dependent claims directed to the method 100, 200.

[0277] Furthermore, the method 100, 200 described above may each have features and / or advantages corresponding to the device A, which are described in this document in connection with the device A.

[0278] The device A is, for example, designed, among other things, to evaluate a (possibly only potentially possible or probable) driving situation, in particular the consequence of the driving situation for the vehicle EgoF and / or user N of the vehicle EgoF with regard to its legal and / or moral consequences (e.g. aspects, significance) or to take the evaluation into account.

[0279] The device A is designed to stimulate a generative process 104 in which the at least one neural network DNN, GPT (located in the vehicle EgoF or outside the vehicle EgoF) generates 104 the evaluation information with regard to the predicted action (e.g., act), in particular in the context of the driving situation. In particular, this can be done at least partially remotely.

[0280] The assessment information may take into account one or more (different) legal and / or moral consequences of the driving situation that at least potentially affects the vehicle, in particular (e.g. explicitly and / or structured according to certain criteria) identifying them.

[0281] The evaluation information may (simultaneously) include or be action information, in particular control information for effecting the action. This may represent a measure to be effected, in particular, it may identify a (e.g., optimized) time, intensity, direction, or a precondition under which the measure is to be executed, suspended, and / or aborted.

[0282] Depending on the legal and / or moral evaluations of the respective evaluation information, at least one functional unit ECU of the vehicle EgoF is then operated 106, in particular influenced or controlled.

[0283] The functional unit ECU can in particular be a unit or system of the vehicle EgoF comprising an output unit, with the aid of which a user N of the vehicle EgoF is informed depending on the evaluations (e.g. in the form of user information).

[0284] Preferably, the operation 106 of the vehicle EgoF comprises influencing, in particular controlling and / or regulating the movement, in particular the longitudinal guidance, lateral guidance, maneuver planning, and / or maneuver execution (e.g., trajectory planning, steering, braking, evasive maneuvers, parking maneuvers, shunting maneuvers, etc.). This can be carried out by controlling the functional unit ECU or an interface within the functional unit ECU depending on the output layer of the at least one neural network DNN, GPT.

[0285] Device A may comprise: - Sensors KS (camera system), RV (front radar), LID (LIDAR), RVL (front left radar), RVR (front right radar), RHR (rear right radar), RHL (rear left radar) of the vehicle EgoF; and / or - a unit (e.g. hardware, processor, software module, functional module) for executing a sensor fusion SF;

[0286] In the Fig. In the embodiment shown in Figure 3, the device A is designed to process the sensor data in order to determine the spatial arrangement and / or movement of the vehicle EgoF and / or the road users VT1, VT2, VT3, VT4, VT5, VT6 and / or other (e.g., moving or immovable) objects O and to generate driving situation data corresponding to the driving situation.

[0287] In particular, the driving situation data can identify a predicted driving situation (which has not yet begun, occurred or ended).

[0288] To process the driving situation data, in particular to transform it 104, the acquisition unit ECU uses, in particular, a tokenizer that generates a sequence of tokens from a data sequence based on sensor data, which can be processed by at least one neural network (DNN, GPT). The character string, in particular the sequence of tokens, represents at least a portion of the driving situation data (in particular, also to be understood as: a portion, e.g., the most important portion, of the information characterizing the driving situation relating to the vehicle EgoF). This can be generated and / or processed step by step (e.g., in at least 4, 8, 16, 32, 64 steps).

[0289] Alternatively or additionally, the sensor data can also be processed with the aid of the at least one neural network DNN, GPT or at least one further neural network DNN, GPT to generate the driving situation data in the form of a character set, in particular the character sequence (transforming 104 into the character set, in particular the character sequence), for example in a text-like form (such as a natural or formalized language). The transformation 103 can be carried out using an image-to-text transformer.

[0290] In particular, the acquisition unit 120 can be designed to process the image data, in particular image sequence data, with the aid of the at least one neural network DNN, GPT or a neural network DNN, GPT (e.g., a further generative pre-trained transformer GPT) operated as an image-to-text transformer or video-to-text transformer. As a result, the acquisition unit ECU can generate a character set, in particular text, from the image data, which reproduces the, in particular semantic, content of the image sequence. This text can form at least part of the driving situation data.

[0291] The device A can further comprise a processing unit ECU which is designed to train 201, train 202 and / or operate 203 at least one neural network DNN, GPT in the vehicle EgoF, in a unit carried along or in a computing unit arranged at a distance (e.g. backend or cloud), which can also belong to the device A. The unit ECU can further be designed to operate 106 (e.g. continuously, event-controlled, cyclically) one or more functionalities of the vehicle EgoF of the vehicle EgoF depending on driving situation data and / or evaluation information.

[0292] The ECU unit further comprises a communication interface for communicating with other road users or traffic infrastructure, such as traffic lights. The ECU unit can receive communication data from the other road users or the traffic infrastructure and incorporate the data into the process. For example, the ECU unit can use the communication data in the EgoF prediction of the driving situation. In particular, the communication data can also be part of the driving situation data, or the driving situation data can be determined and / or predicted based on the communication data.

[0293] For example, device A is configured to control an output unit of the vehicle EgoF in order to generate and / or output user information dependent on the legal and / or moral assessment (e.g., the extent of the legal and / or moral consequences). The output may include or be, for example, text and / or graphics (e.g., a symbol or pictogram or their sequence) and / or animation and / or sound and / or voice output. This can advise the user N against or not against certain decisions E1, E2, E3 and / or confirm or recommend them.

[0294] These can form part of the generated evaluation information or have been generated by the generative pre-trained transformer GPT as at least part of the evaluation information.

[0295] However, the output to the user N can also be auditory, for example as a computer-generated speech output generated by the at least one neural network DNN, GPT, in particular as part of the evaluation data or the information dependent on the evaluation data.

[0296] Alternatively or additionally, in step 106, a unit ECU related to the movement of the vehicle EgoF is controlled to plan a maneuver, to change maneuver planning (e.g., to change a type and / or parameters of the maneuver), in particular to reduce the (otherwise expected) legal and / or moral consequences, to carry out a (specific, selected, adapted) maneuver and / or to adapt the trajectory of the vehicle EgoF depending on the evaluation information.

[0297] The device A may further be designed to carry out any selection of the Fig. 1 and Fig.2 and to carry out the process steps identified as examples in the previous description. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This 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

[0000] ISBN 978-3-406-69610-7

[0146] ISBN 3-8329-1281-9

[0146] ISBN 3-574-27311-8

[0146] ISBN 3-423-50633-4

[0146]

Claims

[1] Method (100) for operating a vehicle (EgoF), comprising: - determining (101) the driving situation data which are characteristic of a driving situation which at least potentially concerns the vehicle (EgoF); - operating (203) a neural network (DNN, GPT) depending on the data on the basis of the driving situation data, wherein evaluation information is generated (104) which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation; - Operating (106) the vehicle (EgoF) depending on the evaluation information. [2] Method (100) according to claim 1, comprising, the determining (101) of driving situation data is or comprises: - predicting (102) one or more driving situations and determining the corresponding driving situation data for the one or more predicted driving situations; and / or - Predicting (102) driving situation data that would characterize one or more driving situations that at least potentially affect the vehicle (EgoF). [3] Method (100) according to claim 1 or 2, comprising - determining (101), in particular predicting (102), driving situation data for several driving situations at least potentially concerning the vehicle (EgoF), in particular several variants or initial scenarios of a current or predicted driving situation; and: - generating (104) the plurality of evaluation information items corresponding to the different driving situations; and: - processing (105), in particular comparing (105) the plurality of evaluation information; and: - taking into account the result of the processing (105) in the method (100, 200). [4] Method (100, 200) according to claim 3, wherein the legal and / or moral consequences of a plurality of driving situations corresponding to the plurality of driving situation data with respect to: - the vehicle (EgoF); and / or - the user (N) of the vehicle (EgoF); and / or one or more different road users (VT1, VT2, VT3, VT4, VT5, VT6) in the surroundings of the vehicle (EgoF); in particular as respective results of respective actions of one or more participants in the driving situation, in particular as actions of one or more different types and / or extents, taken into account, in particular identified. [5] Method (100, 200) according to one of the preceding claims, wherein the at least one neural network (DNN, GPT): - a Generative Pre-trained Transformer (GPT); and / or - a generative, in particular statistical model, in particular a model trained on a specific type of vehicle (EgoF), the specific vehicle (EgoF), a specific user and / or operator of the vehicle (EgoF); comprises, is constructed on the basis of such a model and / or is controlled by such a model. [6] Method (100, 200) according to one of the preceding claims, wherein the at least one neural network (DNN, GPT) is trained (201), in particular retrained (202), depending on data based on: - legal information, in particular concerning traffic law, laws, interpretations of laws, regulations, court rulings; and / or - information relating to morality, in particular ethics, in particular specialist literature, decisions or opinions of commissions, in particular concerning moral theses and / or principles and / or precedents; and / or - information related to subjective opinions, in particular media contributions, particularly concerning the formation of opinions by several people and / or evaluations. [7] Method (100) according to one of the preceding claims, wherein the operation (106) of the vehicle (EgoF) comprises triggering, changing and / or at least partially and / or temporarily suppressing a triggering of a functionality of the vehicle (EgoF). [8] Method (100, 200) according to one of the preceding claims, wherein the at least one neural network (DNN, GPT) is trained (201), in particular retrained (202), depending on: - initial data that predominantly or entirely objectively characterise one or more driving situations from the past, in particular in the form of protocols, witness statements, driving situation data, data based on sensors; and - second data which characterise one or more at least partially, predominantly or entirely subjective legal and / or moral assessments, in particular one or more points of view, opinions, claims, media reports, judgments on the one or more driving situations from the past. [9] Method (100, 200) according to one of the preceding claims, comprising: - transforming (103) the driving situation data into a character set, in particular tokens, in particular comprising tokens representing text information or voice information, wherein the character set characterizes the driving situation, in particular the course and / or one or more consequences of the driving situation; and - training (201), retraining (202) and / or operating (203) the at least one neural network (DNN, GPT) depending on the character set. [10] Method (100, 200) according to one of the preceding claims, wherein the at least one neural network (DNN, GPT) is trained (201), in particular retrained (202) or is to generate evaluation information which predicts and / or anticipates and / or simulates an evaluation which a person, in particular a group of people, would have with regard to one or more driving situations, in particular to one or more possible, expected or predicted consequences of the driving situation. [11] Method (100, 200) for operating a vehicle (EgoF), comprising: Setting up, training (201), in particular retraining (202) a neural network (DNN, GPT) for operating (106) a vehicle (EgoF) on the basis of driving situation data which are characteristic of a driving situation which at least potentially concerns the vehicle (EgoF) in order to generate evaluation information which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation. [12] Neural network (DNN, GPT) for operating a vehicle (EgoF), wherein the neural network (DNN, GPT) is trained (201) and / or retrainable (202) and / or operable (203) depending on data on the basis of driving situation data which are characteristic of a driving situation which at least potentially concerns the vehicle (EgoF), wherein the neural network (DNN GPT) generates (104) evaluation information which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation. [13] Device (A), wherein the device (A) is designed to determine (101) the driving situation data characterizing a driving situation at least potentially concerning a vehicle (EgoF); and is designed: to operate (203) a neural network (DNN, GPT) depending on data on the basis of the driving situation data, wherein evaluation information is generated (104) which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation; and is further designed: to operate (106) the vehicle (EgoF) depending on the evaluation information. [14] Device (A) for operating (106) a vehicle (EgoF), wherein the device (A) is designed to set up, train (201), in particular retrain (202) a neural network (DNN, GPT) on the basis of driving situation data which are characteristic of a driving situation which at least potentially concerns the vehicle (EgoF), to generate evaluation information which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation. [15] Vehicle (EgoF), wherein the vehicle (EgoF) is designed: to determine (101) the driving situation data characterizing a driving situation at least potentially concerning a vehicle (EgoF); and is designed to: train (201), retrain (202) and / or operate (203) a neural network (DNN, GPT), wherein evaluation information is generated (104) which takes into account, in particular characterizes, a legal and / or moral evaluation of the driving situation; and is further designed: to operate (106) the vehicle (EgoF) depending on the evaluation information. [16] Computer program product, in particular comprising the data of the at least one neural network (DNN, GPT), data for setting up, training (201), in particular retraining (202) and / or operating (203) the neural network (DNN, GPT), wherein the computer program product is designed, when executed on at least one computing unit, in particular a computing unit of the device according to claim 13 or 14 or of the vehicle according to claim 15, to carry out one or more steps of the method (100, 200) according to one of claims 1 to 11.

Images