EP4320408A1 - Computerimplementiertes verfahren, computerprogramm und anordnung zum vorhersagen und planen von trajektorien - Google Patents
Computerimplementiertes verfahren, computerprogramm und anordnung zum vorhersagen und planen von trajektorienInfo
- Publication number
- EP4320408A1 EP4320408A1 EP22720428.6A EP22720428A EP4320408A1 EP 4320408 A1 EP4320408 A1 EP 4320408A1 EP 22720428 A EP22720428 A EP 22720428A EP 4320408 A1 EP4320408 A1 EP 4320408A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- embedding
- stat
- features
- tensor
- static
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/38—Electronic maps specially adapted for navigation; Updating thereof
- G01C21/3863—Structures of map data
- G01C21/387—Organisation of map data, e.g. version management or database structures
- G01C21/3878—Hierarchical structures, e.g. layering
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W10/00—Conjoint control of vehicle sub-units of different type or different function
- B60W10/04—Conjoint control of vehicle sub-units of different type or different function including control of propulsion units
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W10/00—Conjoint control of vehicle sub-units of different type or different function
- B60W10/18—Conjoint control of vehicle sub-units of different type or different function including control of braking systems
- B60W10/184—Conjoint control of vehicle sub-units of different type or different function including control of braking systems with wheel brakes
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W10/00—Conjoint control of vehicle sub-units of different type or different function
- B60W10/20—Conjoint control of vehicle sub-units of different type or different function including control of steering systems
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
- B60W30/08—Active safety systems predicting or avoiding probable or impending collision or attempting to minimise its consequences
- B60W30/095—Predicting travel path or likelihood of collision
- B60W30/0956—Predicting travel path or likelihood of collision the prediction being responsive to traffic or environmental parameters
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W50/0097—Predicting future conditions
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
- B60W60/0015—Planning or execution of driving tasks specially adapted for safety
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
- B60W60/0027—Planning or execution of driving tasks using trajectory prediction for other traffic participants
- B60W60/00272—Planning or execution of driving tasks using trajectory prediction for other traffic participants relying on extrapolation of current movement
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
- B60W60/0027—Planning or execution of driving tasks using trajectory prediction for other traffic participants
- B60W60/00276—Planning or execution of driving tasks using trajectory prediction for other traffic participants for two or more other traffic participants
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
- G08G1/0145—Measuring and analyzing of parameters relative to traffic conditions for specific applications for active traffic flow control
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2554/00—Input parameters relating to objects
- B60W2554/20—Static objects
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2554/00—Input parameters relating to objects
- B60W2554/40—Dynamic objects, e.g. animals, windblown objects
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2556/00—Input parameters relating to data
- B60W2556/10—Historical data
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Definitions
- the invention relates to a computer-implemented method, a computer program and an arrangement for predicting and planning trajectories.
- German patent application with the file number 10 2020210 379.8 discloses a hybrid scene representation that models interactions between static and dynamic objects and/or information.
- claims 1, 13 and 14 each solve this problem in that the interaction modeling of road users comprehensively automated driving systems over the entire traffic area and over a predetermined time in the past with the merging of the history of the road users with all static and dynamic Sharing the scene allows prediction of all road users for a specific time in the future.
- One aspect of the invention relates to a computer-implemented method for predicting and planning trajectories.
- the procedure includes the steps
- a first machine learning model which receives the hybrid scene representation as input and is trained or is trained using reference predictions conditions to determine interactions between the static and dynamic environment features, with a function of the first machine learning model on the first layer, the applying the second layer and the third layer and creating an embedding of the rigid static environment features, the state-changing static environment features, and the dynamic environment features, and the embedding being output from the machine learning model; • Determining road user-specific interactions from the common embedding and fusing these with the road user-specific dynamic environment features for each of the road users and obtaining a road user-specific embedding for each of the road users;
- the computer program includes instructions that cause a computer, including a computer of a control unit of a driving system for automated driving functions, to carry out the steps of the method according to the invention when the computer program runs on the computer.
- a further aspect of the invention relates to an arrangement for predicting and planning trajectories.
- the arrangement includes
- an RNN encoder that encodes dynamic environment features including trajectory histories of road users based on real driving data and/or virtual data
- an interaction tensor pooling module that merges the encodings of the RNN and CNN encoders and a hybrid scene representation comprising at least a first layer comprising the rigid static environment features, a second layer comprising the state-changing static environment features and a third layer comprising dynamic environment features comprising the trajectory histories generated;
- a CNN interaction network which determines interactions between the static, dynamic and between the static and dynamic environment features based on the hybrid scene representation, with the CNN interaction network providing a first tensor embedding of the rigid static generating environment features, a second tensor embedding of the state-changing static environment features, and a third tensor embedding of the dynamic environment features, and merging the first, second, and third tensor embeddings into a multi-agent scene tensor;
- an interaction vector extraction module that extracts per traffic participant the features of the multi-agent scene tensor at the location corresponding to the traffic participant's coordinates, merges them with the third tensor embedding of the traffic participant and produces a multi per traffic participant and per scene -Generated agent scene embed;
- an RNN probability decoder that decodes the multi-agent scene embedding and compares the probabilities of the predicted trajectories for each road user and outputs a corresponding value.
- Road users include interactors in scenes of an environment, for example in scenes of a traffic area.
- Road users are, for example, people, such as pedestrians, vehicles, driving systems, and cyclists.
- Driving systems include automated driving systems from automated to autonomous vehicles, road vehicles, people movers, shuttles, robots and drones.
- Road users also include intelligent agents, such as self-driving vehicles or robots.
- Spatial merging means, for example, that spatial coordinates of road users and/or environmental features are represented in pixels of the hybrid scene representation, with one pixel in each of the layers of the hybrid scene representation corresponding to the same route length.
- the environmental features are in pixels Layers and / or represented by feature vectors with spatial anchor points or relative to a reference point.
- the feature vectors have a predetermined spatial anchor point.
- the environmental features are interpreted as pixel values.
- a spatial position of the environmental features is recorded in each layer via a corresponding position on a map. This is advantageous for a spatially corresponding arrangement of the environmental features.
- Surrounding features include houses, streets, in particular street geometry and/or the condition, signs, lane markings, vegetation, mobile road users, vehicles, pedestrians, cyclists.
- Static environmental characteristics are divided into two further categories. Elements that do not change at all or only after a long period of time do not change their status in the short term and are referred to as rigid.
- the rigid static surrounding features are arranged in the first layer.
- the state-changing rigid environment features are placed in the second layer.
- Dynamic environment features affect the moving road users in a scene.
- the coordinates of the road users are used comprehensively positions and/or orientations over a certain period of time in order to generate trajectory histories.
- parameters of a vehicle dynamics or motion dynamics model are used here, for example using a Kalman filter.
- the road users are arranged spatially based on the last coordinate and form the third layer of the dynamic environment features.
- the hybrid scene representation layers a scenario into several layers.
- a real scenario is presented as a hybrid of static and dynamic information.
- the environmental scene according to the invention Representation also called Hybrid Scene Representation for Prediction, abbreviated HSRV.
- the scenario is, for example, an image with i pixels in the x-direction and j pixels in the y-direction, in which the spatial coordinates of the road users are represented in pixels.
- the individual layers can also be represented as images and are arranged congruently with one another, for example the layers are spatially congruently one on top of the other.
- the hybrid scene representation according to the invention can be imagined as a stack of digital photos lying one on top of the other, for example taken from a bird's eye view of an intersection. On the other hand, this stack of images is combined with additional layers of partly purely semantic information that is represented, for example, as pure feature vectors.
- the advantage of the hybrid scene representation according to the invention is that a very large and very flexible amount of information is provided which the first machine learning model can access.
- the variable parameters/weights of the first machine learning model are adjusted, the use of the specific information that is best suited to solving the prediction task then emerges.
- the hybrid scene representation comprises a first layer comprising the regional information on the behavior of the road users and/or weather information, a second layer comprising map information for determining the positions of the road users, a third layer comprising traffic regulation information, and a fourth layer comprising the traffic signs, a fifth layer comprising anchor trajectories, a sixth layer comprising semantically explicit information, a seventh layer comprising semantically latent information and an eighth layer comprising the movement information.
- the first through third layers include the rigid static environmental features.
- Regional information and/or weather information improve the forecast quality. For example, the behavior of road users differs depending on the region. For example, in Germany traffic rules are relatively strictly observed, in Italy rather mildly, in Great Britain people overtake from the right, etc.
- Position data of the road user and/or the environment features are recorded via map information.
- a map section is formed by assigning a value to each pixel of the layer of the representation of the surrounding scene that corresponds to the map information. The values are based on discrete labels on the map, for example numeric codes for street, pedestrian walkway, broken line, double line, etc.
- the rules of right of way are shown next to the map via the traffic regulation information.
- a line is drawn in the middle of each lane. Additional lines are drawn at intersections, which represent all permissible manoeuvres.
- implicitly regulated information such as "Right before left" is overlaid on the signage. Any conflicting rule information is aggregated in this layer to form a consistent rule, so that the rules then in force are treated as having priority.
- the fourth to fifth layers include the state-changing static environment features.
- Traffic advisors include state-changing and stateful traffic advisors.
- Status-changing traffic signs are usually used to summarize signals that are passed on to the driver visually and that can change their status several times in the course of a day. Examples of this category are traffic lights, variable message signs on motorways and entry signs at toll booths. These traffic signs are represented as a pixel value representing the current state in the spatial context of the local scene representation. For reasons of redundancy, such pixel regions are generally not limited to one pixel, but are mapped to a larger number of pixels. The exact size of the expansion is mostly learned from data to an optimum.
- the anchor trajectories combine information from the right of way rules and from the traffic signs that change status.
- the anchor trajectories determined in this way are brought into line with the rules of the status-changing traffic indicators and prioritized accordingly.
- the layer of anchor trajectories can, according to one aspect of the invention, depending on the time required by the traffic participants, for example the driving system, supplement or replace the layers of traffic information and/or traffic regulation information.
- the sixth to eighth layers contain the dynamic environment characteristics.
- Semantically explicit information includes vehicle class, for example trucks, cars, motorcycles, buses, shuttles, bicycles, pedestrians, height and/or width of the objects and/or states of the blinking lights.
- Semantic-latent information cannot be interpreted directly by humans, but is in a certain way implicitly contained in the data.
- the latent information is, for example, continuous numbers with which the robustness against noise signals of discrete classes is increased, for example when a discrete classification varies between truck and car.
- the movement information of the eighth layer includes the trajectory histories.
- Determining the interactions includes predicting possible future interactions, according to one aspect of the invention, based on the eight layers of the hybrid scene representation.
- Interactions concern any interaction between static and static, static and dynamic and dynamic and dynamic environmental features.
- a car is located at an intersection.
- the right of way is regulated by a traffic light.
- One interaction is the traffic light switch. For example, if the traffic light switch shows the car the green traffic light phase and the pedestrian the red traffic light phase, then the other interactions, learned or present in the trajectory histories, are that the pedestrian stops and the car drives into the junction.
- the first machine learning model is used with data pairs of the form
- HSRV_T, GT_T trained. T indicates the number of training data pairs.
- HSRV is the respective hybrid scene representation, on the basis of which the interactions and thus the trajectories are predicted.
- GT is the respective reference prediction, also called ground truth, i.e. the target prediction.
- the optimal parameters for the first machine learning model comprising values for weights are then determined by an optimization method, for example gradient descent. The parameters are optimal when the deviation of the actual predictions output from the first machine learning model from the target predictions is minimized.
- the advantage of processing the hybrid scene representation according to the invention is that information in the second layer changes more frequently than in the first layer.
- the inference time of the first machine learning model is reduced. In this way, the joint embedding, ie a kind of final scene embedding, is generated more quickly.
- the embedding is, for example, embedding in a feature space in which each road user can be identified via coordinates. Since the embedding has the same dimensionality as the hybrid scene representation in terms of spatial resolution, the coordinates used for classification in the hybrid scene representation can be used for each road user in order to obtain the information relevant to the road user from the common embedding.
- the road user-specific interactions are determined from the embedding, for example, in that for each of the road users, characteristics, also called features, of the common embedding are extracted at the point corresponding to the coordinates of the respective road user. These features include all interactions relevant to the respective road user. By fusing these features with the respective road user-specific dynamic features, the prediction of trajectories for the respective road user is calculated based on all interactions of the common embedding.
- each of the trajectory histories are decoded into a plurality of possible predicted trajectories, a large number of possible future modes are calculated, ie a one-to-many mapping is produced.
- the first machine learning model learns probability density functions of the different modes. This makes the one-to-many mapping explicit estimated by learning from multiple modes as opposed to implicit estimation using generative adversarial networks. This is advantageous for a training method of the first machine learning model, since otherwise the mode collapse problem would affect the training.
- the commands of the computer program according to the invention include machine commands, source text or object code written in assembly language, an object-oriented programming language, for example C++, or in a procedural programming language, for example C.
- the computer program is a hardware-independent application program, for example is provided via a data carrier or a data carrier signal using software over the air technology.
- the interaction tensor pooling module and the interaction vector extraction module comprise software and/or hardware components.
- the arrangement relates to computing units that execute the individual encoders, decoders, networks and modules.
- the decoders of the arrangement comprise an attention mechanism.
- the arrangement executes the method according to the invention or the computer program according to the invention.
- a first function of the first machine learning model is applied to the first layer and a first embedding of the rigid static environment features is created.
- a second function is applied to the second layer and a second embedding of the state-changing static environment features is created.
- a third function is applied to the third layer and a third embedding of the dynamic environment features is created.
- the first, second and third embedding will be merged and output as a common embedding from the machine learning model.
- the term embedding refers both to the embedding created with the function applied to all layers together and to the common embedding.
- the rigid static environmental features are processed by layers of a first artificial neural network and embedded in a first tensor.
- the state-changing static environment features are processed by layers of a second artificial neural network and embedded in a second tensor.
- the dynamic environment features are processed by layers of a third artificial neural network and embedded in a third tensor.
- the common embedding is a fourth tensor obtained from a spatial concatenation of the first, second, and third tensors.
- Processing the layers of the first, second and third artificial neural networks is an embodiment of the first, second and third function.
- Activation functions of neuron connections are, for example, non-linear functions.
- the processing of the layers of the first, second and third artificial neural network thus represent non-linear functions.
- the first, second and third tensors differ in a semantic depth.
- the fourth tensor represents a multi-agent scene tensor, where the agents correspond to the road users and include the driving systems.
- the multi-agent scene tensor is calculated based on the hybrid scene representation with the first machine learning model and models the entire interactions of a scenario.
- the first machine learning model is also called the interaction network.
- the multi-agent scene tensor has the same dimensionality as the hybrid scene representation in terms of spatial resolution, the coordinates used to classify the respective tensors in the hybrid scene representation can be used for each road user in order to to obtain information relevant to road users from the joint embedding. Furthermore, the features of the multi-agent scene tensor are used at the point corresponding to these coordinates and thus form a multi-agent scene embedding for each road user. Since for each road user all interactions relevant to the respective road user can be managed by means of the multi- agent-scene tensors are considered, multi-agent scene embedding is also called joint multi-agent scene embedding, abbreviated to JoMASE.
- the multi-agent scene tensor is used to generate a multi-agent scene embedding for each road user in a scene.
- the features of the multi-agent scene tensor are extracted at the point corresponding to the road user's coordinates and merged with the third embedding of the road user.
- the fusion can take place earlier. This extraction is also called interaction vector extraction. Every road user, in particular every driving system, corresponds to a vector in the multi-agent scene tensor. In this way, the road user, in particular the driving system-specific embeddings are preserved.
- the encoded dynamic bird's-eye view environmental features are placed in a spatial tensor that is initialized to 0 and has the same shape, such as width and height, as the image of the encoded static environmental features.
- the encoded dynamic environmental features comprising the trajectory histories are placed in the spatial tensor in relation to their positions in the last time step of their past trajectories.
- This tensor is then concatenated with the image of the encoded static environmental features and the hybrid scene representation is obtained as a combined tensor.
- This information combination is a tensor pooling method and is called interaction tensor pooling.
- This combined tensor is processed by the first machine learning model, which, according to the invention, determines the interactions between the road users and between the road users and the static environmental features while maintaining the locality and outputs the multi-agent scene tensor.
- the first machine learning model comprises skip connections, through which layers are skipped when the machine learning model is processed, for example a double or triple layer jump. Jump connections are also called skip-connections.
- the Hop connections advantageously expose high-level features of interaction.
- the hop connections comprise residual connections that a residual function learns with respect to the layer inputs, such as used in residual neural networks.
- the hop connections comprise chained hop connections. A chained hop connection attempts to reuse features by chaining them into new layers so that more information from previous layers of the network is preserved. This is in contrast to, for example, the residual connections, which instead use element-wise summation to carry over information from earlier layers.
- the fourth tensor is processed by layers of a fourth artificial neural network.
- the fourth artificial neural network outputs an output volume equal in size to the hybrid scene representation.
- the road users are positioned in the output volume based on their actual spatial arrangement.
- the information contained in the fourth tensor, for example in the multi-agent scene tensor, is merged by the fourth artificial neural network.
- the embeddings are generated with convolution networks.
- the first machine learning model is a convolutional network.
- the first, second, third and fourth artificial neural networks are separate convolution networks or individual function blocks in the first machine learning model implemented as a convolution network.
- Convolutional networks also called convolutional neural networks, abbreviated CNN, are particularly advantageous for processing large amounts of data and data that can be represented as images.
- the road user-specific embeddings of past time steps of the trajectory histories of the road users in individual time steps of the decoding weighted differently uses an attention mechanism.
- One aspect of the invention provides an attention module or attention layer in which outputs from the encoder are accumulated.
- the attention module executes an attention algorithm, by means of which the influence of each past time step of the trajectory histories on the current or future trajectories is evaluated during decoding and the most relevant of the past time steps of the trajectory histories are provided to the decoder.
- a further aspect of the invention provides a transformer architecture with self-attention based on sequential encoder and decoder blocks with a similar architecture.
- Each of the encoder blocks includes, for example, a self-attention layer and a feedforward layer.
- Each of the decoder blocks comprises, for example, a self-attention layer, a feedforward layer and an encoder-decoder self-attention layer arranged in between.
- the self-attention algorithm is, for example, in D. Gizlyk, Neural Networks Made Easy (Part 8): Attention Mechanisms, February 8, 2021, https://www.mql5.com/de/artic- Ies/8765#para2 , described.
- Transformer architectures are based, for example, on Bidirectional Encoder Representations from Transformers, abbreviated BERT, or on Generative Pre-trained Transformer, abbreviated GPT.
- the attentional mechanisms improve predictions, especially long-term predictions, and solve the problem of information morphing.
- An advantage of attentional mechanism decoding is better long-term prediction.
- Another advantage of decoding with an attention mechanism is better explainability, since it is possible to find out at each time step of the decoding time which time steps in the past were more influential for the prediction.
- the road user-specific embeddings in a first data stream which predicts various plausible trajectories, and in a second data stream, evaluates the probabilities of the predicted trajectories against each other and outputs a corresponding value, decoded.
- the probabilities are based on the trajectory histories and the calculated interactions.
- the rigid static environment features including map information and the status-changing static environment features including traffic signs and/or anchor trajectories from real data from environment recognition sensors and/or from maps and/or from virtual data with a second machine learning model including layers of a convolution network coded.
- the second machine learning model encodes the static environmental features mentioned, for example, in a semantic feature map, also known as a feature map, in which the individual features are stacked according to the layer structure of the hybrid scene representation.
- the second machine learning model is thus a scene information coder, in particular a CNN coder. Convolution networks are particularly well suited for this.
- Data from environment recognition sensors include raw data and/or data preprocessed, for example with filters, amplifiers, serializers, compression and/or conversion units, from cameras, radar sensors, lidar sensors, ultrasonic sensors, acoustic sensors, Car2X units and/or real-time /Offline Maps.
- the virtual data is generated, for example, using software, hardware, model and/or vehicle-in-the-loop methods. According to a further aspect of the invention, the real data is virtually augmented and/or varied.
- the dynamic environment features including the trajectory histories of the road users, based on real driving data and/or virtual data are coded using a third machine learning model including layers of a recurrent network.
- Recurrent neural networks also known as RNN for short, recognize the time-coded data in the trajectory histories.
- that is recurrent network as a long short-term memory network, abbreviated LSTM, or implemented as a gated recurrent unit network.
- the third machine learning model is thus an RNN encoder.
- the coding of the third machine learning model is overlaid with the semantic feature map of the second machine learning model.
- the road user-specific embeddings are decoded by a fourth machine learning model comprising layers of a recurrent network.
- the fourth machine learning model is thus an RNN decoder.
- the first data stream predicting different plausible trajectories is decoded by a first RNN decoder.
- the first RNN decoder is called a probability decoder.
- the second data stream which compares the probabilities of the predicted trajectories and outputs a corresponding value, is decoded by a second RNN decoder.
- the second RNN decoder is called a trajectory decoder.
- a control unit of one of the driving systems for automated driving functions determines regulation and/or control signals based on the predicted trajectories and provides these signals to actuators for longitudinal and/or lateral guidance of the driving system.
- FIG. 4 shows a representation of road user-specific embeddings according to the invention and 5 shows a flow chart of the method according to the invention.
- FIG. 1 shows an example of a hybrid scene representation HSRV according to the invention.
- a car as an example of a road user R at a junction.
- the car for example, is the ego driving system.
- a pedestrian W At the junction there is a pedestrian W.
- the right of way is controlled by a traffic light L.
- the traffic light circuit L shows the car R the green traffic light phase and the pedestrian W the red one. Above the depiction of this situation from a bird's eye view, the various layers that are essential for predicting the trajectories of road users are shown.
- Layer A shows the regional information.
- Layer B uses the map information, layer C the traffic regulation information.
- the stateful traffic signs and the anchor trajectories are contained in layer D and layer E.
- Layer F describes the semantic characteristics of the individual road users.
- Layer G and layer H contain latent information, with this information in layer G being based on properties that describe the road user and in layer H on the dynamic movement behavior.
- the layers A to E are static layers and describe static environmental features stat of the environmental scene U.
- the layers A to C describe rigid static environmental features stat_1 and the layers D and E state-changing static environmental features stat_2.
- the layers F to H are dynamic layers and describe dynamic environment features dyn of the environment scene U.
- FIG. 2 shows an illustration of the method according to the invention.
- the static environment characteristics stat and the dynamic environment characteristics dyn were included in the hybrid Scene representation HSRV merged.
- the machine learning models according to the invention receive this hybrid scene representation HSRV as input and calculate a specific embedding JoMASE for each road user R. Future trajectories with associated probabilities are decoded from the road user-specific embeddings JoMASE.
- FIG 3 shows an arrangement of a network architecture according to the invention.
- FIG. 4 shows a representation of the road user-specific embeddings JoMASE according to the invention.
- a third machine learning model RNN encoder encodes the trajectory histories TH of road users based on real driving data and/or virtual data.
- a second machine learning model CNN encoder encodes scene information comprising the rigid static environment features stat_1 and the state-changing static environment features stat_2.
- An interaction tensor pooling module ITPM combines the encodings of the RNN and CNN encoders and uses them to generate the hybrid scene representation HSRV as shown in FIG.
- a first machine learning model IntCNN in the form of a convolutional network CNN determines the interactions between the static stat, dynamic dyn and between the static stat and dynamic environment features dyn based on the hybrid scene representation HSRV and merges these interactions.
- the first machine learning model IntCNN creates a first embedding of the rigid static environment features stat_1 in the form of a first tensor embedding, a second embedding of the state-changing static environment features stat_2 in the form of a second tensor embedding and a third embedding of the dynamic environment features dyn in Form of a third tensor embedding.
- the first, second and third embedding are merged into a common embedding M in the form of a multi-agent scene tensor.
- An interaction vector extraction module IVEM extracts the features of the multi-agent scene tensor M for each road user R at the coordinates of the The point corresponding to the point of traffic participant R and merges this with the third tensor embedding of traffic participant R.
- the multi-agent scene embedding JoMASE is generated for each traffic participant R and for each scene.
- a fourth machine learning model RNN trajectory decoder decodes the multi-agent scene embedding JoMASE in a first strand and outputs R predicted trajectories for each road user.
- An RNN probability decoder of the fourth machine learning model decodes the multi-agent scene embedding JoMASE in a second strand and evaluates R probabilities of the predicted trajectories against each other for each road user and outputs a corresponding value in each case.
- the decoders of the fourth machine learning model include, for example, recurrent networks RNN.
- FIG. 5 shows the method according to the invention as a flow chart.
- the provided encoded static stat and dynamic environment features dyn are spatially combined by the interaction tensor pooling module ITPM.
- the hybrid scene representation FISRV is obtained by means of the interaction tensor pooling module ITPM.
- the hybrid scene representation FISRV is processed by the first machine learning model IntCNN.
- the first machine learning model IntCNN determines the first embedding of the rigid static environment features stat_1, the second embedding of the state-changing static environment features stat_2 and the third embedding of the dynamic environment features dyn.
- the first, second and third embedding are merged and output as a common embedding M from the first machine learning model IntCNN.
- the traffic participants R-specific interactions are determined from the common embedding M and with the traffic participants mer R-specific dynamic environment features dyn for each of the Road users R merged. From the fusion, the road user R-specific embedding JoMASE is generated for each of the road users R.
- a method step V5 the road users R-specific embeddings JoMASE are decoded and the predicted trajectories for each of the road users R are obtained, with individual trajectory histories TH being mapped onto a plurality of possible predicted trajectories.
- stat_1 rigid static environment characteristics
- stat_2 state-changing static environment characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mechanical Engineering (AREA)
- Transportation (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Automation & Control Theory (AREA)
- Chemical & Material Sciences (AREA)
- Human Computer Interaction (AREA)
- Combustion & Propulsion (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Analytical Chemistry (AREA)
- Databases & Information Systems (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102021203440.3A DE102021203440A1 (de) | 2021-04-07 | 2021-04-07 | Computerimplementiertes Verfahren, Computerprogramm und Anordnung zum Vorhersagen und Planen von Trajektorien |
| PCT/EP2022/058831 WO2022214414A1 (de) | 2021-04-07 | 2022-04-04 | Computerimplementiertes verfahren, computerprogramm und anordnung zum vorhersagen und planen von trajektorien |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4320408A1 true EP4320408A1 (de) | 2024-02-14 |
Family
ID=81454611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP22720428.6A Pending EP4320408A1 (de) | 2021-04-07 | 2022-04-04 | Computerimplementiertes verfahren, computerprogramm und anordnung zum vorhersagen und planen von trajektorien |
Country Status (3)
| Country | Link |
|---|---|
| EP (1) | EP4320408A1 (de) |
| DE (1) | DE102021203440A1 (de) |
| WO (1) | WO2022214414A1 (de) |
Families Citing this family (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102022201127B4 (de) | 2022-02-03 | 2024-12-05 | Zf Friedrichshafen Ag | Verfahren und Computerprogramm zum Charakterisieren von zukünftigen Trajektorien von Verkehrsteilnehmern |
| CN115817531B (zh) * | 2022-11-25 | 2026-04-28 | 中国第一汽车股份有限公司 | 一种行人运动轨迹的确定方法及确定装置 |
| CN116071923B (zh) * | 2023-01-16 | 2025-05-27 | 天津大学 | 一种基于自适应图融合卷积网络的交通流量预测方法 |
| CN116341721B (zh) * | 2023-03-02 | 2023-10-31 | 石家庄铁道大学 | 一种基于车辆轨迹的运行目的地预测方法及系统 |
| DE102023202995A1 (de) | 2023-03-31 | 2024-10-02 | Robert Bosch Gesellschaft mit beschränkter Haftung | Verfahren zur Auswertung einer Verkehrsszene |
| DE102023108655A1 (de) * | 2023-04-04 | 2024-10-10 | Bayerische Motoren Werke Aktiengesellschaft | Abbilden eines vorbestimmten geografischen Gebiets |
| CN117010265B (zh) * | 2023-04-14 | 2025-06-13 | 北京百度网讯科技有限公司 | 能够进行自然语言交互的自动驾驶模型及其训练方法 |
| CN116540701B (zh) * | 2023-04-19 | 2024-03-05 | 广州里工实业有限公司 | 一种路径规划方法、系统、装置及存储介质 |
| EP4456016A1 (de) * | 2023-04-28 | 2024-10-30 | Aptiv Technologies AG | Verfahren zur vorhersage von trajektorien von verkehrsteilnehmern |
| DE102023205056A1 (de) * | 2023-05-31 | 2024-12-05 | Robert Bosch Gesellschaft mit beschränkter Haftung | Computer-implementiertes Verfahren und System zur Verhaltensplanung eines Teilnehmers einer Verkehrsszene |
| CN116558541B (zh) * | 2023-07-11 | 2023-09-22 | 新石器慧通(北京)科技有限公司 | 模型训练方法和装置、轨迹预测方法和装置 |
| CN116910484B (zh) * | 2023-08-24 | 2025-11-21 | 驭势科技(北京)有限公司 | 轨迹预测方法、装置、电子设备和存储介质 |
| CN117208012A (zh) * | 2023-09-20 | 2023-12-12 | 安徽蔚来智驾科技有限公司 | 车辆轨迹预测方法、控制装置、可读存储介质及车辆 |
| CN117508219B (zh) * | 2023-09-21 | 2025-02-25 | 零束科技有限公司 | 一种车辆路径规划方法及装置 |
| DE102023129397A1 (de) * | 2023-10-25 | 2025-04-30 | Dr. Ing. H.C. F. Porsche Aktiengesellschaft | Verfahren und System zur Identifikation von repräsentativen Streckenabschnitten |
| CN117539253B (zh) * | 2023-11-29 | 2025-03-14 | 北京百度网讯科技有限公司 | 能够遵循指令实现自主脱困的自动驾驶方法、装置和车辆 |
| WO2025147368A1 (en) * | 2024-01-05 | 2025-07-10 | Qualcomm Incorporated | Conditional turn trajectory prediction network for urban intersections |
| DE102024001594A1 (de) | 2024-05-16 | 2024-07-11 | Mercedes-Benz Group AG | Verfahren und Vorrichtung zur Modellierung von dynamischen Objekten und statischen Obiekten beim assistierten und automatisierten Fahren |
| CN118394874B (zh) * | 2024-07-01 | 2024-09-17 | 杭州弘云信息咨询有限公司 | 一种基于大语言模型引导的车辆轨迹预测方法及装置 |
| CN118587898A (zh) * | 2024-08-07 | 2024-09-03 | 电子科技大学长三角研究院(衢州) | 一种基于预训练模型的跨城市交通流预测方法 |
| CN118644989B (zh) * | 2024-08-13 | 2024-12-03 | 浙江清华长三角研究院 | 交互风险预测方法、装置、介质及设备 |
| CN119252018A (zh) * | 2024-09-05 | 2025-01-03 | 深圳职业技术大学 | 交通路口场景下的轨迹预测方法、装置、终端及存储介质 |
| CN119190074A (zh) * | 2024-09-24 | 2024-12-27 | 中国第一汽车股份有限公司 | 一种基于transformer架构的自动驾驶轨迹预测方法及系统 |
| CN119027917A (zh) * | 2024-10-22 | 2024-11-26 | 北京集度科技有限公司 | 一种轨迹规划方法、车辆及计算机程序产品 |
| CN119296331B (zh) * | 2024-12-12 | 2025-03-14 | 泉州师范学院 | 一种基于集群时空动态图网络的轨迹预测方法及系统 |
| CN119828690B (zh) * | 2024-12-17 | 2025-10-21 | 重庆大学 | 一种融合交互性轨迹预测的自动驾驶协同运动规划方法 |
| CN120031917B (zh) * | 2025-02-20 | 2025-10-21 | 无锡惠航智能技术有限公司 | 基于kan的全注意力交互与rnn条件vae的行人轨迹预测方法 |
| CN119840643B (zh) * | 2025-03-18 | 2025-07-04 | 江西五十铃汽车有限公司 | 基于注意力机制的自适应迭代轨迹预测方法及系统 |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11169531B2 (en) * | 2018-10-04 | 2021-11-09 | Zoox, Inc. | Trajectory prediction on top-down scenes |
| DE102019209736A1 (de) | 2019-07-03 | 2021-01-07 | Robert Bosch Gmbh | Verfahren zur Bewertung möglicher Trajektorien |
| DE102020210379A1 (de) | 2020-08-14 | 2022-02-17 | Zf Friedrichshafen Ag | Computerimplementiertes Verfahren und Computerprogrammprodukt zum Erhalten einer Umfeldszenen-Repräsentation für ein automatisiertes Fahrsystem, computerimplementiertes Verfahren zum Lernen einer Prädiktion von Umfeldszenen für ein automatisiertes Fahrsystem und Steuergerät für ein automatisiertes Fahrsystem |
-
2021
- 2021-04-07 DE DE102021203440.3A patent/DE102021203440A1/de active Pending
-
2022
- 2022-04-04 EP EP22720428.6A patent/EP4320408A1/de active Pending
- 2022-04-04 WO PCT/EP2022/058831 patent/WO2022214414A1/de not_active Ceased
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022214414A1 (de) | 2022-10-13 |
| DE102021203440A1 (de) | 2022-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP4320408A1 (de) | Computerimplementiertes verfahren, computerprogramm und anordnung zum vorhersagen und planen von trajektorien | |
| WO2022033810A1 (de) | Computerimplementiertes verfahren und computerprogrammprodukt zum erhalten einer umfeldszenen-repräsentation für ein automatisiertes fahrsystem, computerimplementiertes verfahren zum lernen einer prädiktion von umfeldszenen für ein automatisiertes fahrsystem und steuergerät für ein automatisiertes fahrsystem | |
| DE102022201127B4 (de) | Verfahren und Computerprogramm zum Charakterisieren von zukünftigen Trajektorien von Verkehrsteilnehmern | |
| DE112020005577T5 (de) | Simulieren diverser langfristiger zukünftiger Trajektorien in Strassenszenen | |
| DE102018203583B4 (de) | Verfahren, Fahrerassistenzsystem sowie Kraftfahrzeug zur Prädiktion einer Position oder einer Trajektorie mittels eines graphbasierten Umgebungsmodells | |
| DE102021109395A1 (de) | Verfahren, systeme und vorrichtungen für benutzerverständliche erklärbare lernmodelle | |
| WO2020094534A1 (de) | Verfahren zum trainieren eines künstlichen neuronalen netzes, künstliches neuronales netz, verwendung eines künstlichen neuronalen netzes sowie entsprechendes computerprogramm, maschinenlesbares speichermedium und entsprechende vorrichtung | |
| DE102021213344A1 (de) | Verfahren zum Ermitteln von Agenten-Trajektorien in einem Multi-Agenten-Szenario | |
| DE112022002869T5 (de) | Verfahren und System zur Verhaltensprognose von Akteuren in einer Umgebung eines autonomen Fahrzeugs | |
| DE102021200643B3 (de) | Verfahren zur Umfelderkennung für teilautonome oder autonome Fahrfunktionen eines Kraftfahrzeugs mittels eines neuronalen Netzes | |
| WO2021249677A1 (de) | Verfahren zum koordinieren von verkehrsteilnehmern durch eine servervorrichtung sowie eine servervorrichtung und eine steuerschaltung zum durchführen des verfahrens | |
| DE102022003079A1 (de) | Verfahren zu einer automatisierten Generierung von Daten für rasterkartenbasierte Prädiktionsansätze | |
| DE102023001698A1 (de) | Verfahren zu einer automatisierten Generierung von Daten für rasterkartenbasierte Prädiktionsansätze | |
| DE102024100521A1 (de) | Bildgestützte generierung von beschreibenden und wahrnehmungs-nachrichten von automobilszenen | |
| DE102023200757A1 (de) | Computer-implementiertes Verfahren zur Verhaltensplanung eines zumindest teilautomatisierten Ego-Fahrzeugs mit einem vorgegebenen Navigationsziel | |
| DE102022214341A1 (de) | Verfahren zur Erkennung von mindestens einer Fahrspurlinie und/oder Straßenmarkierung in mindestens einer digitalen Bilddarstellung | |
| DE102022209034A1 (de) | Vorhersage des Verhaltens von Verkehrsteilnehmern basierend auf einer graphischen Darstellung einer Verkehrssituation | |
| DE102021117227A1 (de) | Analysieren eines Kreisverkehrs | |
| DE102024132828A1 (de) | Vision language planning (vlp) modelle mit agentenweisem lernen für autonomes fahren | |
| DE102022131178B3 (de) | Verfahren zum automatisierten Führen eines Fahrzeugs sowie Verfahren zum Erzeugen eines hierzu fähigen Modells des Maschinellen Lernens sowie Prozessorschaltung und Fahrzeug | |
| DE102024000515A1 (de) | Verfahren zur Prädiktion von am Straßenverkehr teilnehmenden Agenten in einer Umgebung eines automatisiert fahrenden Fahrzeugs | |
| DE102024203277A1 (de) | Verfahren zum Vorhersagen von Trajektorien von Verkehrsteilnehmern | |
| DE102023003979A1 (de) | System zur Erfassung, Vorhersage und/oder Überwachung einer Fahrzeugumgebung sowie zur Bewegungsplanung eines Fahrzeugs | |
| DE102023210638A1 (de) | Computerimplementiertes Verfahren zum Trainieren eines Modells sowie Verfahren und System zur Prädiktion des Verhaltens eines Verkehrsteilnehmers | |
| DE102019204187A1 (de) | Klassifizierung und temporale Erkennung taktischer Fahrmanöver von Verkehrsteilnehmern |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20231006 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| RIC1 | Information provided on ipc code assigned before grant |
Ipc: G01C 21/00 20060101AFI20250922BHEP Ipc: B60W 60/00 20200101ALI20250922BHEP Ipc: B60W 30/095 20120101ALI20250922BHEP Ipc: G08G 1/01 20060101ALI20250922BHEP Ipc: G06N 3/02 20060101ALI20250922BHEP Ipc: B60W 10/04 20060101ALI20250922BHEP Ipc: B60W 10/184 20120101ALI20250922BHEP Ipc: B60W 10/20 20060101ALI20250922BHEP Ipc: B60W 50/00 20060101ALI20250922BHEP Ipc: G06N 3/006 20230101ALI20250922BHEP Ipc: G06N 3/044 20230101ALI20250922BHEP Ipc: G06N 3/0455 20230101ALI20250922BHEP Ipc: G06N 3/0464 20230101ALI20250922BHEP Ipc: G06N 3/09 20230101ALI20250922BHEP Ipc: G06N 3/0442 20230101ALI20250922BHEP Ipc: G06N 3/045 20230101ALI20250922BHEP Ipc: G06N 3/08 20230101ALI20250922BHEP |
|
| GRAP | Despatch of communication of intention to grant a patent |
Free format text: ORIGINAL CODE: EPIDOSNIGR1 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: GRANT OF PATENT IS INTENDED |
|
| INTG | Intention to grant announced |
Effective date: 20251112 |