Method for trajectory prediction and method for automatically driving a vehicle
The method uses an attention-based interaction algorithm to preselect relevant vehicles for trajectory prediction, addressing complexity and runtime issues in existing methods by focusing on vehicle interactions, thereby improving prediction quality and efficiency.
Patent Information
- Application Number
- JP2024525762
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2021-11-12
- Filing Date
- 2022-10-10
- Publication Date
- 2024-11-28
- Estimated Expiration
- 2042-10-10
AI Technical Summary
Existing trajectory prediction methods for automated vehicles face challenges with high complexity and runtime dependencies due to the inclusion of all vehicles in the scene, leading to poor prediction results and convergence issues, especially when high-resolution map data is required.
A two-step method involving a machine-trained attention-based interaction algorithm to determine the degree of interaction between vehicles, allowing for preselection of relevant vehicles for trajectory prediction, reducing complexity and execution time by focusing on interactions rather than distance-based selection.
This approach improves the quality and speed of trajectory prediction by selectively training on relevant vehicles, reducing processing costs and enhancing prediction accuracy and convergence, even with complex algorithms.
Smart Images

Figure 2024544122000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a method for the prediction of a trajectory of a vehicle in its surrounding environment according to the preamble of claim 1.
[0002] The present invention further relates to a method for automatically driving a vehicle. [Background technology]
[0003] From the prior art, automated vehicles, for example highly automated or autonomous vehicles, are generally known. In order to be able to meet the safety requirements imposed on such vehicles, it is necessary to predict the future movements of vehicles involved in traffic, also called trajectory prediction. For this trajectory prediction, approaches based on machine learning methods are known.
[0004] In the following, the term "scene" refers to an environmental situation in which at least one further vehicle may be present, as detected by means of the vehicle's environmental detection unit. Additionally, map data of a digital road map may also be used to describe such a scene.
[0005] Unlike rasterization approaches, such as CNN-based approaches (CNN: Convolutional Neural Network), approaches for trajectory prediction that do not perform rasterization of the scene are based on so-called Graph Neural Networks (GNNs) or Self-Attention, also referred to as attention-based approaches in the following. To model the social interactions between vehicles, these approaches form a graph. The nodes of the graph are the encodings or codings of the vehicles in the scene. The nodes are connected to each other via edges.
[0006] Approaches exist that connect all vehicles in the scene with edges so that a complete graph, also called a fully-connected graph, is formed, which means that all vehicles are selected as being included or relevant in the trajectory prediction of the vehicle to be predicted.
[0007] In another approach, the edges of the graph are formed according to the distance between individual vehicles, thus providing a distance-based pre-selection of the vehicles to be considered for the prediction of the target vehicle.
[0008] Y. Ma et al.: Traffic Predict: Trajectory Prediction for Heterogeneous Traffic-Agents; In The Thirty-Third AAAI Conference on Artificial Intelligence, AAAI 2019, 2019, pp. 6120~6127, doi: 10.1609 / aaai.v33i01.33016120 describes a method for predicting future trajectories of road users for navigation of autonomous vehicles. A real-time traffic prediction algorithm based on the so-called Long Short-Term Memory (LSTM) is used. It includes an instance level to learn the movements and interactions of instances. Furthermore, the real-time traffic prediction algorithm also includes a category level to learn the similarity of instances of the same type to refine the prediction. Summary of the Invention [Problem to be solved by the invention]
[0009] It is an object of the present invention to provide a novel method for predicting the trajectory of a vehicle in its surrounding environment and a novel method for automatically driving the vehicle. [Means for solving the problem]
[0010] According to the present invention, the above problem is solved by a method for trajectory prediction having the features recited in claim 1 and a method for autonomously driving a host vehicle having the features recited in claim 9.
[0011] Advantageous embodiments of the invention are the subject of the dependent claims.
[0012] In a method for trajectory prediction of a vehicle in the environment of an ego vehicle, according to the invention, an interaction degree between the vehicles is determined using a machine-trained attention-based interaction algorithm, i.e. the interaction degree is determined for both the interaction between the vehicles in the environment of the ego vehicle and the interaction between the ego vehicle and the vehicles in the environment of the ego vehicle. Furthermore, using the interaction algorithm, individual vehicles are identified as relevant for trajectory prediction from the set of vehicles present in the environment of the ego vehicle and are selected for the trajectory prediction if their respective interaction degrees with at least one vehicle whose trajectory is to be predicted exceed a predefined limit value. In a subsequent learning step, the trajectory prediction algorithm is trained with the vehicles selected as relevant for the trajectory prediction, and in particular only with the vehicles selected as relevant for the trajectory prediction, and the trajectory prediction performed using the trajectory prediction algorithm is performed for the vehicles selected as relevant for the trajectory prediction, in particular only with the vehicles selected as relevant for the trajectory prediction.
[0013] In particular, the method constitutes a so-called two-step approach for vehicle trajectory prediction, in which an interaction-based vehicle pre-selection is first performed, followed by trajectory prediction using a reduced number of vehicles.
[0014] Learning-based trajectory prediction algorithms include map data, e.g., high-resolution digital road maps, and are often based on graph structures, where typically other vehicles detected in the environment of the ego-vehicle form nodes in the graph. Thus, the complexity of such trajectory prediction algorithms scales with the number of vehicles in the scene. This leads to strong runtime dependencies, especially if map information must also be queried for all vehicles in the scene. Furthermore, a large number of irrelevant vehicles in the scene makes the learning process of the trajectory prediction algorithm difficult, e.g., leading to poor prediction results and poor convergence behavior. For example, vehicles at a large distance have no measurable impact on the trajectory prediction of the vehicle to be predicted. Thus, once all agents or vehicles in the scene are selected, the trajectory prediction algorithm needs to independently identify which vehicles have a beneficial added value to the prediction and which do not.
[0015] In contrast, the method uses an interaction algorithm to realize an initial pre-selection of vehicles related to the trajectory prediction algorithm. The interaction algorithm is characterized by low complexity, since a particularly advantageous method does not require map data of high-resolution digital road maps to determine the interaction. Thus, the complexity can be reduced by a limited selection of vehicles, especially for complex and possibly map-based trajectory prediction algorithms. The graph-based approach also reduces the execution time, since the number of nodes in the graph can be reduced.
[0016] In contrast to distance-based selection, the pre-selection of relevant vehicles is not made according to the distance of the relevant vehicles, but according to the degree of interaction of the relevant vehicles with the vehicle to be predicted, so that it can be taken into account, for example, that the future trajectory of a vehicle is strongly influenced by a vehicle traveling ahead at a certain distance, whereas it is hardly influenced by a vehicle traveling behind in the same direction.
[0017] Thus, the method achieves an improved run time of the final trajectory prediction algorithm compared to approaches without pre-selection. Also, the pre-selection of relevant vehicles can reduce the training time of the final trajectory prediction algorithm. Furthermore, an improved quality of the final trajectory prediction can be realized compared to approaches without pre-selection and compared to approaches with distance-based pre-selection. The indicators of vehicle interactions can additionally be used for other tasks, e.g., impacting other road users.
[0018] Due to the pre-selection and the associated reduction in the number of selected vehicles, even very complex trajectory prediction algorithms can be used with low processing costs and short execution times. The trajectory prediction algorithms used here are, for example, - C.Tang and RRSalakhutdinov: “Multiple Futures Prediction”; In: Advances in Neural Information Processing Systems,2019,vol.32, - T.Salzmann, B.Ivanovic, P.Chakravarty and M.Pavone: “Trajectron++: Dynamically-Feasible Trajectory Forecasting with Heterogeneous Data”; In: Computer Vision-ECCV 2020,Cham,2020,pp.683-700, - M.Liang et al.: “Learning Lane Graph Representations for Motion Forecasting”; In: Computer Vision-ECCV 2020,Cham,2020,pp.541-556, - S. Khandelwal, W. Qi, J. Singh, A. Hartnett and D. Ramanan: “What-If Motion Prediction for Autonomous Driving” 2020, and / or - Y.Ma, X.Zhu, S.Zhang, R.Yang, W.Wang and D.Manocha: “TrafficPredict:Trajectory Prediction for Heterogeneous Traffic-Agents”; In The Thirty-Third AAAI Conference on Artificial Intelligence, AAAI 2019,2019,pp.6120-6127,doi:10.1609 / aaai.v33i01.33016120 It is formed according to
[0019] In one possible configuration of the method, the degree of interaction between the vehicles is determined pairwise for each two vehicles.
[0020] In one possible further configuration of the method, during machine training of the interaction algorithm, the dependencies between the vehicles are learned pairwise and the interaction algorithm is used to weight how significant the mutual influence of the vehicles is for each pair in order to form an interaction degree, so that during subsequent operation of the interaction algorithm the interaction between the vehicles can be easily and reliably determined and the interaction degree can be determined.
[0021] In one possible alternative configuration of the method, in machine training of the interaction algorithm, trajectory predictions are generated that are used to train the interaction algorithm, thereby achieving effective training of the interaction algorithm and achieving very high accuracy and reliability of the interaction algorithm in determining the interaction degree.
[0022] In one further possible configuration of the method, the generated trajectory predictions are used in machine training of the interaction algorithm for implicit learning of the degree of interaction.
[0023] In one alternative possible configuration of the method, during operation of the machine-trained interaction algorithm, all vehicles in the scene are encoded using long short-term memory (also called long short-term memory) forming the nodes of a complete graph (also called fully connected graph). The use of long short-term memory realizes the memory of previous experiences and thus a long-lasting short-term memory, since the principle behavior of the graph is encoded with corresponding weights. Here, the complete graph can be used to completely represent the relationships between the vehicles.
[0024] In one possible alternative configuration of the method, the nodes are connected to each other via edges and the distance between the vehicles is used as an edge feature, so that, for example, the distance between the vehicles can also be taken into account when determining each interaction degree.
[0025] In one possible configuration of the method, the trajectory prediction algorithm uses map data of a high-resolution digital road map for trajectory prediction. By using such a high-resolution digital road map, the current location and past movements of the vehicle detected and located in the scene can be stored. Here, the road map can include road topology, traffic signs, traffic signals (e.g., traffic light signals), pedestrian crossings and other information, thus achieving highly accurate and reliable trajectory prediction.
[0026] In the method according to the invention for automatically driving an ego-vehicle, in the method for predicting a trajectory of a vehicle in the surrounding environment of the ego-vehicle according to the above, the trajectory of the vehicle in the surrounding environment of the ego-vehicle is predicted and the predicted trajectory is taken into account in the automated lateral and / or longitudinal control of the ego-vehicle in the automated driving of the ego-vehicle. By using the method for predicting a trajectory of a vehicle, a particularly reliable and safe automated driving, e.g. highly automated or autonomous driving, of the ego-vehicle can be realized.
[0027] In the following, an embodiment of the present invention will be explained in detail with reference to the drawings. [Brief description of the drawings]
[0028] [Figure 1] FIG. 1 is a schematic block diagram of an apparatus for trajectory prediction. [Diagram 2] FIG. 1 is a schematic block diagram of an apparatus for forming an interaction algorithm; [Diagram 3] FIG. 10 shows a schematic diagram of the deviation of the average distance error of the two trajectory predictions from the final trajectory prediction algorithm results for different numbers of vehicles in the scene. [Figure 4] FIG. 10 shows a schematic diagram illustrating the deviation of the final distance error of the two trajectory predictions from the final trajectory prediction algorithm results for different numbers of vehicles in the scene. [Diagram 5] FIG. 10 shows a schematic diagram illustrating the deviation of the misclassification rates of the two trajectory predictions from the final trajectory prediction algorithm results for different numbers of vehicles in the scene. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0029] In all the drawings, the same reference numerals are used to designate corresponding parts.
[0030] In FIG. 1, a block diagram of a possible embodiment of a device 1 for trajectory prediction of a number of vehicles F1 to Fn in the surrounding environment of an ego vehicle is shown.
[0031] The data memory 2 stores data describing the scene in the environment surrounding the vehicle. These data include environmental data detected by an environmental sensor system of the vehicle, other vehicles and / or traffic infrastructure, not shown in detail, and map data KD of a high resolution digital road map. In particular, the environmental data here also includes data relating to the vehicles F1 to Fn.
[0032] In order to minimize the processing effort when predicting the trajectories of the vehicles F1 to Fn in the surrounding environment of the host vehicle, initial interactions are determined between the vehicles F1 to Fn whose trajectories are to be determined and all other vehicles F1 to Fn in the scene, i.e., including the host vehicle.
[0033] This determination is made by a machine-trained attention-based interaction algorithm 3, which determines multiple interaction degrees I1-Im between the vehicles F1-Fn, in particular between the vehicle F1-Fn whose trajectory T is predicted and the vehicles F1-Fn present in the surrounding environment. The interaction degrees I1-Im between the vehicles F1-Fn are determined for each pair of two vehicles F1-Fn (for every two of the vehicles F1-Fn).
[0034] This interaction algorithm 3 is characterized by a short execution time and low complexity and focuses only on the interactions between the vehicles F1 to Fn. The map data KD of a high-resolution digital road map is not required for the determination of the interaction degrees I1 to Im. Here, the interaction degrees I1 to Im represent so-called attention weights and are formed by the so-called Self-Attention 3.1 of the attention-based interaction algorithm 3.
[0035] In particular, when determining the interaction degrees I1 to Im, a trajectory prediction TP' is also generated which is used to train the interaction algorithm 3. Here, the trajectory prediction TP' is used in machine training of the interaction algorithm 3, for example in implicit learning of the interaction degrees I1 to Im.
[0036] Note that the interaction degrees I1-Im are not "labelable" and therefore cannot be explicitly learned from the data. Therefore, an attention-based interaction algorithm 3 is used that implicitly learns the interaction degrees I1-Im between the vehicles F1-Fn. In contrast, the training object, i.e. the predicted trajectory T, is "labelable". However, the interactions between the vehicles F1-Fn are learned implicitly.
[0037] In particular, the interaction algorithm 3 learns to use, in particular, pairwise dependencies between the individual vehicles F1-Fn to determine the interaction degrees I1-Im. That is, learning of pairwise dependencies between the vehicles F1-Fn is performed. Here, the attention mechanism used "weights" each pair how strongly the vehicles F1-Fn influence each other. The attention weights used here take values ranging from 0.0 to 1.0, for example. Thus, it is possible to learn from the population of the data set how important each surrounding vehicle F1-Fn in the sequence or scene is for predicting the trajectory T of the selected vehicle F1-Fn and / or the ego vehicle.
[0038] After the training of the interaction algorithm 3 is completed, this interaction algorithm 3 is used in particular in the so-called Forward-Pass, in which the trajectory predicted by the interaction algorithm 3 plays a subordinate role. Instead, attention weights are extracted here for the predicted vehicles F1-Fn, which form interaction degrees I1-Im. These indicate for each other vehicle F1-Fn in the sequence how important each of these other vehicles F1-Fn is for the prediction of the selected vehicle F1-Fn and / or the ego vehicle.
[0039] Based on the respectively determined interaction degrees I1-Im, a vehicle pre-selection A(F1-Fn) is performed by the selection unit 4. That is, each vehicle F1-Fn from the set of vehicles F1-Fn present in the surrounding environment of the ego-vehicle is identified as a vehicle relevant for the trajectory prediction and selected for the trajectory prediction if the interaction degrees I1-Im of each of the individual vehicles F1-Fn with the other vehicles F1-Fn exceed a predefined limit value. Now, for each vehicle F1-Fn whose trajectory T is predicted, an individual selection of the relevant vehicles F1-Fn is performed. This is realized since the interaction degrees I1-Im are formed in pairs. For example, only a certain number of vehicles F1-Fn with the highest attention weights are selected or all vehicles F1-Fn with attention weights of 0.3 or more are selected.
[0040] Only if a vehicle pre-selection A(F1-Fn) has been made is an arbitrarily complex trajectory prediction algorithm 5 used, which also uses map data KD from a high resolution digital road map to finally predict, based on the pre-selected vehicles F1-Fn and the map data KD, a trajectory T of at least one of the vehicles F1-Fn in the environment surrounding the ego-vehicle.
[0041] Both the training of the trajectory prediction algorithm 5 and its operation on the autonomously capable ego-vehicle with the final predicted trajectory T are performed only on the basis of a vehicle pre-selection A(F1-Fn) with a reduced number of interacting vehicles F1-Fn. The vehicle pre-selection A(F1-Fn) allows the final trajectory prediction algorithm 5 to focus only on relevant or interacting vehicles F1-Fn and also facilitates the data-driven learning process, especially due to better convergence. This process is also simplified if the final trajectory prediction algorithm 5 additionally uses map data KD from a digital road map. The reason is that so-called "queries" from the road map, e.g. distance to the next stopping point or maximum permitted speed, can be very cumbersome in terms of time aspects, so that by pre-selecting the relevant vehicles F1-Fn the number of "queries" can be significantly reduced.
[0042] Fig. 2 shows a block diagram of a possible embodiment of an apparatus for forming the interaction algorithm 3 according to Fig. 1. Here, the interaction algorithm 3 is designed as an attention-based trajectory prediction algorithm implemented. In a manner not shown in detail, other configurations of the interaction algorithm 3 are possible, as long as the interaction degrees I1-Im are determined on an attention basis.
[0043] In the input encoder 6, all the vehicles F1 to Fn are encoded using so-called LSTM encoders 6.1 to 6.n equipped with long-term and short-term memory. During the encoding, the nodes h1 to hn of the complete graph G are encoded. 0 ~h n 0That is, the resulting encoding is a so-called fully connected graph G with a node h1 0 ~h n 0 Node h1 0 ~h n 0 is edge e 1,2 , e 1,n , e 2,1 , e 2,n , e n,1 , e n,2 The distance between the vehicles F1 to Fn is calculated as an additional edge feature, i.e., the edge e 1,2 , e 1,n , e 2,1 , e 2,n , e n,1 , e n,2 is used to form
[0044] Here, the graph G is divided into multiple layers L1 to L2 in the interaction module 7 that is the core of the interaction algorithm 3. g Furthermore, the interaction module is formed by the so-called parallel self-attention-heads SAL1 to SAL2, which are necessary for this method. h Using these, so-called Multi-Head-Self-Attention is performed following Ashish Vaswani et al.: “Attention Is All You Need” (https: / / arxiv.org / abs / 1706.03762).
[0045] Self-Attention Head SAL1~SAL h The layers L1 to L g The vector v obtained using (Lg) is provided, and in this case, based on the combination unit 7.1, the self-attention heads SAL1 to SAL hA combination and linearization of the output results is performed using the metric metric, thus determining features a1-an formed as latent encodings, which are updated based on the interaction degrees I1-Im and are also used for training. The features a1-an are then decoded into trajectories T to realize an implicit learning of the interaction degrees I1-Im. That is, the pairwise dependencies and / or weights between the vehicles F1-Fn in the scene, and thus the interaction degrees I1-Im between the agents, i.e. between the vehicles F1-Fn, are learned in the interaction module 7. Each interaction degree I1-Im is formed as a pairwise weight as a result of a so-called softmax function, for example according to Ashish Vaswani et al.: “Attention Is All You Need” (https: / / arxiv.org / abs / 1706.03762). The interaction degrees I1-Im are then further used to form the features a1-an. In particular, the interaction degrees I1-Im are stored in a matrix M of nxn format.
[0046] The output decoder 8 generates trajectory predictions TP' for the vehicles F1-Fn from the interaction degrees I1-Im using so-called linear residual decoders 8.1-8.z, which are used only for training the interaction algorithm 3. When applied to the trajectory prediction algorithm 5, the trajectory predictions TP' are not used, and the vehicle preselection A(F1-Fn) is performed using only the attention weights, i.e. the interaction degrees I1-Im. The residual layer is understood to be a layer consisting of several neuron layers. Unlike a conventional multilayer perceptron (abbreviated as MLP), the residual layer introduces so-called "skip connections". This allows the encoding of the first layer to be added to the encoding of the last layer by linear operations. This allows linear backpropagation of the error during training. This shows good results, especially in deep neural networks.
[0047] 3-5 show three established metrics used to determine the quality of an orbit prediction.
[0048] The metrics include the change in the minimum average displacement error ΔminADE as a function of a fixed number C of vehicles F1-Fn per scene (hatched bars) for trajectory predictions determined based on a vehicle pre-selection A(F1-Fn) by the trajectory prediction algorithm 5 described in Figures 1 and 2, and for trajectory prediction algorithms where the number C of vehicles F1-Fn is limited using a distance-based scheme (unhatched bars), as shown in Figure 3. The minimum average displacement error represents how far, on average, each calculated position of the respective trajectory T is from the true position.
[0049] Further, the metrics include the change in the minimum final displacement error ΔminFDE as a function of a fixed number C of vehicles F1-Fn per scene, as shown in Fig. 4 for trajectory predictions determined based on a vehicle pre-selection A(F1-Fn) by the trajectory prediction algorithm 5 described in Fig. 1 and Fig. 2 (hatched bars) and for trajectory prediction algorithms where the number C of vehicles F1-Fn is limited using a distance-based scheme (unhatched bars). Here, the minimum final displacement error represents the deviation of the prediction from the true trajectory T of the respective last prediction step.
[0050] Furthermore, the metrics include the change in the misclassification rate ΔMR as a function of a fixed number C of vehicles F1-Fn per scene, as shown in Fig. 5. The scenes are for trajectory predictions determined based on a vehicle pre-selection A(F1-Fn) by the trajectory prediction algorithm 5 described in Figs. 1 and 2 (hatched bars) and for trajectory prediction algorithms where the number C of vehicles F1-Fn is limited using a distance-based scheme (unhatched bars).
[0051] For all metrics, lower is better.
[0052] Based on these metrics, it is compared how the interaction-based vehicle pre-selection A (F1-Fn) described above performs compared to distance-based (Euclidean) selection for the already established final trajectory prediction algorithm 5.
[0053] Each zero value is given by the result of the final trajectory prediction algorithm 5, which has been trained and tested on the entire dataset. Only deviations from this baseline are represented. Here, the number C represents a fixed number for which the vehicles F1 to Fn have been reduced in each scene.
[0054] These metrics clearly show that for each value of the number C, using the trajectory prediction algorithm 5 based on the above interaction-based pre-selection A (F1 to Fn) leads to better results than when using distance-based selection, i.e., an improvement in the quality of the final trajectory prediction can be achieved compared to the approach using distance-based pre-selection.
[0055] The minimum mean displacement error change ΔminADE and the minimum final displacement error change ΔminFDE show improved performance compared to the final trajectory prediction algorithm 5 trained on the entire dataset due to the reduced complexity in the number C of 5, 7 and 9 vehicles F1-Fn, which can be seen from the negative values, i.e. an improved quality of the final trajectory prediction can be achieved compared to the approach without pre-selection.
[0056] Furthermore, by the interaction-based pre-selection A(F1-Fn), the execution time of the final trajectory prediction algorithm 5 can be achieved compared to the trajectory prediction algorithm 5 based on the entire scene including all the vehicles F1-Fn. [Prior art documents] [Non-patent literature]
[0057] [Non-Patent Document 1] Ye.
Claims
1. A method for predicting the trajectory of a vehicle (F1 to Fn) in a surrounding environment of the vehicle, comprising: - the interaction degrees (I1 to Im) between said vehicles (F1 to Fn) are determined using a machine-trained attention-based interaction algorithm (3); using said interaction algorithm (3), individual vehicles (F1-Fn) are identified from the set of vehicles (F1-Fn) present in the environment of said ego-vehicle as being relevant for said trajectory prediction and are selected for said trajectory prediction if their respective degrees of interaction (I1-Im) with at least one of said vehicles (F1-Fn) for which a trajectory (T) is predicted exceed a predefined limit value; - in a subsequent learning step, a trajectory prediction algorithm (5) is trained using said vehicles (F1 to Fn) selected as vehicles relevant for said trajectory prediction, the trajectory prediction carried out by means of the trajectory prediction algorithm (5) is carried out for the vehicles (F1 to Fn) selected as the vehicles relevant for said trajectory prediction; A method comprising:
2. 2. A method according to claim 1, characterized in that the degrees of interaction (I1 to Im) between the vehicles (F1 to Fn) are determined pairwise for each two vehicles (F1 to Fn).
3. 3. The method according to claim 1 or 2, characterized in that during the machine training of the interaction algorithm (3), the dependencies between the vehicles (F1-Fn) are learned pairwise and the extent to which the vehicles (F1-Fn) have a mutual influence is weighted pairwise to form an interaction degree (I1-Im) using the interaction algorithm (3).
4. Method according to any one of claims 1 to 3, characterised in that the machine training of the interaction algorithm (3) generates trajectory predictions which are used to train the interaction algorithm (3).
5. 5. The method according to claim 4, characterized in that the generated trajectory predictions are used in the machine training of the interaction algorithm (3) for implicit learning of the interaction degrees (I1-Im).
6. During the operation of the machine-trained interaction algorithm (3), all vehicles (F1 to Fn) in the scene are encoded using long and short-term memory, and during the encoding, the nodes (h 1 0 ~h n 0 6. The method according to claim 1, wherein a cation exchange layer is formed.
7. The node (h 1 0 ~h n 0 ) is the edge (e 1,2 , e 1,n , e 2,1 , e 2,n , e n,1 , e n,2 7. The method according to claim 6, characterized in that the vehicles (F1-Fn) are connected to each other via a distance between the vehicles (F1-Fn) and the distance between the vehicles (F1-Fn) is used as an edge feature.
8. Method according to any one of the preceding claims, characterised in that the trajectory prediction algorithm (5) uses map data (KD) of a high resolution digital road map for trajectory prediction.
9. A method for automatically driving a vehicle, comprising: A method according to any one of claims 1 to 8, in which a trajectory (T) of a vehicle (F1 to Fn) in the surrounding environment of the ego-vehicle is predicted, the predicted trajectory (T) is taken into account during automated lateral and / or longitudinal control of the ego vehicle in the automated driving of the ego vehicle, A method comprising:
Citation Information
Patent Citations
Vehicle control system, method thereof and travel support server
JP2018206036A
Vehicle control system and method, and travel support server
JP2019164812A
Systems and methods for providing cooperation-aware lane change control in dense traffic
JP2021046193A
Engine combined use electric excavator
KR102322221B1
Vehicle path prediction
US20200331465A1
