Parameter estimation method, device, equipment and storage medium for autonomous driving

Abstract

The present invention relates to the field of autonomous driving control and discloses a parameter estimation method, apparatus, device, and storage medium for autonomous driving. The method comprises: obtaining historical driving data and, based on the historical driving data, performing quantile modeling on obstacle hazardness to obtain a first evaluation model; evaluating obstacle aggressiveness using the first evaluation model based on obstacle setting parameters in the historical driving data to obtain an evaluation result, and generating a second evaluation model based on the evaluation result; obtaining a first aggressiveness value of a target obstacle and, based on the first aggressiveness value, calculating an estimated parameter corresponding to the target obstacle using the second evaluation model. The present invention improves the flexibility of obstacle parameter collection and the accuracy of evaluation.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving control, and in particular to a parameter estimation method, apparatus, device and storage medium for autonomous driving. Background Art

[0002] With the advancement of autonomous driving technology, the perception, computational decision-making, and control execution aspects of autonomous driving algorithms are constantly being upgraded. This has enabled autonomous driving algorithms to mature and adapt to a wider range of environments, handle complex driving scenarios, and avoid obstacles for safe navigation. The richness and accuracy of underlying driving data are crucial for the iteration of autonomous driving algorithms. The ability to acquire data from a wide range of driving scenarios to train autonomous driving algorithms ensures more stable and rapid iteration.

[0003] Current data collection for driving scenarios, especially data on various obstacles within them, typically relies on road testing or expanding upon this data. This obstacle data collection relies on the actual road environment, making it difficult to collect data on obstacle formations in unconventional or dangerous scenarios. This means existing obstacle parameter collection methods are inflexible. Summary of the Invention

[0004] The main purpose of the present invention is to solve the technical problem that the obstacle parameter collection method in the existing autonomous driving technology is not flexible enough.

[0005] A first aspect of the present invention provides a parameter estimation method for autonomous driving, comprising: obtaining historical driving data, and performing quantile modeling on obstacle hazardness based on the historical driving data to obtain a first evaluation model; based on the setting parameters of the obstacles in the historical driving data, using the first evaluation model to evaluate the obstacle aggressiveness to obtain an evaluation result, and generating a second evaluation model based on the evaluation result; obtaining a first aggressiveness value of a target obstacle, and using the second evaluation model to calculate the estimated parameters corresponding to the target obstacle based on the first aggressiveness value.

[0006] Optionally, in a first implementation manner of the first aspect of the present invention, the quantile modeling of the obstacle hazard is performed based on the historical driving data to obtain a first evaluation model, including: inputting the historical driving data into a preset pre-trained model, and extracting the dangerous behavior characteristics corresponding to the vehicle in each scene in the historical driving data through the input layer in the pre-trained model; estimating the collision duration between the vehicle and the obstacle in each scene respectively through the fully connected layer in the pre-trained model based on the dangerous behavior characteristics; estimating multiple percentile values ​​of the collision between the vehicle and the obstacle through the output layer in the pre-trained model based on the collision duration between the vehicle and the obstacle; and updating the pre-trained model based on the multiple percentile values ​​of the collision between the vehicle and the obstacle to obtain a first evaluation model.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the obstacle aggressiveness is evaluated based on the setting parameters of the obstacles in the historical driving data using the first evaluation model, and the evaluation results include: according to the setting parameters of the obstacles in the historical driving data, simulating the aggressive behavior information of the obstacles in each scene in the historical driving data; and according to the aggressive behavior information, evaluating the second aggressiveness value of the obstacles in each scene separately through the first evaluation model and using it as the evaluation result.

[0008] Optionally, in a third implementation manner of the first aspect of the present invention, the simulating and setting the aggressive behavior information of the obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data includes: adjusting the setting parameters of the obstacles in each scene in the historical driving data according to a preset parameter range to obtain at least one set of parameter combinations corresponding to each scene; based on the parameter combination, generating at least one simulated obstacle corresponding to each obstacle in each scene, and setting the aggressive behavior information of the simulated obstacle.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the evaluating, based on the aggressive behavior information, using the first evaluation model, respectively the second aggressiveness value of the obstacle in each scenario and providing the evaluation result includes: calculating, based on the aggressive behavior information, using the first evaluation model, respectively the collision duration between the obstacle and the vehicle in each scenario and the corresponding percentile value;

[0010] Compare the collision duration between the obstacle and the vehicle with the corresponding percentile value, and determine the maximum percentile value of the collision between the obstacle and the vehicle based on the comparison result; convert the maximum percentile value into the corresponding second aggressiveness value and use it as the evaluation result.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, generating a second evaluation model based on the evaluation result includes: using a preset gradient training model to fit the correlation between the setting parameters and the second aggressiveness value; updating the gradient training model according to the correlation until the gradient training model converges, thereby obtaining a second evaluation model.

[0012] Optionally, in a sixth implementation manner of the first aspect of the present invention, the calculation of the estimated parameters corresponding to the target obstacle based on the first aggressiveness value using the second evaluation model includes: using the second evaluation model to search for multiple groups of setting parameter sets within a preset approximate range that are associated with the first aggressiveness value, wherein each group of setting parameter sets corresponds to multiple setting parameters of the same type; cross-combining the setting parameters in each group of setting parameters, and calculating the third aggressiveness value corresponding to each cross-combination of the setting parameters; comparing the first aggressiveness value with each third aggressiveness value, and setting the cross-combination of the setting parameters corresponding to the third aggressiveness value that has the smallest difference from the first aggressiveness value as the estimated parameters corresponding to the target obstacle.

[0013] The second aspect of the present invention provides a parameter estimation device for autonomous driving, comprising: a modeling module for acquiring historical driving data, and performing quantile modeling of obstacle hazardness based on the historical driving data to obtain a first evaluation model; an evaluation module for evaluating the obstacle aggressiveness using the first evaluation model based on the setting parameters of the obstacle in the historical driving data, obtaining an evaluation result, and generating a second evaluation model based on the evaluation result; a calculation module for acquiring a first aggressiveness value of a target obstacle, and calculating, based on the first aggressiveness value, an estimated parameter corresponding to the target obstacle using the second evaluation model.

[0014] Optionally, in a first implementation manner of the second aspect of the present invention, the modeling module includes: an extraction unit for inputting the historical driving data into a preset pre-training model, and extracting the dangerous behavior characteristics corresponding to the vehicle in each scene in the historical driving data through the input layer in the pre-training model; a duration estimation unit for estimating the collision duration between the vehicle and the obstacle in each scene according to the dangerous behavior characteristics through the fully connected layer in the pre-training model; a hazard estimation unit for estimating multiple percentile values ​​of the collision between the vehicle and the obstacle according to the collision duration between the vehicle and the obstacle through the output layer in the pre-training model; and an updating unit for updating the pre-training model based on the multiple percentile values ​​of the collision between the vehicle and the obstacle to obtain a first evaluation model.

[0015] Optionally, in a second implementation of the second aspect of the present invention, the evaluation module includes: a setting unit for simulating and setting the aggressive behavior information of the obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data; an aggressiveness evaluation unit for evaluating the second aggressiveness value of the obstacles in each scene according to the aggressive behavior information through the first evaluation model and using it as the evaluation result.

[0016] Optionally, in a third implementation method of the second aspect of the present invention, the setting unit is further used to: adjust the setting parameters of the obstacles in each scene in the historical driving data according to a preset parameter range to obtain at least one set of parameter combinations corresponding to each scene; based on the parameter combination, generate at least one simulated obstacle corresponding to each obstacle in each scene, and set the aggressive behavior information of the simulated obstacle.

[0017] Optionally, in a fourth implementation of the second aspect of the present invention, the aggressiveness assessment unit is further used to: calculate the collision duration between the obstacle and the vehicle and the corresponding percentile value for each scenario based on the aggressive behavior information through the first assessment model; compare the collision duration between the obstacle and the vehicle with the corresponding percentile value, and determine the maximum percentile value of the collision between the obstacle and the vehicle based on the comparison result; convert the maximum percentile value into a corresponding second aggressiveness value and use it as an assessment result.

[0018] Optionally, in a fifth implementation of the second aspect of the present invention, the evaluation module further includes: a fitting unit, used to use a preset gradient training model to fit the correlation between the setting parameters and the second aggressiveness value; a generation unit, used to update the gradient training model according to the correlation until the gradient training model converges and stops to obtain a second evaluation model.

[0019] Optionally, in a sixth implementation of the second aspect of the present invention, the calculation module includes: a search unit, used to use the second evaluation model to search for multiple groups of setting parameter sets within a preset approximate range that are associated with the first aggression value, wherein each group of setting parameter sets corresponds to multiple setting parameters of the same type; a calculation unit, used to cross-combine the setting parameters in each group of setting parameters, and calculate the third aggressiveness value corresponding to each cross-combination of the setting parameters; a comparison unit, used to compare the first aggression value and each third aggression value, and set the cross-combination of the setting parameters corresponding to the third aggression value that has the smallest difference from the first aggression value as the estimated parameter corresponding to the target obstacle.

[0020] The third aspect of the present invention provides a parameter estimation device for autonomous driving, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the parameter estimation device for autonomous driving executes the above-mentioned parameter estimation method for autonomous driving.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions that, when executed on a computer, enable the computer to execute the above-mentioned parameter estimation method for autonomous driving.

[0022] The technical solution provided by this invention uses historical driving data to create quantile modeling of obstacle hazard. The resulting first assessment model is then used to evaluate the aggressiveness of the defined simulated obstacle, and a second assessment model is then fitted. Finally, after the required aggressiveness is determined, obstacle parameters are directly estimated based on this second assessment model. This objective definition of the simulated obstacle's aggressiveness is achieved by comparing the dangerousness of the simulated obstacle with that in the historical driving data. For a specific obstacle with a defined aggressiveness, the obstacle's parameters can be accurately estimated, improving the flexibility and accuracy of obstacle parameter collection. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of a first embodiment of a parameter estimation method for autonomous driving according to an embodiment of the present invention;

[0024] Figure 2 2 is a schematic diagram of a second embodiment of a parameter estimation method for autonomous driving according to an embodiment of the present invention;

[0025] Figure 3 2 is a schematic diagram of a third embodiment of a parameter estimation method for autonomous driving according to an embodiment of the present invention;

[0026] Figure 4 Schematic diagram of an embodiment of a parameter estimation device for autonomous driving according to an embodiment of the present invention;

[0027] Figure 5 2 is a schematic diagram of another embodiment of a parameter estimation device for autonomous driving according to an embodiment of the present invention;

[0028] Figure 6 Schematic diagram of an embodiment of a parameter estimation device for autonomous driving in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] Embodiments of the present invention provide a parameter estimation method, apparatus, device, and storage medium for autonomous driving. These methods obtain historical driving data and perform quantile modeling of obstacle hazardness based on the historical driving data to obtain a first evaluation model. The first evaluation model is used to evaluate obstacle aggressiveness based on obstacle setting parameters in the historical driving data, obtaining an evaluation result, and a second evaluation model is generated based on the evaluation result. Furthermore, a first aggressiveness value of a target obstacle is obtained and, based on the first aggressiveness value, the second evaluation model is used to calculate the estimated parameters corresponding to the target obstacle. This present invention improves the flexibility of obstacle parameter collection and the accuracy of evaluation.

[0030] The terms "first," "second," "third," "fourth," and the like (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A first embodiment of a parameter estimation method for autonomous driving according to an embodiment of the present invention includes:

[0032] 101. Obtain historical driving data, and perform quantile modeling on obstacle hazard based on the historical driving data to obtain a first assessment model;

[0033] It is understood that the execution subject of the present invention can be an autonomous driving parameter estimation device, a terminal, or a server, and the specific implementation is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0034] In this embodiment, the historical driving data can be data collected by the autonomous vehicle during past road tests, or data collected by the autonomous vehicle during simulated driving in software, including information related to the host vehicle, obstacles, and maps. For example, the host vehicle's speed, acceleration, and driving direction; the static and dynamic information, speed and acceleration of obstacles; and road data.

[0035] In this embodiment, after obtaining historical driving data, the quantiles of the obstacle's danger are modeled based on the collected information. Here, the obstacle danger refers to the danger posed by the obstacle to the autonomous driving vehicle. The modeling result is a danger value of a preset quantile (such as the 100th percentile), and a first evaluation model is obtained, in which the danger value can be represented by parameters such as THW (Time Headway) and TTC (Time to Collision).

[0036] 102. Based on the setting parameters of the obstacles in the historical driving data, using the first evaluation model to evaluate the aggressiveness of the obstacles, obtain an evaluation result, and generate a second evaluation model based on the evaluation result;

[0037] In this embodiment, multiple scenario-defining parameters are pre-set as obstacle parameters, including the size of the obstacle padding around the host vehicle (range 0 to 5); the maximum deceleration the host vehicle can achieve to avoid a collision (range -6 to 0, in km / h); and the minimum deceleration required to avoid a collision (range -6 to 0, in km / h). These three parameters are randomly combined within this range to generate a preset number of simulation scenarios for each scenario in the historical driving data.

[0038] After generating multiple sets of parameters randomly generated according to the set parameters and corresponding simulation scenarios, the first evaluation model is used to evaluate the aggressiveness of the obstacle to the main vehicle. The aggressiveness of the obstacle can be defined by the relative danger of the obstacle, that is, the first evaluation model is directly used to evaluate the aggressiveness of the obstacle, and the final evaluation result of the aggressiveness of the obstacle is determined according to the transformation with the relative danger in the first evaluation model.

[0039] In this embodiment, after the evaluation result of the aggressiveness of the obstacle is generated, the relationship between the input setting parameters, that is, the relationship between the input setting parameters and the evaluation result, is fitted as the second evaluation model.

[0040] 103. Obtain a first aggressiveness value of the target obstacle, and calculate an estimated parameter corresponding to the target obstacle using the second evaluation model based on the first aggressiveness value.

[0041] In this embodiment, after the second evaluation model is obtained, the application phase can be carried out, where a first aggressiveness value of the target obstacle is obtained, and the simulation parameters of the driving data are found to achieve the first aggressiveness value.

[0042] For example, for obstacles like vehicles, we want to find the average aggressiveness, that is, the setting with a first aggressiveness value of 0.7. Then, based on the three first aggressiveness values, we reverse-calculate the corresponding setting parameter combinations using the second evaluation model, such as the padding size of 4 around the host vehicle, the maximum deceleration of the host vehicle to avoid a collision (assuming the maximum possible deceleration of the host vehicle to avoid a collision is -0.4 km / h), and the minimum deceleration required by the obstacle to avoid a collision (range: -5 km / h).

[0043] In this embodiment of the present invention, historical driving data is used to generate quantile modeling of obstacle hazard. The resulting first assessment model is then used to evaluate the aggressiveness of the defined simulated obstacle, and a second assessment model is generated through fitting. Finally, after the required aggressiveness is determined, obstacle parameters are directly estimated based on this second assessment model. This objective definition of the aggressiveness of the simulated obstacle is achieved by comparing the dangerousness of the simulated obstacle with that of the historical driving data. For a specific obstacle with a defined aggressiveness, the obstacle parameters can be accurately estimated, improving the flexibility and accuracy of obstacle parameter collection.

[0044] See also Figure 2 A second embodiment of the parameter estimation method for autonomous driving according to the present invention includes:

[0045] 201. Obtain historical driving data, input the historical driving data into a preset pre-trained model, and extract dangerous behavior features corresponding to vehicles in various scenarios in the historical driving data through an input layer of the pre-trained model;

[0046] 202. Estimate the collision duration between the vehicle and the obstacle in each of the scenarios using the fully connected layer in the pre-trained model based on the dangerous behavior characteristics;

[0047] 203. Estimate multiple percentile values ​​of the collision between the vehicle and the obstacle through the output layer of the pre-trained model according to the collision duration between the vehicle and the obstacle;

[0048] 204. Based on multiple percentile values ​​of collision between the vehicle and the obstacle, update the pre-trained model to obtain a first evaluation model;

[0049] In this embodiment, when building the first evaluation model, the pre-set pre-trained model is fine-tuned through the evaluation set. The evaluation set of the pre-trained model is the historical driving data, and the first evaluation model is trained through the historical driving data.

[0050] In this embodiment, the input layer of the pre-trained model is first defined to extract dangerous behaviors for obstacle hazard assessment. The dangerous behavior characteristics may include the relative position between the obstacle and the main vehicle, the type of obstacle, and the speed of the main vehicle. Then, the fully connected layer is used to estimate the collision time (TTC) between the vehicle and the obstacle in each scenario of the evaluation set. Finally, the output layer is used to classify the collision time estimated by the fully connected layer, and the numerical distribution of the collision time at each percentile. For example, given a set of input dangerous behavior characteristics, the value of the fifth percentile in the distribution of the obstacle TTC can be estimated when this set of information is met.

[0051] Specifically, refer to Figure 3 The diagram below shows the architecture of the first evaluation model. The model consists of an input layer, three fully connected layers, and an output layer. The widths of the fully connected layers can be 64, 32, and 16, respectively. The output layer has 99 neurons, each used to estimate the 99th percentile. Each neuron in the output layer uses the pinball loss function with corresponding parameters to simultaneously estimate different percentiles.

[0052] 205. Simulate and set the aggressive behavior information of the obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data;

[0053] 206. Based on the aggressive behavior information, use the first evaluation model to evaluate the second aggressiveness value of the obstacle in each scene and use it as an evaluation result;

[0054] In this embodiment, by defining obstacle variables in historical driving data, which may include the relative position of the obstacle to the host vehicle, the obstacle type, and the host vehicle speed, and then randomly generating setting parameters within a range of values, scenarios with different obstacle behaviors are simulated. In multiple simulation scenarios, the aggressiveness of different obstacle behaviors, i.e., the second aggressiveness value, is estimated through the modeled first evaluation model.

[0055] Specifically, when simulating the aggressive behavior information of obstacles in various scenes in the historical driving data, the following steps are included:

[0056] 1) adjusting the obstacle setting parameters for each scene in the historical driving data according to a preset parameter range to obtain at least one set of parameter combinations corresponding to each scene;

[0057] 2) Based on the parameter combination, at least one simulated obstacle corresponding to each obstacle in each scene is generated, and aggressive behavior information of the simulated obstacle is set.

[0058] In this embodiment, when simulating scenarios with different obstacle behaviors, a set of setting parameters is randomly generated for each of the three setting parameters, and a set of parameter combinations is used as a simulation scenario. For example, if (padding around the main vehicle, maximum deceleration of the main vehicle, maximum deceleration of the obstacle) is (2, -3, -4), then it is a set of parameter combinations used as the aggressive behavior information of the simulated obstacle.

[0059] Specifically, when the first evaluation model is used to evaluate the second aggressiveness value of obstacles in each scene, the following steps are included:

[0060] 1) Based on the aggressive behavior information, using the first assessment model, calculate the collision duration between the obstacle and the vehicle and the corresponding percentile value for each scenario;

[0061] 2) Comparing the collision duration between the obstacle and the vehicle with the corresponding percentile value, and determining the maximum percentile value of the collision between the obstacle and the vehicle based on the comparison result;

[0062] 3) Convert the maximum percentile value into a corresponding second aggressiveness value and use it as the evaluation result.

[0063] In this embodiment, the aggressiveness of an obstacle can be defined using relative risk: if the TTC of an obstacle is less than the TTC of X% of obstacles in similar situations, its second aggressiveness value is X / 100. For example, if the TTC of a vehicle is less than 95% of vehicles in similar situations, its second aggressiveness value is 0.95.

[0064] Furthermore, each obstacle in each scenario that is close to the host vehicle (within a distance of less than 10 meters) can be evaluated using the first evaluation model. The TTC in the simulated scenario is then compared with the corresponding TTC percentile to calculate the aggressiveness of each obstacle.

[0065] 207. Using a preset gradient training model, fitting the correlation between the setting parameter and the second aggressiveness value;

[0066] 208. Update the gradient training model according to the association relationship until the gradient training model converges, thereby obtaining a second evaluation model.

[0067] In this embodiment, a historical driving data set is used as an evaluation set when building a first evaluation model, which includes a large number of scenarios. When the obstacle behavior of the scenario is adjusted by setting parameters, each scenario generates corresponding multiple simulation scenarios. After evaluation by the first evaluation model, the second aggressiveness value obtained by evaluating each group of parameter combinations randomly generated by setting parameters can be determined. Here, the relationship between the parameter combination and the second aggressiveness value can be fitted through a gradient training model.

[0068] Specifically, for each type of obstacle, the relationship between the setting parameters and the second aggressiveness value can be fitted by training a gradient boosted tree (Gradient BoostedTrees) model. First, initialize the relationship between the setting parameters and the second aggressiveness value, and estimate the constant value that minimizes the loss function, which is the constant value of the gradient training model. Then calculate the value of the negative gradient of the preset loss function in the current gradient training model and use it as an estimate of the residual. Then estimate the regression tree leaf node area to fit the approximate value of the residual. Then use linear search to estimate the value of the leaf node area to minimize the loss function. Update the regression tree through the loss function combined with the correlation relationship to obtain the final output second evaluation model to evaluate the relationship between the setting parameters and the second aggressiveness value.

[0069] 209. Obtain a first aggressiveness value of the target obstacle, and use the second evaluation model to search for multiple sets of setting parameters within a preset approximate range that are correlated with the first aggressiveness value, where each set of setting parameters corresponds to multiple setting parameters of the same type.

[0070] 210. Cross-combining the setting parameters in each set of setting parameters, and calculating the third aggressiveness value corresponding to each cross-combination of the setting parameters;

[0071] 211. Compare the first aggression value with each third aggression value, and set the setting parameter of the cross combination corresponding to the third aggression value having the smallest difference from the first aggression value as the estimation parameter corresponding to the target obstacle.

[0072] In this embodiment, when searching for estimated parameters corresponding to the input first aggression value using the second evaluation model, in addition to directly using the second evaluation model for evaluation, a grid method can also be used within the trained model to find combinations of setting parameters corresponding to the desired first aggression value (aggression level) to improve evaluation diversity and accuracy. For example, for obstacles such as vehicles, if the input first aggression values ​​are 0.7, 0.8, and 0.9, a grid is generated within the value range of each setting parameter (e.g., approximately 10 equally spaced points including the maximum and minimum values). The grid then searches for setting parameter combinations and calculates the corresponding third aggression value for each setting parameter combination such that the corresponding fitted aggression is closest to the first aggression values ​​of 0.7, 0.8, and 0.9.

[0073] In this embodiment of the present invention, a quantile model is used to extract obstacle behavior characteristics from historical driving data. The results are used to define the aggressiveness of each obstacle behavior in the simulation, and then appropriate setting parameters are found to set the obstacles in the simulation evaluation set. When the aggressiveness level of the obstacle is known, the obstacle parameters can be set more flexibly, comprehensively, and accurately.

[0074] The above describes the parameter estimation method of the automatic driving in the embodiment of the present invention. The following describes the parameter estimation device of the automatic driving in the embodiment of the present invention. Figure 4 In one embodiment of the present invention, an apparatus for estimating parameters of an autonomous driving system includes:

[0075] A modeling module 401 is configured to obtain historical driving data and perform quantile modeling of obstacle hazard based on the historical driving data to obtain a first evaluation model;

[0076] an evaluation module 402 for evaluating the aggressiveness of obstacles using the first evaluation model based on the setting parameters of the obstacles in the historical driving data, obtaining an evaluation result, and generating a second evaluation model based on the evaluation result;

[0077] The calculation module 403 is configured to obtain a first aggressiveness value of the target obstacle, and calculate an estimation parameter corresponding to the target obstacle using the second evaluation model according to the first aggressiveness value.

[0078] In this embodiment of the present invention, historical driving data is used to generate quantile modeling of obstacle hazard. The resulting first assessment model is then used to evaluate the aggressiveness of the defined simulated obstacle, and a second assessment model is generated through fitting. Finally, after the required aggressiveness is determined, obstacle parameters are directly estimated based on this second assessment model. This objective definition of the aggressiveness of the simulated obstacle is achieved by comparing the dangerousness of the simulated obstacle with that of the historical driving data. For a specific obstacle with a defined aggressiveness, the obstacle parameters can be accurately estimated, improving the flexibility and accuracy of obstacle parameter collection.

[0079] See also Figure 5 Another embodiment of the parameter estimation device for autonomous driving according to the embodiment of the present invention includes:

[0080] A modeling module 401 is configured to obtain historical driving data and perform quantile modeling of obstacle hazard based on the historical driving data to obtain a first evaluation model;

[0081] an evaluation module 402 for evaluating the aggressiveness of obstacles using the first evaluation model based on the setting parameters of the obstacles in the historical driving data, obtaining an evaluation result, and generating a second evaluation model based on the evaluation result;

[0082] The calculation module 403 is configured to obtain a first aggressiveness value of the target obstacle, and calculate an estimation parameter corresponding to the target obstacle using the second evaluation model according to the first aggressiveness value.

[0083] Specifically, the modeling module 401 includes:

[0084] An extraction unit 4011 is configured to input the historical driving data into a preset pre-trained model, and extract the dangerous behavior features corresponding to the vehicles in each scene in the historical driving data through an input layer in the pre-trained model;

[0085] a duration estimation unit 4012 for estimating the collision duration between the vehicle and the obstacle in each of the scenarios based on the dangerous behavior characteristics and using the fully connected layer in the pre-trained model;

[0086] a risk estimation unit 4013 for estimating, based on the collision duration between the vehicle and the obstacle, a plurality of percentile values ​​of the collision probability between the vehicle and the obstacle through the output layer of the pre-trained model;

[0087] The updating unit 4014 is configured to update the pre-trained model based on a plurality of percentile values ​​of collisions between the vehicle and the obstacle to obtain a first evaluation model.

[0088] Specifically, the evaluation module 402 includes:

[0089] A setting unit 4021 is configured to simulate and set the aggressive behavior information of obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data;

[0090] The aggressiveness evaluation unit 4022 is configured to evaluate the second aggressiveness value of the obstacle in each scene according to the aggressiveness behavior information and using the first evaluation model as an evaluation result.

[0091] Specifically, the setting unit 4021 is further configured to:

[0092] Adjusting the obstacle setting parameters for each scene in the historical driving data according to a preset parameter range to obtain at least one set of parameter combinations corresponding to each scene;

[0093] Based on the parameter combination, at least one simulated obstacle corresponding to each obstacle in each scene is generated, and aggressive behavior information of the simulated obstacle is set.

[0094] Specifically, the aggressiveness assessment unit 4022 is further configured to:

[0095] Calculating, using the first evaluation model, the collision duration between the obstacle and the vehicle and the corresponding percentile value for each scenario based on the aggressive behavior information;

[0096] Comparing the collision duration between the obstacle and the vehicle with the corresponding percentile value, and determining the maximum percentile value of the collision between the obstacle and the vehicle based on the comparison result;

[0097] The maximum percentile value is converted into a corresponding second aggressiveness value and used as an evaluation result.

[0098] Specifically, the evaluation module 402 further includes:

[0099] A fitting unit 4023 is configured to fit the correlation between the setting parameter and the second aggressiveness value using a preset gradient training model;

[0100] The generating unit 4024 is configured to update the gradient training model according to the association relationship until the gradient training model converges, thereby obtaining a second evaluation model.

[0101] Specifically, the calculation module 403 includes:

[0102] a searching unit 4031 configured to use the second assessment model to search for a plurality of setting parameter sets within a preset approximate range that are correlated with the first aggression value, wherein each setting parameter set corresponds to a plurality of setting parameters of the same type;

[0103] a calculation unit 4032, configured to cross-combine the setting parameters in each set of setting parameters and calculate a third aggressiveness value corresponding to each cross-combination of the setting parameters;

[0104] The comparing unit 4033 is configured to compare the first aggression value with each third aggression value, and set the setting parameter of the cross combination corresponding to the third aggression value having the smallest difference from the first aggression value as the estimation parameter corresponding to the target obstacle.

[0105] In this embodiment of the present invention, a quantile model is used to extract obstacle behavior characteristics from historical driving data. The results are used to define the aggressiveness of each obstacle behavior in the simulation, and then appropriate setting parameters are found to set the obstacles in the simulation evaluation set. When the aggressiveness level of the obstacle is known, the obstacle parameters can be set more flexibly, comprehensively, and accurately.

[0106] above Figure 4 and Figure 5The parameter estimation device for autonomous driving in an embodiment of the present invention is described in detail from the perspective of modular functional entities. The parameter estimation device for autonomous driving in an embodiment of the present invention is described in detail from the perspective of hardware processing.

[0107] Figure 6 : This is a schematic diagram of the structure of an autonomous driving parameter estimation device provided by an embodiment of the present invention. The autonomous driving parameter estimation device 600 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (for example, one or more processors) and memory 620, and one or more storage media 630 (for example, one or more mass storage devices) storing application programs 633 or data 632. The memory 620 and storage medium 630 may be temporary storage or permanent storage. The program stored in the storage medium 630 may include one or more modules (not shown in the figure), each of which may include a series of instruction operations in the autonomous driving parameter estimation device 600. Furthermore, the processor 610 may be configured to communicate with the storage medium 630 to execute the series of instruction operations in the storage medium 630 on the autonomous driving parameter estimation device 600.

[0108] The parameter estimation device 600 for autonomous driving may further include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 6 The structure of the parameter estimation device for autonomous driving shown does not constitute a limitation on the parameter estimation device for autonomous driving, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0109] The present invention also provides a parameter estimation device for autonomous driving, wherein the computer device includes a memory and a processor, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the parameter estimation method for autonomous driving in the above-mentioned embodiments.

[0110] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the parameter estimation method for autonomous driving.

[0111] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0112] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc., various media that can store program code.

[0113] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A parameter estimation method for autonomous driving, characterized in that: The parameter estimation method for autonomous driving includes: Acquiring historical driving data, and performing quantile modeling on obstacle hazardness based on the historical driving data to obtain a first assessment model; Based on the setting parameters of the obstacles in the historical driving data, using the first evaluation model to evaluate the aggressiveness of the obstacles to obtain an evaluation result, and generating a second evaluation model based on the evaluation result; Obtaining a first aggressiveness value of a target obstacle, and calculating an estimated parameter corresponding to the target obstacle using the second evaluation model based on the first aggressiveness value; The obstacle aggressiveness is evaluated using the first evaluation model based on the setting parameters of the obstacles in the historical driving data, and the evaluation results include: simulating the aggressive behavior information of the obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data; and evaluating the second aggressiveness value of the obstacles in each scene according to the aggressive behavior information through the first evaluation model and using it as the evaluation result.

2. The parameter estimation method for autonomous driving according to claim 1, characterized in that: The step of performing quantile modeling on the obstacle hazard based on the historical driving data to obtain a first evaluation model includes: Inputting the historical driving data into a preset pre-training model, and extracting the dangerous behavior features corresponding to the vehicles in each scene in the historical driving data through the input layer of the pre-training model; According to the dangerous behavior characteristics, the collision duration between the vehicle and the obstacle in each scenario is estimated through the fully connected layer in the pre-trained model; estimating, based on the collision duration between the vehicle and the obstacle, a plurality of percentile values ​​of the collision between the vehicle and the obstacle through the output layer of the pre-trained model; The pre-trained model is updated based on multiple percentile values ​​of collisions between the vehicle and the obstacle to obtain a first evaluation model.

3. The parameter estimation method for autonomous driving according to claim 1, wherein: The simulating and setting the aggressive behavior information of the obstacles in each scene in the historical driving data according to the setting parameters of the obstacles in the historical driving data includes: Adjusting the obstacle setting parameters for each scene in the historical driving data according to a preset parameter range to obtain at least one set of parameter combinations corresponding to each scene; Based on the parameter combination, at least one simulated obstacle corresponding to each obstacle in each scene is generated, and aggressive behavior information of the simulated obstacle is set.

4. The parameter estimation method for autonomous driving according to claim 1, wherein: The second aggressiveness value of the obstacle in each scene is evaluated respectively by the first evaluation model based on the aggressive behavior information, and the evaluation results include: Calculating, using the first evaluation model, the collision duration between the obstacle and the vehicle and the corresponding percentile value for each scenario based on the aggressive behavior information; Comparing the collision duration between the obstacle and the vehicle with the corresponding percentile value, and determining the maximum percentile value of the collision between the obstacle and the vehicle based on the comparison result; The maximum percentile value is converted into a corresponding second aggressiveness value and used as an evaluation result.

5. The parameter estimation method for autonomous driving according to any one of claims 3 to 4, characterized in that: Generating a second evaluation model based on the evaluation result includes: Using a preset gradient training model to fit the correlation between the setting parameter and the second aggressiveness value; The gradient training model is updated according to the association relationship until the gradient training model converges, thereby obtaining a second evaluation model.

6. The parameter estimation method for autonomous driving according to claim 5, characterized in that: Calculating the estimated parameter corresponding to the target obstacle using the second evaluation model according to the first aggressiveness value includes: Using the second evaluation model, searching for multiple sets of setting parameters within a preset approximate range that are correlated with the first aggressiveness value, wherein each set of setting parameters corresponds to multiple setting parameters of the same type; Cross-combining the setting parameters in each set of setting parameters, and calculating the third aggressiveness value corresponding to each cross-combination of the setting parameters; The first aggressiveness value is compared with each third aggressiveness value, and the setting parameter of the cross combination corresponding to the third aggressiveness value having the smallest difference from the first aggressiveness value is set as the estimated parameter corresponding to the target obstacle.

7. A parameter estimation device for autonomous driving, characterized in that: The automatic driving parameter estimation device comprises: a modeling module, configured to obtain historical driving data and perform quantile modeling of obstacle hazardness based on the historical driving data to obtain a first evaluation model; an evaluation module configured to evaluate the aggressiveness of the obstacle using the first evaluation model based on the setting parameters of the obstacle in the historical driving data, obtain an evaluation result, and generate a second evaluation model based on the evaluation result; a calculation module for obtaining a first aggressiveness value of a target obstacle and, based on the first aggressiveness value, calculating an estimated parameter corresponding to the target obstacle using the second evaluation model; the evaluation module comprising: a setting unit for simulating and setting aggressiveness behavior information of obstacles for each scenario in the historical driving data based on setting parameters of the obstacles in the historical driving data; and an aggressiveness evaluation unit for evaluating, based on the aggressiveness behavior information and using the first evaluation model, the second aggressiveness value of the obstacle for each scenario as an evaluation result.

8. A parameter estimation device for autonomous driving, characterized in that: The parameter estimation device for autonomous driving includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the autonomous driving parameter estimation device performs the steps of the autonomous driving parameter estimation method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the parameter estimation method for autonomous driving as described in any one of claims 1 to 6 are implemented.

Images