CN113954870B

CN113954870B - Automatic driving vehicle behavior decision optimization system based on digital twinning technology

Info

Publication number: CN113954870B
Application number: CN202111248526.8A
Authority: CN
Inventors: 庞诏文; 陈振斌; 卢家怿; 冯鑫杰; 高灵飞; 杨峥
Original assignee: Hainan University
Current assignee: Hainan University
Priority date: 2021-10-26
Filing date: 2021-10-26
Publication date: 2023-03-03
Anticipated expiration: 2041-10-26
Also published as: CN113954870A

Abstract

The application provides an automatic driving vehicle behavior decision optimization system based on a digital twin technology. The system comprises: a digital twin platform and a parallel processing cloud platform. The digital twin platform is used for establishing a digital twin system corresponding to the first vehicle and simulating a first behavior decision to be executed by the first vehicle by using the digital twin system. The parallel processing cloud platform judges whether the first behavior decision meets the safety condition or not based on simulation data obtained by the simulation of the digital twin platform, and obtains a second behavior decision of the first vehicle again by using the decision optimization model under the condition that the first behavior decision does not meet the safety condition. The system carries out real-time simulation calculation on the behavior decision of the first vehicle through the digital twin platform, and realizes real-time dynamic optimization on the behavior decision through the parallel processing cloud platform, so that the safety of automatic driving of the first vehicle in the whole life cycle can be improved.

Description

Automatic driving vehicle behavior decision optimization system based on digital twin technology

Technical Field

The application relates to the technical field of automatic driving, in particular to an automatic driving vehicle behavior decision optimization system based on a digital twin technology.

Background

With the deep advance of internet, big data, cloud computing and artificial intelligence technologies, the current society enters the intelligent era of mutual object interconnection and mutual object intelligence. The automobile industry is deeply integrated with new-generation information technologies such as high-performance computing chips, artificial intelligence, internet of things and the like, and the automatic driving technology gradually becomes the focus of the industry. The automatic driving can effectively reduce traffic risks, improve driving safety, save a large amount of social time cost and labor cost, improve urban operation efficiency and is a time for upgrading the industry with one-time changing personality of the automobile manufacturing industry and automobile travel service.

The automatic driving automobile integrates multiple disciplines such as vehicle engineering, artificial intelligence, computer science, automatic control and the like, and relates to multiple disciplines, multiple scenes and multiple fields. At present, in automatic driving, when a finite-state machine model is used for making a decision on the behavior of a vehicle, the finite-state machine model cannot make a targeted decision according to the time-varying state and environment of the vehicle, so that a decision result has certain limitation, and the driving safety of the vehicle can be reduced.

Disclosure of Invention

The application provides an automatic driving vehicle behavior decision optimizing system based on a digital twin technology, safety verification is carried out on the behavior decision of a vehicle through a digital twin system corresponding to the vehicle, the behavior decision of the vehicle is determined again under the condition that the verification is not passed, and the driving safety of the vehicle can be improved.

In a first aspect, an embodiment of the present application provides an automated driving vehicle behavior decision optimization system. The system comprises: a digital twin platform and a parallel processing cloud platform connected with the first vehicle.

The digital twin platform is used for building a digital twin system corresponding to the first vehicle by using a digital twin technology, and configuring a simulation environment of the digital twin system based on first data of the first vehicle at the current time and first data of historical time; the first data comprises a first behavioral decision and second data; the first data is determined by the first vehicle.

Wherein the digital twin platform is further configured to simulate the first vehicle with the digital twin system to execute a first behavior decision of the first vehicle at the current time in the simulation environment, determine simulation data of the first vehicle; the simulation data is indicative of a safety of the first vehicle;

the parallel processing cloud platform is used for judging whether the first behavior decision of the current time meets a safety condition or not based on the simulation data, and determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data of the current time under the condition that the safety condition is not met.

According to the scheme, the simulation environment of the digital twin system is configured according to the first data of the current time and the historical time, so that the digital twin system can be consistent with the actual working environment of the first vehicle, and the modeling and simulation of the full life cycle of the first vehicle are realized. The method comprises the steps of simulating a first behavior decision of a vehicle in real time by using a digital twin system, judging whether the first behavior decision meets a safety condition or not by parallelly processing real-time simulation data based on the digital twin platform of a cloud platform, and obtaining a second behavior decision of the vehicle again under the condition that the first behavior decision does not meet the safety condition so as to improve the safety of automatic driving of the vehicle.

In a possible implementation, the parallel processing cloud platform is further configured to send the second behavior decision to the first vehicle and send the second behavior decision to the digital twin platform; wherein the digital twin platform simulates the first vehicle with the digital twin system performing the second behavioral decision

In a possible implementation manner, the vehicle includes a sensing module, a decision module, a chassis control module, and/or a domain controller module, and the digital twin system includes a sensing model corresponding to the sensing module, a decision model corresponding to the decision module, a chassis control model corresponding to the chassis control module, and/or a domain control model corresponding to the domain controller module.

In one possible embodiment, the second data includes: control signals, perception data, planning data, driving state data and intermediate calculation data.

The perception data are collected by the perception module and comprise road data, radar data, environment data and map data.

Wherein the sensing module may include: the system comprises an image sensor for collecting road traffic data and map data, a radar for collecting radar data and a first sensor for collecting environmental data. The application does not limit the type and number of sensors included in the sensing module.

In a possible embodiment, the first behavior decision is determined by the decision module from the second data.

In one possible embodiment, the determining, using a decision optimization model, a second behavior decision of the first vehicle at the current time based on the first data at the current time comprises:

extracting feature data from the first data of the current time;

and inputting the characteristic data into the decision optimization model to obtain a second behavior decision of the current time output by the decision optimization model.

In one possible embodiment, the parallel processing cloud platform is further configured to store first data of the first vehicle at a historical time, and iteratively update parameters of the decision optimization model according to the first data at the historical time.

In one possible implementation, the parallel processing cloud platform is further configured to:

predicting first data of the first vehicle at a next time based on the first data of the current time;

determining, using the decision optimization model, a second behavioral decision for the first vehicle at the next time based on the first data at the next time.

In a second aspect, the application further provides an optimization method for behavior decision of an autonomous vehicle. The method comprises the following steps:

acquiring first data of a first vehicle at the current time and the historical time; the first data comprises a first behavioral decision and second data;

configuring a simulation environment of the digital twin system based on first data of the first vehicle at the current time and the historical time, wherein the digital twin system is built by using a digital twin technology;

simulating the first vehicle to execute a first behavior decision at the current time by using the digital twin system in the simulation environment, and determining simulation data of the first vehicle; the simulation data is indicative of safety of the first vehicle;

and under the condition that the first behavior decision of the current time does not meet the safety condition based on the simulation data, determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data of the current time.

In a third aspect, the application further provides an automatic driving vehicle behavior decision optimization device. The device includes:

the acquisition module is used for acquiring first data of a first vehicle at the current time and the historical time; the first data comprises a first behavioral decision and second data;

a configuration module to configure a simulation environment of the digital twin system based on first data of the first vehicle at a current time and a historical time; the digital twinning system is built by utilizing a digital twinning technology;

a simulation module, configured to simulate, in the simulation environment, the first vehicle to execute a first behavior decision of the current time by using the digital twin system, and determine simulation data of the first vehicle; the simulation data is indicative of safety of the first vehicle;

and the optimization module is used for determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data of the current time under the condition that the first behavior decision of the current time is judged not to meet safety conditions based on the simulation data.

Drawings

FIG. 1 is a flow chart of a method for optimizing an automated driving vehicle behavior decision provided by an embodiment of the present application;

FIG. 2 is a flowchart of a training method of a decision optimization model according to an embodiment of the present disclosure;

FIG. 3 is a schematic structural diagram of an automated driving vehicle behavior decision optimization provided by an embodiment of the present application;

FIG. 4 is a schematic structural diagram of an automatic driving vehicle behavior decision optimizing device according to an embodiment of the present disclosure;

FIG. 5 is a schematic structural diagram of a model training apparatus for training a decision optimization model according to an embodiment of the present disclosure;

fig. 6 is a schematic structural diagram of a computing device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

In the description of the embodiments of the present application, the words "exemplary," "for example," or "for instance" are used to mean serving as an example, instance, or illustration. Any embodiment or design described herein as "exemplary," "for example," or "for example" is not necessarily to be construed as preferred or advantageous over other embodiments or designs. Rather, use of the words "exemplary," "e.g.," or "exemplary" is intended to present relevant concepts in a concrete fashion.

In the description of the embodiments of the present application, the term "and/or" is only one kind of association relationship describing an associated object, and means that three relationships may exist, for example, a and/or B may mean: a exists alone, B exists alone, and A and B exist at the same time. In addition, the term "plurality" means two or more unless otherwise specified. For example, the plurality of systems refers to two or more systems, and the plurality of screen terminals refers to two or more screen terminals.

Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or to implicitly indicate the indicated technical feature. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically stated.

The automatic driving is a comprehensive system integrating perception, decision planning and control execution functions, fully considers the coordination planning of vehicles and traffic environment, and is an important component of a future intelligent traffic system. The behavior decision belongs to a decision planning layer, and is to perform behavior judgment according to acquired sensing data, perform behavior prediction on the position, speed, direction and the like of an identified object, and make a corresponding control strategy of the vehicle. The automatic driving specifically converts the control strategy into information of time dimension and space dimension, and outputs a macroscopic decision instruction based on the information for a vehicle execution module to execute more specifically, so that the aim of making a driving decision instead of a human driver can be fulfilled.

In the case where the vehicle predicts the driving behavior using the finite state machine model, different driving states and a conversion relationship between the driving states may be described by constructing a finite directed communication diagram to obtain the finite state machine model, which generates a driving state at the next time according to the current driving state. The model has the advantages of simplicity and feasibility, and has the defects of neglecting the dynamic property and uncertainty of the environment and being only suitable for simpler scenes.

Based on the above analysis, the embodiment of the present application provides a vehicle behavior decision method based on a digital twin. The method can be used for decision equipment, and aims to verify the behavior decision of the vehicle by using a digital twin system of the vehicle, and re-determine the behavior decision of the vehicle under the condition that the set safety condition is not met, so that the driving safety of the vehicle is improved.

Fig. 1 is a method for optimizing an automatic driving vehicle behavior decision according to an embodiment of the present disclosure. As shown in fig. 1, the method includes steps S101 to S104 as follows.

Step S101, first data of the first vehicle at the current time and the historical time are obtained.

The first vehicle can send the first data to the decision-making device in real time so as to optimize the behavior decision of the first vehicle. The first data comprises a first behavior decision and second data, wherein the second data comprises control signals, perception data, planning data, driving state data and the like, so that full-scale data modeling of the vehicle is realized.

The first vehicle is provided with a sensing module, a decision-making module, a chassis control module, a domain controller module and the like. The chassis control module includes, but is not limited to, a conventional powertrain module, a new energy powertrain module, a driveline module, a body system module, and an electronic system module.

The perception data may include road data, radar data, environmental data, and map data. The environment data may include, among other things: air temperature, air speed, humidity, illumination intensity. The radar data may include mechanical lidar data, solid state lidar data, millimeter wave radar data, and infrared radar data. The road data may include: road topology, traffic light signals and traffic signs. The environmental data may include: the traffic data may include traffic light signals and traffic signs, and may include external factors such as air temperature, wind speed, humidity, illumination intensity, road topology data, traffic data, obstacle data, map data, location data, and the like.

The control signal may include: the system comprises a brake signal, a steering signal, a driving signal, a gear signal, a vehicle speed signal, a driving direction signal and a torque output signal of a chassis control, a target brake signal, a target steering signal, a target driving signal, a target gear signal, a target vehicle speed signal and the like of a domain controller.

The planning data may include: global travel paths and local travel paths, etc.

The driving state data may include: travel speed, acceleration, chassis rotation angle, etc.

Also, the sensing module of the first vehicle may include a sensor for acquiring the aforementioned various sensing data, for example, the sensor may include an image sensor, a radar, and the like.

The first vehicle may obtain a first behavior decision of the first vehicle based on the aforementioned second data by a decision module arranged on the first vehicle. The first behavior decision may also be obtained by a decision model in a corresponding digital twin system of the first vehicle arranged on the decision device based on the aforementioned second data. Specifically, when the decision device receives first data sent by the first vehicle, the first data is used for configuring a simulation environment of the digital twin system, and a first behavior decision of the first vehicle can be obtained by using the digital twin system in the simulation environment.

Optionally, a data flow data transfer bus may be used to connect and transfer data between the first vehicle and the decision-making device. The form of the information flow data transmission bus includes, but is not limited to, limited transmission technology, wireless transmission technology, quantum transmission technology. The wired transmission technology includes but is not limited to CAN bus transmission technology, flexray bus transmission technology and MOST bus transmission technology, and the wireless transmission technology includes but is not limited to bluetooth connection transmission, 4G network connection transmission, 5G network connection transmission, WLAN connection transmission. The problem of expected functional safety is well solved by adopting abundant and various data transmission forms, namely, when one data output fails, another data transmission form can be effectively adopted for replacing the data output.

The second data may further include: the vehicle time-varying data mainly comprises the control signal, the sensing data, all real-time data and additional data outside a driving path, so that the completeness and the effectiveness of a vehicle running system can be better improved;

the second data may further include: and (4) system operation mechanism characteristics. The system operation mechanism features mainly refer to theoretical methods or mechanisms in the field or category, and may include artificial intelligence algorithms or empirical formulas used for calculation and derivation during operation of decision-making equipment, and effective methods for analyzing and updating real-time rolling iteration data, the artificial intelligence algorithms include, but are not limited to, random forests for supervised learning, naive Bayes classifiers, least squares methods, logistic regression, support vector machines, ensemble learning, etc., clustering algorithms for unsupervised learning, K-means algorithms, SVD matrix analysis algorithms, etc., and one or more arbitrary combinations of reinforcement learning, deep learning, and machine learning algorithms, and the calculations of such theoretical methods or empirical formulas are obtained by current intelligent calculation methods or a large number of experimental experiences, and have strong applicability to the scheme of the present application.

Step S102, configuring a simulation environment of the digital twin system based on first data of the first vehicle at the current time and the historical time.

The decision device may establish a digital twin system corresponding to the first vehicle using digital twin technology in advance. The digital twin system comprises a perception model corresponding to the perception module, a decision model corresponding to the decision module, a chassis control model corresponding to the chassis control module and/or a domain control model corresponding to the domain controller module. In particular, the decision device may digitally model portions of the vehicle in a simulation environment using CAD drawing, CAE approximation numerical analysis or finite element methods based on physical data of the vehicle including, but not limited to, geometry, material properties, circuit structure and connection relationships, to obtain the same digital twin system as an actual vehicle in the simulation environment.

In one example, the corresponding digital twin system of the first vehicle may also be sent to the decision device by the model device after the construction based on the digital twin technology is completed.

And when the decision-making equipment receives first data sent by the first vehicle, configuring the simulation environment of the digital twin system according to the first data.

Step S103, in a simulation environment, simulating the first vehicle by using the digital twin system to execute a first behavior decision of the first vehicle at the current time, and determining simulation data of the first vehicle.

After obtaining a first behavior decision of a first vehicle, the decision device inputs the first behavior decision into a decision model of the digital twin system, the decision model sends an instruction in the first behavior decision to a domain controller module and a chassis controller module in the digital twin system, and simulation calculation is carried out in a simulation environment to obtain simulation data. The simulation data indicates safety of the first vehicle.

The simulation data comprises a data set characterized in the running process of the digital twin platform and all data required by the parallel processing cloud platform when the behavior decision optimization is carried out.

For example, the simulation data may include: a distance between the first vehicle and a second vehicle, and/or a distance between the first vehicle and an obstacle; wherein the second vehicle comprises a vehicle surrounding the first vehicle.

And step S104, judging whether the first behavior decision of the current time meets the safety condition of the first vehicle based on the simulation data.

And the decision device analyzes and compares the simulation data with preset safety conditions of the first vehicle after obtaining the simulation data output by the digital twin system. When the simulation data does not match the safety condition, it is determined that the first behavior decision at the current time does not satisfy the safety condition of the first vehicle, and the decision device performs the following step S106. When the simulation data matches the safety condition, it is determined that the first behavior decision at the current time satisfies the safety condition of the first vehicle, and the decision device directly performs the subsequent step S107.

Wherein the safety conditions of the first vehicle may be set according to the simulation data and the actual safety requirements. For example, based on the simulation data shown previously, the safety conditions may be set as: the distance between the first vehicle and the second vehicle is larger than 1m, and the distance between the first vehicle and the obstacle is larger than 30 cm.

And S105, determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data at the current time.

When the decision device determines that the first behavior decision at the current time does not satisfy the safety condition, the decision device may re-determine, based on the first data obtained in step S101, a behavior decision of the first vehicle, that is, a second behavior decision, by using a pre-trained decision optimization model, and execute step S106. Specifically, the decision device may extract feature data from the first data, input the feature data into the decision optimization model, and obtain a second behavior decision output by the decision optimization model.

The training process of the decision optimization model is described in detail in the following with reference to fig. 4, and will not be described herein again.

And step S106, sending the first behavior decision or the second behavior decision to the first vehicle.

In one example, when the first behavior decision for the current time satisfies the safety condition and the first behavior decision is obtained by the decision module of the first vehicle, the decision device may send an instruction to the first vehicle instructing the first vehicle to perform autonomous driving control in accordance with the first behavior decision.

In one example, when the first behavior decision at the current time does not satisfy the safety condition, as shown in fig. 2, the decision device may send a second behavior decision to the first vehicle, and the first vehicle performs automatic driving control according to the second behavior decision.

Fig. 2 is a flowchart of a training method of a decision optimization model according to an embodiment of the present disclosure. As shown in fig. 4, the method includes steps S201 to S203 as follows.

Step S201, obtaining a plurality of training samples of a first vehicle, wherein each training sample comprises historical first data and a historical behavior decision corresponding to the historical first data.

The database records the data and behavior decision of the first vehicle in the whole life cycle. The model device may obtain a plurality of training samples of the first vehicle in a database. The first vehicle can send the sensing data and the corresponding behavior decision to the model device for storage after executing the behavior decision each time.

The historical first data and its corresponding historical behavior decision may be an actual driving strategy of the first vehicle at the historical time.

And S202, extracting characteristic data in each historical first data.

The model device may extract feature data in each of the historical first data using a preset feature extraction method.

And S203, updating parameters of the decision optimization model by using the characteristic data corresponding to each historical first data and the historical behavior decision.

The decision optimization model can be constructed by adopting one algorithm of a supervised learning algorithm and/or an unsupervised learning algorithm, or can be obtained by combining the algorithms. Wherein, the supervised learning can comprise random forest, naive Bayes classifier, least square method, logistic regression, support vector machine, convolution neural network, BP neural network, etc., and the unsupervised learning can comprise clustering algorithm, K-mean algorithm, SVD matrix analysis algorithm, etc.

Taking supervised learning as an example, after the model device obtains an initial decision optimization model, the feature data corresponding to each piece of historical first data may be input into the decision optimization model, so as to obtain a behavior prediction policy output by the decision optimization model. And the model equipment can calculate the loss value between the behavior prediction strategy corresponding to each historical first data and the historical behavior strategy based on a preset loss function, and adjust the parameters of the decision optimization model according to the obtained loss value until the training condition of the decision optimization model is met. The training condition may include that a loss value between a behavior prediction strategy and a historical behavior strategy output by the decision optimization model is smaller than a threshold.

Based on the embodiment of the method for optimizing the behavior decision of the autonomous driving vehicle shown in fig. 1, the present application also provides a system for optimizing the behavior decision of the autonomous driving vehicle.

Fig. 3 is a schematic structural diagram of an automated driving vehicle behavior decision optimization system provided in the present application. As shown in fig. 3, the system includes: a digital twin platform 301 and a parallel processing cloud platform 302 connected to a first vehicle.

The digital twin platform 301 is used for building a corresponding digital twin system of the first vehicle by using a digital twin technology, and the digital twin system is consistent with an actual physical system of the first vehicle, and specific description can be referred to the description in the foregoing method embodiment.

The digital twin platform 301 is further configured to configure a simulation environment of the digital twin system based on the first data of the first vehicle at the current time and the historical time, so that the simulation environment of the digital twin system is consistent with the working environment of the first vehicle. The first data comprises the second data and the first behavior decision.

The digital twin platform 301 is further configured to obtain a first behavior decision of the first vehicle at the current time, and simulate the first vehicle by using the digital twin system to execute the first behavior decision of the first vehicle at the current time, so as to determine simulation data of the first vehicle; the simulation data is indicative of safety of the first vehicle; the first data is uploaded by the first vehicle.

The parallel processing cloud platform 302 is configured to determine whether the first behavior decision at the current time satisfies a safety condition based on the simulation data, and re-determine a second behavior decision of the first vehicle at the current time based on the first data at the current time by using a decision optimization model based on the first data at the current time when the first behavior decision at the current time does not satisfy the safety condition.

The parallel processing cloud platform 302 is further configured to send the second behavior decision to the first vehicle and to send the second behavior decision to the digital twin platform; wherein the digital twin platform simulates a first vehicle with the digital twin system to perform a second behavioral decision.

In one example, the parallel processing cloud platform 302 is further configured to store historical first data of the first vehicle, enabling monitoring of full-life cycle data of the first vehicle. The historical data can effectively provide the functions of model optimization, data iteration updating, data analysis, method improvement, function improvement, experience summary and the like, so that the behavior decision method of the automatic driving automobile is improved. For example, the parallel processing cloud platform 302 may iteratively update parameters of the decision optimization model based on the historical first data.

In one example, parallel processing cloud platform 302 is further to: predicting first data of the first vehicle at a next time based on the first data of the current time; determining, using the decision optimization model, a second behavior decision for the first vehicle at the next time based on the first data for the next time.

In one example, the parallel processing cloud platform 302 may send the second behavior decision to a terminal display device of the first vehicle for display.

In an example, the behavior decision sent by the parallel processing cloud platform may be in a finished vehicle state report form, and the content may include, but is not limited to, a current state of an actual finished vehicle system, sensing system detection data, path planning data, vehicle fault information, a finished vehicle system safety online evaluation result, timeliness early warning of each system, abnormal condition monitoring, and a maintenance suggestion, except for a specific execution instruction.

In the system, through rolling iterative optimization among the parallel processing cloud platform, the digital twin platform and the whole vehicle running system, the digital twin platform feeds back an integration and analysis evaluation result to the virtual simulation system, and after the virtual simulation system is updated in an iterative manner, the parallel processing cloud platform transmits data to the actual whole vehicle, so that the whole vehicle-cloud-virtual loop interaction is realized, and the behavior decision optimization method for the automatic driving vehicle is more accurate, efficient and safe.

In addition, the system has the functions of dynamically optimizing behavior decision-making capability, improving the data processing efficiency of a sensing system, analyzing the path planning of the whole vehicle, monitoring the running state of a chassis system of the whole vehicle and system fault information, and optimizing a control method of the whole vehicle, achieves behavior decision-making optimization, whole vehicle safety guarantee and online monitoring management in the whole life cycle, can greatly improve the safety and reliability of the automatically driven vehicle, and saves the development cost.

The parallel processing cloud platform can store and calculate a large amount of data, can also realize vehicle real-time state monitoring and state prediction in a certain time and space, and realizes synchronous optimization, state estimation, fault assessment and risk prediction of a physical object and a virtual simulation system, and performance monitoring and safety management of a full life cycle. In addition, under the high-efficiency computing power and processing of the parallel processing cloud platform, the obtained data is subjected to secondary development, key indexes and parameters of a real object and a simulation model are continuously computed, and multi-objective optimization including a behavior decision method of an actual finished automobile running system can be further promoted.

Based on the foregoing vehicle behavior decision method embodiment shown in fig. 2, the present application further provides a vehicle behavior decision device.

Fig. 4 is a schematic structural diagram of an automatic vehicle driving behavior decision optimizing device 400 according to an embodiment of the present application. The optimization apparatus 400 may apply the aforementioned decision device. As shown in fig. 4, the optimizing apparatus 400 includes: an acquisition module 401, a configuration module 402, a simulation module 403 and an optimization module 404.

The obtaining module 401 is configured to obtain first data of a first vehicle at a current time and a historical time, where the first data includes second data and a first behavior decision.

Wherein the configuration module 402 is configured to configure a simulation environment of the digital twin system based on first data of the first vehicle at a current time and a historical time; the digital twinning system is built by utilizing a digital twinning technology.

Wherein the simulation module 403 is configured to simulate the first vehicle with the digital twin system to execute the first behavior decision in the simulation environment, and determine simulation data of the first vehicle; the simulation data indicates safety of the first vehicle.

Wherein the optimization module 404 is configured to determine a second behavior decision of the first vehicle at the current time using a decision optimization model based on the first data at the current time if it is determined based on the simulation data that the first behavior decision at the current time does not satisfy a safety condition.

For specific implementation processes of each functional module in the optimization apparatus 400, reference may be made to the description in the embodiment of the method shown in fig. 1, and details are not described here again.

Based on the foregoing embodiment of the training method for the decision optimization model shown in fig. 2, the present application further provides a model training apparatus.

Fig. 5 is a schematic structural diagram of a vehicle behavior model training apparatus 500 according to an embodiment of the present application. The model training apparatus 500 can be applied to the model device described above. As shown in fig. 5, the model training apparatus 500 includes: an acquisition module 501, a feature module 502, and a training module 503.

The obtaining module 501 is configured to obtain a plurality of training samples of a first vehicle, where each training sample includes historical first data and a historical behavior decision corresponding to the historical first data.

The feature module 502 is configured to extract feature data in each historical first data.

The training module 503 is configured to update parameters of the decision optimization model by using the feature data and the historical behavior decision corresponding to each historical first data.

The specific implementation process of each functional module of the model training apparatus 500 can be described in the embodiment of the method shown in fig. 2, and is not described herein again.

Fig. 6 is a schematic hardware structure diagram of a computing device 600 according to an embodiment of the present application. The computing device 600 may be the decision device or the model device described above. The computing device 600 may also be a digital twin platform or a parallel processing cloud platform as described above.

Referring to fig. 6, the computing device 600 includes a processor 601, a memory 602, a communication interface 603, and a bus 804, and the processor 601, the memory 602, and the communication interface 603 are connected to each other by the bus 804. The processor 601, memory 602, and communication interface 603 may also be connected using connections other than bus 804.

The memory 602 may be various types of storage media, such as Random Access Memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), programmable ROM (PROM), erasable PROM (EPROM), electrically Erasable PROM (EEPROM), flash memory, optical storage, hard disk, and the like.

Where the processor 601 may be a general-purpose processor, the general-purpose processor may be a processor that performs certain steps and/or operations by reading and executing content stored in a memory, such as the memory 602. For example, a general purpose processor may be a Central Processing Unit (CPU). The processor 601 may include at least one circuit to perform all or part of the steps of the method shown in fig. 1 or fig. 2.

The communication interface 603 includes input/output (I/O) interfaces, physical interfaces, logical interfaces, and the like for realizing interconnection of devices inside the computing apparatus 600, and interfaces for realizing interconnection of the computing apparatus 600 with model apparatuses (e.g., other computing apparatuses or user apparatuses). The physical interface may be an ethernet interface, a fiber optic interface, an ATM interface, or the like.

The bus 804 may be any type of communication bus, such as a system bus, for interconnecting the processor 601, the memory 602, and the communication interface 603.

The above devices may be respectively disposed on separate chips, or at least a part or all of the devices may be disposed on the same chip. Whether each device is separately disposed on a different chip or integrated on one or more chips will often depend on the needs of the product design. The embodiment of the present application does not limit the specific implementation form of the above device.

The computing device 600 shown in fig. 6 is merely exemplary, and in implementations, the computing device 600 may include other components, which are not listed here.

In the above embodiments, all or part of the implementation may be realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to be performed in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a Solid State Disk (SSD)), among others.

It is to be understood that the various numerical references referred to in the embodiments of the present application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of the present application. It should be understood that, in the embodiment of the present application, the sequence numbers of the foregoing processes do not imply an order of execution, and the order of execution of the processes should be determined by their functions and inherent logic, and should not constitute any limitation to the implementation process of the embodiment of the present application.

The above-mentioned embodiments, objects, technical solutions and advantages of the present application are further described in detail, it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention, and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made on the basis of the technical solutions of the present application should be included in the scope of the present application.

Claims

1. An autonomous vehicle behavior decision optimization system, the system comprising: a digital twin platform and a parallel processing cloud platform connected with a first vehicle; the digital twin platform is used for building a digital twin system corresponding to the first vehicle by using a digital twin technology, and configuring a simulation environment of the digital twin system based on first data of the first vehicle at the current time and first data of historical time; the first data comprises a first behavioral decision and second data; the first data is determined by the first vehicle; the first vehicle comprises a perception module, a decision module, a chassis control module and/or a domain controller module, and the digital twin system comprises a perception model corresponding to the perception module, a decision model corresponding to the decision module, a chassis control model corresponding to the chassis control module and/or a domain control model corresponding to the domain controller module;

the second data comprises control signals generated by a chassis control module and/or a domain controller module, sensing data acquired by the sensing module, intermediate calculation data, driving planning data of the first vehicle, driving state data, all real-time data, additional data and system operation mechanism characteristics;

the method comprises the steps that a decision-making device carries out digital modeling on each part of a first vehicle by adopting a CAD (computer-aided design) drawing method, a CAE (computer-aided engineering) approximate numerical analysis method or a finite element method in a simulation environment on the basis of physical data of the vehicle, including geometric dimensions, material attributes, a circuit structure and a connection relation, so as to obtain a digital twin system;

the digital twin platform is further used for simulating the first vehicle by utilizing the digital twin system to execute a first behavior decision of the first vehicle at the current time in the simulation environment, and determining simulation data of the first vehicle; the first behavior decision is obtained by a decision module arranged on the first vehicle based on the second data, the simulated data being indicative of safety of the first vehicle;

the parallel processing cloud platform is used for judging whether a first behavior decision of the current time meets a safety condition or not based on the simulation data, and determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data of the current time under the condition that the first behavior decision of the first vehicle does not meet the safety condition, and sending the second behavior decision to the first vehicle in a finished vehicle state report form; the finished vehicle state report comprises the current state of a finished vehicle system, sensing system detection data, path planning data, vehicle fault information, finished vehicle system safety online evaluation results, timeliness early warning of each system, abnormal condition monitoring and maintenance suggestions;

the parallel processing cloud platform is further used for sending the second behavior decision to the first vehicle and synchronously sending the second behavior decision to the digital twin platform; wherein the digital twin platform simulates the first vehicle with the digital twin system performing the second behavioral decision; and executing synchronous iterative optimization among the parallel processing cloud platform, the digital twin platform and the first vehicle running system.

2. The system of claim 1, wherein the simulation data comprises: the method comprises the steps of collecting data characterized in the running process of the digital twin platform and parallelly processing all data required by the cloud platform when behavior decision optimization is carried out; a distance between the first vehicle and a second vehicle, and/or a distance between the first vehicle and an obstacle, wherein the second vehicle comprises a vehicle surrounding the first vehicle.

3. The system of claim 1, wherein the control signals, sensory data, planning data, driving state data, and intermediate calculation data comprise:

wherein the perception data comprises road data, radar data, environmental data, and map data;

the control signals comprise a braking signal, a steering signal, a driving signal, a gear signal, a vehicle speed signal, a driving direction signal and a torque output signal controlled by the chassis, and a target braking signal, a target steering signal, a target driving signal, a target gear signal and a target vehicle speed signal of the domain controller;

the intermediate calculation data comprises process data generated by an artificial intelligence algorithm or an empirical formula of calculation derivation during the operation of the decision-making equipment;

the planning data includes: a global travel path and a local travel path;

the driving state data comprises: travel speed, acceleration, and chassis rotation angle.

4. The system of claim 1, wherein the determining, using a decision optimization model, a second behavior decision for the first vehicle at the current time based on the first data at the current time comprises:

extracting feature data from the first data of the current time;

5. The system according to claim 4, wherein the decision optimization model is constructed using a supervised learning algorithm and/or an unsupervised learning algorithm; the supervised learning comprises a random forest, a naive Bayes classifier, a least square method, logistic regression, a support vector machine, a convolutional neural network and a BP neural network, and the unsupervised learning comprises a clustering algorithm, a K-mean algorithm and an SVD matrix analysis algorithm.

6. The system of claim 1, wherein the parallel processing cloud platform is further configured to store the first data of the first vehicle at the historical time, and iteratively update parameters of the decision optimization model based on the first data of the first vehicle at the historical time.

7. The system of claim 1, wherein the parallel processing cloud platform is further configured to:

predicting the first data of the first vehicle at a next time based on the first data of the current time;

determining, with the decision optimization model, a second behavior decision for the first vehicle at the next time based on the first data for the next time.

8. An automated driving vehicle behavior decision optimization method for use with the system of any one of claims 1-7, the method comprising:

the second data comprises control signals generated by a chassis control module and/or a domain controller module, sensing data acquired by the sensing module, intermediate calculation data, driving planning data of the first vehicle, driving state data, all real-time data, additional data and system operation mechanism characteristics; configuring a simulation environment of a digital twin system based on first data of the first vehicle at the current time and the historical time, wherein the digital twin system is built by utilizing a digital twin technology;

the method comprises the following steps that on the basis of physical data including geometric dimensions, material attributes, a circuit structure and a connection relation, a CAD drawing method, a CAE approximate numerical analysis method or a finite element method is adopted in a simulation environment, and digital modeling is conducted on each part of a vehicle through decision-making equipment to obtain a digital twin system;

simulating the first vehicle to execute a first behavior decision at the current time by using the digital twin system in the simulation environment, and determining simulation data of the first vehicle; the simulation data is indicative of the first vehicle safety;

under the condition that the first behavior decision of the current time does not meet the safety condition based on the simulation data, determining a second behavior decision of the first vehicle at the current time by using a decision optimization model based on the first data of the current time, and sending the second behavior decision to the first vehicle in the form of a complete vehicle state report; the finished vehicle state report comprises the current state of the finished vehicle system, sensing system detection data, path planning data, vehicle fault information, finished vehicle system safety online evaluation results, timeliness early warning of each system, abnormal condition monitoring and maintenance suggestions.