CN108268740B - Virtual acquisition method and device for big data of automatic driving training - Google Patents
Virtual acquisition method and device for big data of automatic driving training Download PDFInfo
- Publication number
- CN108268740B CN108268740B CN201810122666.2A CN201810122666A CN108268740B CN 108268740 B CN108268740 B CN 108268740B CN 201810122666 A CN201810122666 A CN 201810122666A CN 108268740 B CN108268740 B CN 108268740B
- Authority
- CN
- China
- Prior art keywords
- driving environment
- target
- targets
- gene
- driving
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 title claims abstract description 32
- 108090000623 proteins and genes Proteins 0.000 claims abstract description 125
- 238000011156 evaluation Methods 0.000 claims abstract description 68
- 206010064571 Gene mutation Diseases 0.000 claims abstract description 16
- 230000007613 environmental effect Effects 0.000 claims description 43
- 230000014509 gene expression Effects 0.000 claims description 31
- 230000002068 genetic effect Effects 0.000 claims description 24
- 230000035772 mutation Effects 0.000 claims description 8
- 238000010586 diagram Methods 0.000 description 16
- 238000005516 engineering process Methods 0.000 description 7
- 238000004088 simulation Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000013480 data collection Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007614 genetic variation Effects 0.000 description 2
- 101150084750 1 gene Proteins 0.000 description 1
- FGRBYDKOBBBPOI-UHFFFAOYSA-N 10,10-dioxo-2-[4-(N-phenylanilino)phenyl]thioxanthen-9-one Chemical compound O=C1c2ccccc2S(=O)(=O)c2ccc(cc12)-c1ccc(cc1)N(c1ccccc1)c1ccccc1 FGRBYDKOBBBPOI-UHFFFAOYSA-N 0.000 description 1
- 101150028074 2 gene Proteins 0.000 description 1
- TVEXGJYMHHTVKP-UHFFFAOYSA-N 6-oxabicyclo[3.2.1]oct-3-en-7-one Chemical compound C1C2C(=O)OC1C=CC2 TVEXGJYMHHTVKP-UHFFFAOYSA-N 0.000 description 1
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 101150073444 rd gene Proteins 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Hardware Design (AREA)
- Physiology (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Genetics & Genomics (AREA)
- Automation & Control Theory (AREA)
- Aviation & Aerospace Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Traffic Control Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The embodiment of the invention discloses a virtual acquisition method and a virtual acquisition device for big data of automatic driving training. The method comprises the following steps: determining gene codes of candidate driving environment targets for automatic virtual driving; performing automatic driving evaluation on each candidate driving environment target, and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result; aiming at each group of parent driving environment targets, performing gene cross operation and/or gene mutation operation on gene codes of the group of parent driving environment targets to generate child driving environment targets of the group of parent driving environment targets; and if the terminal condition is not met, taking the generated child driving environment targets as new candidate driving environment targets, returning to execute automatic driving evaluation operation, and selecting new parent driving environment targets from the new candidate driving environment targets according to an evaluation result. The embodiment of the invention can reduce the difficulty of large data acquisition of automatic driving, reduce the acquisition cost and increase the operability.
Description
Technical Field
The embodiment of the invention relates to the technical field of automatic driving, in particular to a virtual acquisition method and device for big data of automatic driving training.
Background
With the development of science and technology and the improvement of people's pursuit of living quality, people hope to get rid of fatigue driving activities, and automatic driving technology comes along. Automatic driving, also known as unmanned driving, is a technology for realizing unmanned driving through an intelligent system, and the information of the road environment needs to be comprehensively known when safe and reliable road driving capacity is needed in the automatic driving process.
In the existing automatic driving technology, a multi-sensor technology is generally adopted to acquire environmental information, for example, a camera is used to acquire image information of a road environment, a laser radar is used to acquire three-dimensional information of a road, and the like. In order to acquire a large amount of data under various complex environments, a method for acquiring environmental data in an actual road is generally adopted, and a vehicle is usually carried with a plurality of sensors to run under the actual road environment to acquire the data. The diversity of data is ensured by collecting enough road information.
The existing road environment big data acquisition method for automatic driving is generally influenced by environmental conditions, and relatively comprehensive data is difficult to acquire. For example, environmental data for snowy, rainy and hail days, environmental data for different lighting conditions, etc. In these cases, it is difficult to obtain sufficiently comprehensive environmental data due to the influence of uncontrollable natural environmental conditions. Even for data collection in a common natural environment, it is often necessary to spend a lot of time and economic cost to collect field data in order to ensure the diversity of data. In addition, there is a certain risk in actual road data collection. For example, during the automated driving field data collection process, some traffic accidents may occur. At the same time, the data set should contain some bursty emergency data. Such as a sudden pedestrian accident, a sudden vehicle rear-end accident, and the like. The data of these accidents cannot be acquired in the data acquisition of the actual environment.
Disclosure of Invention
The embodiment of the invention provides a virtual acquisition method and a virtual acquisition device for big data of automatic driving training, which can more simply and conveniently realize more comprehensive acquisition of the big data of automatic driving, reduce the acquisition difficulty, reduce the acquisition cost and increase the operability.
In a first aspect, an embodiment of the present invention provides a virtual acquisition method for big data of automatic driving training, including:
determining gene codes of candidate driving environment targets for automatic virtual driving;
performing automatic driving evaluation on each candidate driving environment target, and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result;
aiming at each group of parent driving environment targets, performing gene cross operation and/or gene mutation operation on gene codes of the group of parent driving environment targets to generate child driving environment targets of the group of parent driving environment targets;
and if the terminal condition is not met, taking the generated child driving environment targets as new candidate driving environment targets, returning to execute automatic driving evaluation operation, and selecting new parent driving environment targets from the new candidate driving environment targets according to an evaluation result.
In a second aspect, an embodiment of the present invention further provides a virtual collecting device for big data of automatic driving training, where the device includes:
the gene coding module is used for determining the gene codes of all candidate driving environment targets for automatic virtual driving;
the driving evaluation module is used for automatically evaluating the driving of each candidate driving environment target and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result;
the evolution module is used for carrying out gene cross operation and/or gene mutation operation on the gene codes of the set of parent driving environment targets aiming at each set of parent driving environment targets to generate child driving environment targets of the set of parent driving environment targets;
and the return evaluation module is used for taking each generated child driving environment target as a new candidate driving environment target if the termination condition is not met, returning to execute the automatic driving evaluation operation, and selecting a new parent driving environment target from the new candidate driving environment targets according to the evaluation result.
The embodiment of the invention generates the virtual automatic driving environment by simulating the target in the automatic driving environment and carrying out gene coding, automatic driving evaluation and gene crossing and/or gene variation on the target. According to the technical scheme provided by the embodiment of the invention, the generated automatic driving environment can be called through the virtual engine to generate different automatic driving virtual big data, so that more comprehensive automatic driving big data can be more simply and conveniently acquired, the acquisition difficulty is reduced, the acquisition cost is reduced and the operability is improved.
Drawings
FIG. 1 is a flowchart of a virtual collecting method for big data of automatic driving training according to a first embodiment of the present invention;
FIG. 2 is a schematic diagram of the gene coding of environmental factors of automatic driving according to a first embodiment of the present invention;
FIG. 3 is a distribution diagram of a target location in accordance with a first embodiment of the present invention;
FIG. 4 is a schematic diagram of the gene coding of the target factors of the automatic driving vehicle according to the first embodiment of the present invention;
FIG. 5 is a simplified exemplary diagram of an autonomous driving environment in accordance with a first embodiment of the invention;
FIG. 6 is a flowchart of a virtual collecting method for big data of automatic driving training according to a second embodiment of the present invention;
FIG. 7 is a diagram showing an example of evolution of gene codes of environmental factors for automatic driving in accordance with a second embodiment of the present invention;
FIG. 8 is an example diagram of the evolution phenotype of the gene coding of the environmental factors of the automatic driving vehicle in the second embodiment of the present invention;
FIG. 9 is a diagram showing an example of the evolution of gene codes of target factors for automatic driving in a second embodiment of the present invention;
FIG. 10 is an exemplary view of an evolution phenotype of gene codes of the automatic driving objective factors in the second embodiment of the present invention;
FIG. 11 is a flowchart of a virtual collecting method for big data of automatic driving training according to a third embodiment of the present invention;
fig. 12 is a schematic structural diagram of a virtual acquisition device for big data of automatic driving training in the fourth embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.
Example one
Fig. 1 is a flowchart of a virtual collecting method for big auto-driving training data according to a first embodiment of the present invention, where this embodiment is applicable to a virtual collecting situation of big auto-driving training data, and the method may be executed by a virtual collecting device for big auto-driving training data, and the method may specifically include:
and step 110, determining the gene codes of the candidate driving environment targets of the automatic virtual driving.
In this embodiment, a highly simulated virtual engine, such as Unreal, may be used to simulate an autonomous driving environment. Unreal is a shorthand of Unreal Engine, is a game Engine, adopts the technologies of instant light track tracking, high dynamic range illumination technology, virtual displacement and the like, can calculate two hundred million polygon calculations in real time per second, and has very high simulation degree of real environment. Other virtual simulation platforms with high simulation degree can also be used for simulating the driving environment of automatic driving.
The candidate driving environment target can comprise at least one environmental factor of environmental factors possibly encountered by real driving environments such as weather, road conditions, trees, buildings, traffic lights and the like, and at least one target factor of environmental vehicles and pedestrians.
Specifically, the gene codes of the candidate driving environment targets of the automatic virtual driving are determined, and for example, two environment factors, namely weather and road conditions, can be selected for gene coding, as shown in table 1; both the environmental vehicle and the pedestrian can be genetically encoded as shown in table 2.
TABLE 1 Gene coding Table for weather and road conditions
TABLE 2 Gene coding Table for environmental vehicles and pedestrians
As shown in table 1, gene coding is performed on different weather conditions such as semi-cloudy days, light rain and the like in the weather, and gene coding of other different weather conditions such as foggy days and the like can be added according to needs; similarly, gene codes of different road conditions may be added, that is, gene codes of different environmental factors may be set according to needs, as shown in fig. 2, where fig. 2 is a schematic diagram of gene codes of environmental factors of automatic driving according to an embodiment of the present invention.
As shown in table 2, in the gene coding for the environmental vehicles and pedestrians, the types, numbers, and positions of the targets are set. Where the types are between the same object, there may be different types, for example, vehicles may include dollies and vans, etc., pedestrians may include children and adults, etc.; the number is the number of different targets; in the gene coding of the position, 10 genes may be set, the corresponding positions of the target in the virtual environment image may be divided into 1024 positions, and may be position combinations of 32 × 32 units in the x and y directions, respectively, as shown in fig. 3, fig. 3 is a distribution diagram of the target position in the first embodiment of the present invention, and different target positions may be generated by changing according to the position distribution in fig. 3. In addition to the two target factors of the vehicle and the pedestrian, other target factors, such as pets, etc., may be added as required, as shown in fig. 4, where fig. 4 is a schematic gene coding diagram of the automatic driving target factor in the first embodiment of the present invention.
In this embodiment, by setting the position of the target, the establishment of a special driving environment, such as a sudden pedestrian accident or a sudden rear-end collision accident, can be realized, and specifically, if the position of the pedestrian and the vehicle or the position of the vehicle and the vehicle is changed from a longer distance to a shorter distance in a short time. In a similar way, other special conditions can be set according to needs, and compared with the prior art, the cost is greatly reduced.
And 120, performing automatic driving evaluation on each candidate driving environment target, and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result.
The specific process of the automatic driving assessment may be as follows: establishing a corresponding automatic virtual driving environment for a plurality of driving environment targets (comprising different environment factors and target factors) through a simulation platform, and performing automatic driving evaluation according to the performance of an automatic driving evaluation vehicle in the automatic virtual driving environment. For example, if the automated driving evaluation vehicle is traveling well in a driving environment without a collision or other accident, the automated driving evaluation value corresponding to the driving environment may be high, for example, 100 points; if the automated driving evaluation vehicle is continuously stopped or collided in another driving environment, the automated driving evaluation value corresponding to the driving environment may be low, for example, 0 point. The automatic driving evaluation value can be used for setting evaluation conditions and evaluation intervals as required, scoring different performances, and setting an analysis statistical formula to obtain a final automatic driving evaluation value.
Specifically, a plurality of candidate driving environment targets may be randomly selected for the automatic driving evaluation, and a preset number of candidate driving environment targets may be selected as the parent driving environment targets according to the evaluation result, for example, 5 candidate driving environment targets arranged in front may be selected as the parent driving environment targets according to the evaluation result, that is, the automatic driving evaluation value. The preset number can be set as required.
And 130, aiming at each group of parent driving environment targets, performing gene cross operation and/or gene mutation operation on the gene codes of the group of parent driving environment targets to generate child driving environment targets of the group of parent driving environment targets.
Wherein, the gene cross operation can be the exchange of gene codes of two different positions, and the gene mutation operation can be the modification of the gene code of a certain position.
Specifically, the parent driving environment targets determined in step 120 may be grouped two by two, and a genetic crossover operation and/or a genetic mutation operation may be performed on the genetic code of each group to generate child driving environment targets corresponding to the parent driving environment targets of the group.
And 140, if the termination condition is not met, taking the generated each child driving environment target as a new candidate driving environment target, returning to execute the automatic driving evaluation operation, and selecting a new parent driving environment target from the new candidate driving environment targets according to the evaluation result.
The termination condition may be that the number of times of generating the child driving environment target is greater than a number threshold, or that a difference between the automatic driving evaluation value and the evaluation expectation value of the generated child driving environment target is less than a difference threshold, where the number threshold and the difference threshold may be set according to different requirements of training big data.
Specifically, if the termination condition is not satisfied, the generated child driving environment targets are used as new candidate driving environment targets, and the process returns to the step 120 of executing, that is, the automatic driving evaluation operation is executed, a new parent driving environment target is selected from the new candidate driving environment targets according to the evaluation result, the gene coding of the new parent driving environment target is subjected to gene crossing operation and/or gene mutation operation, and a corresponding new child driving environment target is generated until the termination condition is satisfied.
In this embodiment, if the termination condition is satisfied, a certain constraint condition may be set for the child driving environment target so that the child driving environment target is consistent with the real automatic driving environment. The constraint condition may be that the automated driving evaluation vehicle is different in position from the target factor and/or that the environmental vehicle and the pedestrian are different in position from the target factor, and the like. The constraint condition can be set according to different requirements of training big data, for example, the constraint condition can also be that two pedestrians cannot have the same position, but can have adjacent positions, etc.
In this embodiment, if the termination condition is satisfied and the corresponding constraint condition is set, the environment factor and the target factor in the final child driving environment target may be called in the simulation platform (e.g., unregeal), so as to generate different automatic driving training data. Fig. 5 is a schematic diagram of an example of an autonomous driving environment according to a first embodiment of the present invention, as shown in fig. 5, the environment is a rural road environment under sunny days, in which pedestrians, street lamps and trees are arranged.
In the embodiment, the automatic virtual driving environment is generated by simulating the target in the automatic driving environment, performing gene coding, automatic driving evaluation and gene crossing and/or gene variation on the target, setting corresponding constraint conditions, and generating different automatic driving training data through the simulation platform. The technical scheme provided by the embodiment of the invention can more simply and conveniently realize more comprehensive automatic driving big data acquisition, reduce the acquisition difficulty, reduce the acquisition cost and increase the operability.
Optionally, determining the gene code of each candidate driving environment target for the automatic virtual driving may include: determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target; and combining the gene expression of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene code of the candidate driving environment target.
Example two
Fig. 6 is a flowchart of a virtual collecting method of big data of automatic driving training in the second embodiment of the present invention. On the basis of the above embodiments, the virtual acquisition method of the automatic driving training big data is further optimized in this embodiment. Correspondingly, the method of this embodiment may specifically include:
and step 210, determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target.
Specifically, a plurality of candidate driving environment targets can be randomly selected, and for each candidate driving environment target, the gene expression of each environment target factor included in the candidate driving environment target is determined, and the specific gene code is shown in table 1 and fig. 2. For example, the candidate driving environment target can select two environmental factors of weather and road conditions.
And step 220, combining the gene expressions of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene codes of the candidate driving environment targets.
Specifically, the gene expressions of the environmental target factors in each candidate driving environment target determined in step 210 may be combined according to the arrangement order of the environmental target factors, so as to obtain the gene code of the candidate driving environment target. Wherein, the arrangement sequence can be set according to the requirement. For example, if the environmental factors of a candidate driving environment target are weather and road conditions, the weather is semi-cloudy, and the road conditions are highway, referring to table 1, the gene code of the candidate driving environment target is 00000.
And step 230, performing automatic driving evaluation on each candidate driving environment target, and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result.
And step 240, performing gene cross operation on the gene codes of the set of parent driving environment targets aiming at each set of parent driving environment targets to obtain child driving environment targets of the set of parent driving environment targets.
Wherein the process of performing a genetic crossover operation on the genetic codes of the set of ancestor driving environment targets may comprise: selecting at least one gene coding position adjacent as a crossover position; swapping gene expression of a first parent genetic environment target of the set of parent genetic environment targets at the intersection location with gene expression of a second parent genetic environment target at the intersection location.
Specifically, at least one adjacent gene coding position may be selected as the intersection position, for example, fig. 7 is an exemplary diagram of the evolution of the gene coding of the driving environment factor in the second embodiment of the present invention, and as shown in fig. 7, parents may select the 2 nd and 3 rd gene coding positions as the intersection positions. And exchanging the gene expression of the first parent gene environment target in the set of parent gene environment targets at the intersection position with the gene expression of the second parent gene environment target at the intersection position, illustratively, as shown in fig. 7, exchanging the gene 00 of the first parent gene environment target (left side) in the parent at the intersection position with the gene 11 of the second parent gene environment target (right side) at the intersection position, and performing a genetic variation operation on the gene 1 of the last position of the second parent gene environment target, i.e., modifying, to obtain corresponding child gene expressions of 01100 and 10010.
It should be noted that, for each set of parent driving environment targets, the child driving environment targets of the set of parent driving environment targets can be obtained by performing gene cross operation on the gene codes of the set of parent driving environment targets. In addition, the method can also carry out gene mutation operation on any parent driving environment target or any child driving environment target to obtain a new child driving environment target.
And 250, modifying the gene expression of at least one position of at least one child driving environment target to obtain the gene code of the new child driving environment target.
Specifically, the genetic variation operation is performed on the gene codes of the driving environment targets of the children obtained in step 240, that is, the gene expression of at least one position of at least one driving environment target of the children is modified, so as to obtain the gene codes of the driving environment targets of the new children. Illustratively, referring to figure 7, the genes at the first position of the first child genetic environment target (left side) of the child, i.e., the second generation's parent, may be modified to obtain new child gene expressions of 10010 and 11100.
It should be noted that, for each set of parent driving environment targets, the gene codes of the set of parent driving environment targets may be subjected to gene cross operation or gene mutation operation to obtain the child driving environment targets of the set of parent driving environment targets, or the gene codes of the set of parent driving environment targets may be subjected to gene cross operation and gene mutation operation to obtain the child driving environment targets of the set of parent driving environment targets.
Fig. 8 is an example diagram of the evolution phenotype of the gene codes of the driving environment factors for automatic driving in the second embodiment of the present invention, which corresponds to the gene codes in fig. 7 one by one, and it can be seen that the environment factors of the driving environment targets of the group of parents are half-cloudy expressways and fine country roads, and the environment factors of the driving environment targets of the children obtained through two times of gene crossing operations and gene mutation operations are snowy urban roads and fine expressway, which are completely different from the environment factors of the driving environment targets of the parents.
Fig. 9 is a diagram showing an example of the evolution of the gene codes of the target factors of the automatic driving vehicles according to the second embodiment of the present invention, and as shown in fig. 9, the target factors are vehicles and pedestrians. Fig. 10 is an example of the evolution phenotype of the gene codes of the driving environment objectives of the automatic vehicle in the second embodiment of the present invention, which corresponds to the gene codes in fig. 9 one by one, and the gene codes of the objective factors of the driving environment objectives of the parent set obtain the objective factors of different child driving environment objectives through two gene crossover operations and gene mutation operations.
And step 260, if the termination condition is not met, taking the generated each child driving environment target as a new candidate driving environment target, returning to execute the automatic driving evaluation operation, and selecting a new parent driving environment target from the new candidate driving environment targets according to the evaluation result.
Specifically, if the termination condition is not satisfied, the generated child driving environment targets are used as new candidate driving environment targets, and the process returns to the step 230, i.e., the automatic driving evaluation operation is performed, a new parent driving environment target is selected from the new candidate driving environment targets according to the evaluation result, the gene coding of the new parent driving environment target is subjected to gene crossing operation and/or gene mutation operation, and the corresponding new child driving environment target is generated until the termination condition is satisfied.
In this embodiment, if the termination condition is satisfied, a certain constraint condition may be set for the child driving environment target so that the child driving environment target is consistent with the real automatic driving environment.
In this embodiment, if the termination condition is satisfied and the corresponding constraint condition is set, the environment factor and the target factor in the final child driving environment target may be called in the simulation platform (e.g., unregeal), so as to generate different automatic driving training data.
In the embodiment, the automatic virtual driving environment is generated by simulating the target in the automatic driving environment, performing gene coding, automatic driving evaluation and gene crossing and/or gene variation on the target, setting corresponding constraint conditions, and generating different automatic driving training data through the simulation platform. The technical scheme that this embodiment provided can realize more comprehensive collection of autopilot big data more simply conveniently to reduce the degree of difficulty of gathering, reduce the collection cost and increase maneuverability.
EXAMPLE III
The present embodiment may provide an example based on the above embodiments, and further describes the evolution process of the driving environment in the virtual acquisition method for big data of automatic driving training.
In this embodiment, an idea of an evolutionary algorithm may be adopted, where the evolutionary algorithm is an algorithm simulating the evolution of a natural living organism, and the process includes selecting excellent individuals as parents, and generating children from the parents by means of intersection and mutation, and the evolutionary algorithm may include the following steps: 1) given a set of initial solutions; 2) evaluating the performance of the current set of solutions; 3) selecting a certain number of solutions from the current set of solutions as a basis for the iterated solutions; 4) then, operating the solution to obtain an iterated solution; 5) and stopping if the solutions meet the requirements, otherwise, re-operating the solutions obtained by iteration as the current solutions.
Fig. 11 is a flowchart of a virtual collection method for big data of automatic driving training in the third embodiment of the present invention, and as shown in fig. 11, the technical solution of this embodiment may be an evolution process, which corresponds to a process of an evolution algorithm, and specifically may include:
And 340, selecting individual road environment, selecting a preset number of parents with higher selectivity, wherein the preset number can be set as required.
And 350, performing individual crossing, namely performing gene crossing operation on the selected ancestor individuals with preset quantity.
And step 360, carrying out individual mutation, namely carrying out gene mutation operation on the parent individuals after the gene cross operation in the step 350 to generate new child individuals.
And step 370, ending, and judging whether the ending condition is met. The termination condition may be set to a specified evolution time, a specified evolution algebra, or the performance of each child individual satisfying a certain requirement, and may be set as needed. If the termination condition is met, the evolution process is complete and step 380 is entered. If not, the process can return to step 330 again to loop until the termination condition is met.
And step 380, ending.
In this example, excellent ancestral genes are selected, excellent ancestral genes are generated from the excellent ancestral genes through gene exchange and/or genetic mutation, and new excellent ancestral genes are further generated from the excellent ancestral genes through continuous iteration. The technical scheme that this embodiment provided adopts the thought of evolutionary algorithm, can produce the virtual driving environment of automatic driving of a plurality of differences, and constantly evolution and study for the driving environment that produces constantly evolves until satisfying the requirement of training big data, thereby on this virtual driving environment of automatic driving's basis, more comprehensive automatic driving big data's collection is realized more simply conveniently, and reduces the degree of difficulty of gathering, reduces the collection cost and increases maneuverability.
Example four
Fig. 12 is a schematic structural diagram of a virtual collecting device for big data of automatic driving training in a fourth embodiment of the present invention, where the device may include:
the gene coding module 410 is used for determining the gene codes of all candidate driving environment targets for automatic virtual driving;
the driving evaluation module 420 is configured to perform automatic driving evaluation on each of the candidate driving environment targets, and select a parent driving environment target from each of the candidate driving environment targets according to an evaluation result;
the evolution module 430 is used for performing gene cross operation and/or gene mutation operation on the gene codes of the set of parent driving environment targets aiming at each set of parent driving environment targets to generate child driving environment targets of the set of parent driving environment targets;
and the return evaluation module 440 is configured to, if the termination condition is not satisfied, take the generated child driving environment targets as new candidate driving environment targets, return to execute the automatic driving evaluation operation, and select a new parent driving environment target from the new candidate driving environment targets according to an evaluation result.
Further, the evolution module 430 may include a gene crossing unit, which may be configured to:
selecting at least one gene coding position adjacent as a crossover position;
swapping gene expression of a first parent genetic environment target of the set of parent genetic environment targets at the intersection location with gene expression of a second parent genetic environment target at the intersection location.
Further, the evolution module 430 may further include a genetic cross mutation unit, and the genetic cross mutation unit may be configured to:
performing gene cross operation on the gene codes of the set of parent driving environment targets to obtain child driving environment targets of the set of parent driving environment targets;
modifying the gene expression of at least one position of at least one said child driving environment target to obtain the gene coding of the new child driving environment target.
Further, the termination condition may be that the number of times of generating the child driving environment target is greater than a number threshold, or that a difference between the automatic driving evaluation value and the evaluation expectation value of the generated child driving environment target is less than a difference threshold.
Further, the gene coding module 410 can be specifically used for:
determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target;
and combining the gene expression of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene code of the candidate driving environment target.
Further, the candidate driving environment target may include at least one environmental factor of weather, road conditions, trees, buildings and traffic lights, and at least one target factor of environmental vehicles and pedestrians.
Further, the automatic driving evaluation vehicle is located differently from the target factor in which the locations of the surrounding vehicles and pedestrians are different.
The virtual acquisition device for big automatic driving training data provided by the embodiment of the invention can execute the virtual acquisition method for big automatic driving training data provided by any embodiment of the invention, and has the following beneficial effects: the cost is low, and road data do not need to be acquired on site by using a vehicle-mounted platform; the operability is high, the method is not limited by natural conditions, safety and the like, and road data of different natural conditions, different experimental conditions and the like in a simulated environment can be collected at any time; the data is comprehensive, and the existing automatic driving road data acquisition method cannot be used for investing enough vehicles to acquire data of various road conditions and environments all over the world.
The virtual acquisition device for the big automatic driving training data provided by the embodiment of the invention can execute the virtual acquisition method for the big automatic driving training data provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (13)
1. A virtual acquisition method for big data of automatic driving training is characterized by comprising the following steps:
determining gene codes of candidate driving environment targets for automatic virtual driving;
performing automatic driving evaluation on each candidate driving environment target, and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result;
aiming at each group of parent driving environment targets, performing gene cross operation and/or gene mutation operation on gene codes of the group of parent driving environment targets to generate child driving environment targets of the group of parent driving environment targets;
if the terminal condition is not met, taking the generated child driving environment targets as new candidate driving environment targets, returning to execute automatic driving evaluation operation, and selecting new parent driving environment targets from the new candidate driving environment targets according to an evaluation result; wherein the determining of the gene codes of the candidate driving environment targets for the automatic virtual driving comprises the following steps:
determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target;
and combining the gene expression of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene code of the candidate driving environment target.
2. The method of claim 1, wherein performing a genetic crossover operation on the genetic codes of the set of ancestor driving environment targets comprises:
selecting at least one gene coding position adjacent as a crossover position;
swapping gene expression of a first parent genetic environment target of the set of parent genetic environment targets at the intersection location with gene expression of a second parent genetic environment target at the intersection location.
3. The method of claim 1, wherein performing genetic crossover and mutation operations on the genetic code of the set of ancestral driving environment targets comprises:
performing gene cross operation on the gene codes of the set of parent driving environment targets to obtain child driving environment targets of the set of parent driving environment targets;
modifying the gene expression of at least one position of at least one said child driving environment target to obtain the gene coding of the new child driving environment target.
4. The method of claim 1, wherein the termination condition is that the number of times the child driving environment target is generated is greater than a number threshold, or a difference between the automatic driving evaluation value and the evaluation expectation value of the generated child driving environment target is less than a difference threshold.
5. The method of claim 1, wherein the candidate driving environment targets include at least one environmental factor selected from the group consisting of weather, road conditions, trees, buildings, and traffic lights, and at least one target factor selected from the group consisting of environmental vehicles and pedestrians.
6. The method of claim 5, wherein the automated driving assessment vehicle is located differently than the target factor, wherein the target factor is located differently between the environmental vehicle and the pedestrian.
7. A virtual collection device of automatic driving training big data, characterized by includes:
the gene coding module is used for determining the gene codes of all candidate driving environment targets for automatic virtual driving;
the driving evaluation module is used for automatically evaluating the driving of each candidate driving environment target and selecting a parent driving environment target from each candidate driving environment target according to an evaluation result;
the evolution module is used for carrying out gene cross operation and/or gene mutation operation on the gene codes of the set of parent driving environment targets aiming at each set of parent driving environment targets to generate child driving environment targets of the set of parent driving environment targets;
the return evaluation module is used for taking each generated child driving environment target as a new candidate driving environment target if the termination condition is not met, returning to execute automatic driving evaluation operation, and selecting a new parent driving environment target from the new candidate driving environment targets according to an evaluation result;
the gene coding module can be specifically used for determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target; and combining the gene expression of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene code of the candidate driving environment target.
8. The apparatus of claim 7, wherein the evolution module comprises a gene crossing unit configured to:
selecting at least one gene coding position adjacent as a crossover position;
swapping gene expression of a first parent genetic environment target of the set of parent genetic environment targets at the intersection location with gene expression of a second parent genetic environment target at the intersection location.
9. The apparatus of claim 7, wherein the evolution module further comprises a genetic cross mutation unit configured to:
performing gene cross operation on the gene codes of the set of parent driving environment targets to obtain child driving environment targets of the set of parent driving environment targets;
modifying the gene expression of at least one position of at least one said child driving environment target to obtain the gene coding of the new child driving environment target.
10. The apparatus of claim 7, wherein the termination condition is that the number of times the child driving environment target is generated is greater than a number threshold, or a difference between the automatic driving evaluation value and the evaluation expectation value of the generated child driving environment target is less than a difference threshold.
11. The device according to claim 7, wherein the gene coding module is specifically configured to:
determining the gene expression of each environment target factor contained in each candidate driving environment target aiming at each candidate driving environment target;
and combining the gene expression of the environmental target factors according to the arrangement sequence of the environmental target factors to obtain the gene code of the candidate driving environment target.
12. The apparatus of claim 11, wherein the candidate driving environment targets comprise at least one environmental factor selected from the group consisting of weather, road conditions, trees, buildings, and traffic lights, and target factors comprising at least one of environmental vehicles and pedestrians.
13. The apparatus of claim 12, wherein the automated driving assessment vehicle is located differently than the target factor, wherein the target factor is located differently between the environmental vehicle and the pedestrian.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810122666.2A CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810122666.2A CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108268740A CN108268740A (en) | 2018-07-10 |
| CN108268740B true CN108268740B (en) | 2021-11-16 |
Family
ID=62773863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810122666.2A Active CN108268740B (en) | 2018-02-07 | 2018-02-07 | Virtual acquisition method and device for big data of automatic driving training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108268740B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108959813B (en) * | 2018-07-26 | 2021-01-15 | 北京理工大学 | Simulation modeling method for intelligent vehicle road navigation environment model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080297377A1 (en) * | 2007-05-28 | 2008-12-04 | High Tech Computer Corp. | Keypad structure and electronic device using the same |
| CN103218489A (en) * | 2013-04-11 | 2013-07-24 | 浙江大学 | Method for simulating vehicle personalized driving characteristic based on video sample |
| CN106154834A (en) * | 2016-07-20 | 2016-11-23 | 百度在线网络技术（北京）有限公司 | For the method and apparatus controlling automatic driving vehicle |
-
2018
- 2018-02-07 CN CN201810122666.2A patent/CN108268740B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080297377A1 (en) * | 2007-05-28 | 2008-12-04 | High Tech Computer Corp. | Keypad structure and electronic device using the same |
| CN103218489A (en) * | 2013-04-11 | 2013-07-24 | 浙江大学 | Method for simulating vehicle personalized driving characteristic based on video sample |
| CN106154834A (en) * | 2016-07-20 | 2016-11-23 | 百度在线网络技术（北京）有限公司 | For the method and apparatus controlling automatic driving vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108268740A (en) | 2018-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11210200B2 (en) | Software validation for autonomous vehicles | |
| US11036232B2 (en) | Iterative generation of adversarial scenarios | |
| CN112307566B (en) | Vehicle simulation test method, device, equipment and storage medium | |
| CN109101907B (en) | Vehicle-mounted image semantic segmentation system based on bilateral segmentation network | |
| CN108416783B (en) | Road scene segmentation method based on full convolution neural network | |
| US20190382013A1 (en) | Vehicle control device, vehicle control method, and storage medium | |
| CN110716529A (en) | Automatic generation method and device for automatic driving test case | |
| CN111797001A (en) | Method for constructing automatic driving simulation test model based on SCANeR | |
| CN111310765A (en) | Laser point cloud semantic segmentation method and device | |
| Liu et al. | A large-scale simulation dataset: Boost the detection accuracy for special weather conditions | |
| CN115048972B (en) | Traffic scene deconstruction classification method and virtual-real combined automatic driving test method | |
| CN115098079B (en) | Radar detection model determination method, system, electronic device and readable storage medium | |
| CN110763483A (en) | Automatic generation method and device of security level test scene library | |
| CN107590766A (en) | A kind of method of discrimination of the land used combination form related to road traffic accident risk | |
| CN112987711A (en) | Optimization method of automatic driving regulation algorithm and simulation testing device | |
| CN115165387A (en) | Control method, device and system for testing performance of automatic driving whole vehicle | |
| CN108268740B (en) | Virtual acquisition method and device for big data of automatic driving training | |
| CN116597690B (en) | Highway test scene generation method, equipment and medium for intelligent network-connected automobile | |
| CN117521389A (en) | Vehicle perception test method based on vehicle-road collaborative perception simulation platform | |
| CN111368465A (en) | Unmanned decision-making method based on ID3 decision tree | |
| CN115855531A (en) | Test scene construction method, device and medium for automatic driving automobile | |
| Cutrone et al. | A framework for identifying and simulating worst-case animal-vehicle interactions | |
| CN114987495A (en) | Man-machine hybrid decision-making method for highly automatic driving | |
| Li | Exploration of the Assessment for AVP Algorithm Training in Underground Parking Garages Simulation Scenario | |
| CN116680186B (en) | Vehicle-road collaborative automatic driving simulation test method, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |