CN110991095A - Training method and device for vehicle driving decision model - Google Patents
Training method and device for vehicle driving decision model Download PDFInfo
- Publication number
- CN110991095A CN110991095A CN202010145000.6A CN202010145000A CN110991095A CN 110991095 A CN110991095 A CN 110991095A CN 202010145000 A CN202010145000 A CN 202010145000A CN 110991095 A CN110991095 A CN 110991095A
- Authority
- CN
- China
- Prior art keywords
- environment information
- sample
- trained
- virtual
- driving decision
- 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.)
- Granted
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W2050/0001—Details of the control system
- B60W2050/0019—Control system elements or transfer functions
- B60W2050/0028—Mathematical models, e.g. for simulation
- B60W2050/0031—Mathematical model of the vehicle
Landscapes
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Human Computer Interaction (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Traffic Control Systems (AREA)
Abstract
The specification discloses a training method and a device for a vehicle driving decision model. The present specification generates virtual environment information (i.e., sample environment information) similar to the actual environment information from the pre-trained first GAN, inputs the sample environment information to the pre-trained second GAN, generates a virtual driving decision (i.e., a sample driving decision) matching the sample environment information, and trains a vehicle driving decision model using the sample environment information and the sample driving decision corresponding to the sample environment information. The training mode reduces the dependence of the vehicle driving decision model on historical data, and when the vehicle is in an environment with a complex traffic condition, the driving decision suitable for the environment where the vehicle is located and the vehicle driving state can be output, so that the generalization of the vehicle driving decision model is strong.
Description
Technical Field
The specification relates to the technical field of unmanned driving, in particular to a training method and a training device for a vehicle driving decision model.
Background
At present, the intellectualization of vehicles is an important component of artificial intelligence technology, and the function of the vehicles in social production and life is increasingly prominent, so that the vehicles become one of the main directions for guiding the development of traffic technology.
In the prior art, unmanned vehicles and vehicles with driving assistance functions (hereinafter, collectively referred to as "vehicles") often adopt preset decision models, and a driving decision adapted to an environment where the vehicle is located is obtained according to analysis of the environment, so that the vehicle can drive according to the driving decision. It can be seen that the accuracy of the driving decision output by the decision model is determined by the training of the decision model.
Currently, in the training process of decision models, the models need to be trained according to a large amount of known expert data. The expert data generally includes the motion of the vehicle and the environment of the vehicle at each time point, and the model can be trained by taking the motion of the vehicle at the current time and the environment of the vehicle as input and taking the motion of the vehicle at the next time as a label.
However, in the actual operation process, the acquisition of the expert data is complicated, and after the acquisition, the expert data needs to be manually screened and subjected to processing such as labeling, so that the workload is large. Moreover, if the model is trained only according to the expert data, the trained decision model has strong dependency on the expert data, the training effect of the decision model is affected, the generalization of the model is poor, and the trained decision model cannot output a driving decision which is suitable for the environment where the vehicle is located and the driving state of the vehicle, so that the driving safety of the vehicle is endangered. The above drawbacks will be more apparent when the vehicle is in an environment with more complex traffic conditions.
Disclosure of Invention
The embodiment of the specification provides a training method and a training device for a vehicle driving decision model, so as to partially solve the problems in the prior art.
The embodiment of the specification adopts the following technical scheme:
the training method for the vehicle driving decision model provided by the specification comprises the following steps:
inputting actual environment information in historical data into a first generator in a pre-trained first generative countermeasure network GAN to obtain virtual environment information output by the first generator;
inputting the virtual environment information into a first discriminator in the first GAN, so that the first discriminator judges whether the virtual environment information is actual environment information;
taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputting the sample environment information to a second generator in a second GAN trained in advance to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator;
inputting the virtual driving decisions to a second discriminator in the second GAN, so that the second discriminator respectively judges whether each virtual driving decision matches with the sample environmental information;
taking the virtual driving decision which is determined by the second discriminator to be matched with the sample environment information as a sample driving decision;
and training a vehicle driving decision model to be trained according to the sample environment information and the sample driving decision matched with the sample environment information.
Optionally, inputting the actual environment information in the historical data to a first generator in a pre-trained first generative confrontation network GAN to obtain the virtual environment information output by the first generator, and specifically including: scrambling actual environment information in the historical data; and inputting the scrambled actual environment information into a first generator in a first pre-trained GAN to obtain the virtual environment information output by the first generator.
Optionally, the pre-training of the first GAN specifically includes: inputting actual environment information in historical data into a first generator in the first GAN to be trained to obtain virtual environment information output by the first generator to be trained; inputting the virtual environment information output by the first generator to be trained into a first discriminator in the first GAN to be trained, so that the first discriminator to be trained judges whether the virtual environment information output by the first generator to be trained is actual environment information; and training the first GAN until the first generator to be trained and the first discriminator to be trained reach Nash balance.
Optionally, the pre-training of the second GAN specifically includes: taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputting the sample environment information to a second generator in the second GAN to be trained to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained; inputting each virtual driving decision corresponding to the sample environmental information and output by the second generator to be trained into a second discriminator in the second GAN to be trained, so that the second discriminator to be trained respectively judges whether each virtual driving decision is matched with the sample environmental information; and training the second GAN by using the second arbiter to be trained to determine that the probability of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained is the maximum, and determining whether the accuracy of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained with the sample environment information is the maximum by the second arbiter to be trained.
Optionally, the second determiner is configured to determine, for each virtual driving decision, a matching degree between the virtual driving decision and the sample environment information, and determine that each virtual driving decision matches the sample environment information when the matching degree between the virtual driving decision and the sample environment information is greater than a specified threshold, and determine that each virtual driving decision does not match the sample environment information when the matching degree between the virtual driving decision and the sample environment information is not greater than the specified threshold; training the second GAN, specifically comprising: and in the process of training the second GAN, when a specified condition is met, increasing the specified threshold until the specified threshold reaches a preset threshold.
Optionally, the actual environment information or the virtual environment information both include: at least one of a road map, a traffic route, a traffic light, a speed of the vehicle, and status information of obstacles in the environment perceived by the vehicle; the virtual driving decision comprises: and (5) vehicle pose.
Optionally, scrambling actual environment information in the history data specifically includes: and scrambling at least one of traffic signal lamps, the speed of the vehicle and state information of obstacles in the environment sensed by the vehicle in the actual environment information.
The present specification provides a training device for a vehicle driving decision model, including:
the first input module is used for inputting actual environment information in historical data to a first generator in a pre-trained first generative countermeasure network GAN to obtain virtual environment information output by the first generator;
a first judgment module, configured to input the virtual environment information to a first discriminator in the first GAN, so that the first discriminator judges whether the virtual environment information is actual environment information;
the second input module is used for taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information and inputting the sample environment information into a second generator in a second GAN trained in advance to obtain each virtual driving decision which is output by the second generator and corresponds to the sample environment information;
a second determining module, configured to input the virtual driving decisions to a second determiner in the second GAN, so that the second determiner determines whether each virtual driving decision matches the sample environment information;
a determination module, configured to take the virtual driving decision determined by the second discriminator to match the sample environmental information as a sample driving decision;
and the training module is used for training a vehicle driving decision model to be trained according to the sample environment information and the sample driving decision matched with the sample environment information.
The present specification provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-described method for training a vehicle driving decision model.
The electronic device provided by the specification comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor realizes the training method of the vehicle driving decision model when executing the program.
The embodiment of the specification adopts at least one technical scheme which can achieve the following beneficial effects:
in the embodiment of the specification, virtual environment information (i.e., sample environment information) similar to the actual environment information is generated by pre-trained first GAN, the sample environment information is input to pre-trained second GAN to generate a virtual driving decision (i.e., a sample driving decision) matched with the sample environment information, and then a vehicle driving decision model is trained by using the sample environment information and the sample driving decision corresponding to the sample environment information.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:
fig. 1 is a schematic diagram of a training system architecture of a vehicle driving decision model according to an embodiment of the present disclosure;
fig. 2 is a schematic flowchart of a method for training a vehicle driving decision model according to an embodiment of the present disclosure;
fig. 3 is a schematic structural diagram of a training device for a vehicle driving decision model according to an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of an electronic device provided in an embodiment of this specification.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.
The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.
In the embodiment of the present disclosure, a vehicle driving decision model may be trained by using a system architecture as shown in fig. 1, where the system architecture mainly includes a first Generative Adaptive Network (GAN), a second GAN, and a vehicle driving decision model. The first GAN comprises a first generator and a first discriminator, and the second GAN comprises a second generator and a second discriminator.
Inputting actual environment information to a first generator, so that the first generator generates virtual environment information; and inputting the virtual environment information into a second discriminator to enable the second discriminator to judge whether the input virtual environment information is actual environment information or not, and feeding back the judgment result to the first generator. The first generator takes the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputs the sample environment information to the second generator, so that the second generator generates each virtual driving decision for the sample environment information; and inputting each virtual driving decision into a second discriminator to enable the second discriminator to judge whether each virtual driving decision is matched with the sample environment information or not, and feeding back a judgment result to a second generator. The second generator takes the virtual driving decision that the second discriminator determines to match the sample environmental information as a sample driving decision. And training the vehicle driving decision model through the sample environment information and a sample driving decision corresponding to the sample environment information.
The above process will be described in detail with reference to the accompanying drawings. As shown in fig. 2, fig. 2 is a schematic flowchart of a training method for a vehicle driving decision model provided in the present specification, where the schematic flowchart includes:
s100: and inputting the actual environment information in the historical data into a first generator in a pre-trained first generative countermeasure network GAN to obtain the virtual environment information output by the first generator.
The historical data may be vehicle status and actual environmental information when the driver drives the vehicle under different road conditions. The actual environment information may include at least one of a road map, a traffic route, a traffic signal, a speed of the vehicle, and status information of obstacles in the environment perceived by the vehicle. The obstacle can be a dynamic obstacle (such as other vehicles and pedestrians) and a static obstacle (such as a rail), and the state information of the obstacle can be the pose, the shape and the like of the obstacle.
The actual environment information can be acquired through a sensor carried by the vehicle and can also be acquired through a control system of the vehicle. In addition, the actual environment information may also be determined by the cloud device based on where the vehicle is located. In short, there are various ways to acquire history data such as actual environment information, and the embodiment of the present specification is not limited to this.
As shown in fig. 1, a first generator trained in advance outputs virtual environment information by actual environment information in history data. Although the virtual environment information is not the actual environment information, the virtual environment information is similar to the actual environment information, the diversity of data can be improved, the generation process is more efficient, the cost in historical data acquisition is greatly reduced, the dependence on the historical data in the subsequent vehicle running decision model training is also reduced, and the generalization is stronger.
S102: the virtual environment information is input to a first discriminator in the first GAN, so that the first discriminator judges whether the virtual environment information is actual environment information.
The pre-trained first discriminator can use actual environment information in historical data as a reference standard to judge virtual environment information output by the pre-trained first generator, if the virtual environment information is judged to be the actual environment information, the virtual environment information output by the first generator is similar to the actual environment information, and the virtual environment information can be subsequently used for training a vehicle driving decision model.
S104: and taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputting the sample environment information into a second generator in a second GAN trained in advance to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator.
Since the sample environment information is the virtual environment information determined by the first discriminator as the actual environment information (i.e., the degree of similarity between the virtual environment information and the actual environment information is extremely high), the second generator outputs each virtual driving decision corresponding to the sample environment information. Wherein the second generator identifies virtual driving decisions suitable for execution by the vehicle under the sample environmental information, referred to as virtual driving decisions "corresponding to" the sample environmental information. For a sample environment information, there may be one or more virtual driving decisions corresponding thereto, and the embodiments of the present specification are not limited thereto.
S106: and inputting each virtual driving decision into a second discriminator in the second GAN, so that the second discriminator respectively judges whether each virtual driving decision is matched with the sample environment information.
Continuing with the above example, as shown in fig. 1, although the second generator is capable of outputting virtual driving decisions corresponding to the sample environmental information, the virtual driving decisions that the second generator deems appropriate for the sample environmental information may not all match the sample environmental information in consideration of the actual traffic conditions being more complex. At this time, the second discriminator which needs to be trained in advance respectively judges whether each virtual driving decision is matched with the sample environment information.
S108: and taking the virtual driving decision which is determined by the second determiner to be matched with the sample environment information as a sample driving decision.
The virtual driving decision which is determined by the pre-trained second discriminator to be matched with the sample environment information is used as a sample driving decision, and when a subsequent vehicle driving decision model is trained, the accuracy of the model can be improved.
S110: and training the vehicle driving decision model to be trained according to the sample environment information and the sample driving decision matched with the sample environment information.
The sample environmental information and the sample driving decision matched with the sample environmental information can be used as labels to train the vehicle driving decision model to be trained. If the vehicle driving decision model is a reinforcement learning model, the training process of the vehicle driving decision model may include: acquiring current vehicle state information and current environment information; taking the sample environment information and the sample driving decision matched with the sample environment information as labels; inputting current vehicle state information and current environment information into a vehicle running decision model to be trained to obtain a vehicle running decision output by the vehicle running decision model to be trained; determining the reward of the vehicle driving decision according to the vehicle driving decision and the label; and training the vehicle driving decision model to be trained by taking the reward maximization as a training target to obtain the trained vehicle driving decision model.
The vehicle state information may include, among other things, a vehicle pose (e.g., a position of the vehicle, a vehicle attitude, an orientation of the vehicle, etc.), a vehicle speed, and the like. The current vehicle state information can be acquired through a sensor of the vehicle and can also be acquired through a control system of the vehicle. The current environment information may also be obtained in the above manner, which is not described herein again. The output vehicle running decision can be embodied in the form of a running track.
In addition, the vehicle running decision model to be trained can be trained according to the actual environment information in the historical data, the vehicle state information in the historical data, the sample environment information and the sample running decision matched with the sample environment information. That is, the training of the vehicle driving decision model combines both the historical data with strong dependency and the stochastic data similar to the historical data (i.e., the sample environmental information, the actual environmental information matched with the sample environmental information, etc.), and combines the data with strong dependency and large randomness to train the model, so that the diversity of the data is improved, the dependency of the model on the historical data is reduced, and the generalization of the vehicle driving decision model can be improved.
Embodiments of the present description employ a GAN network to generate virtual environment information. The GAN network trains the generator and the discriminator respectively according to the maximum and minimum game principle, so that data generated by the GAN network has authenticity. In addition, because the GAN network is based on unsupervised learning, compared with a mode of training a model through a large amount of historical data, the method greatly reduces the cost, provides richer data for model training, increases the diversity of data of the training model, and can solve the problems of extremely high cost and low generalization of the trained model when the model is trained through the historical data only. Taking the traffic signal lamp (red lamp, green lamp, yellow lamp) processing as an example, if the information similar to the actual traffic signal lamp is generated by the similarity method, a blue lamp may appear, although the similarity between the blue lamp and the green lamp is high, the blue lamp does not meet the actual situation, and if the GAN network is adopted to process the traffic signal lamp, the similar problem does not appear.
The embodiment of the present specification generates virtual environment information by one GAN, and then inputs the generated virtual environment information to another GAN, so that it generates a virtual driving decision matching the virtual environment information. Because the nature of the environmental information is different from that of the driving decision, the environmental information required in the subsequent model training can be the environmental information similar to the actual environmental information, but the required driving decision is not the driving track similar to the actual driving track but is adapted to the driving track of the environmental information, and the accuracy of the model can be improved only by training the model through the data.
It can be seen that the training purpose of one GAN is to generate virtual environment information similar to the actual environment information, and the training purpose of another GAN is to generate virtual driving decisions adapted to the virtual environment information. In the embodiment of the specification, two GANs are trained respectively based on two different purposes, so that the training pertinence is stronger, the structure of each GAN is simpler, the training speed is increased, and the accuracy of the model is also improved. However, if the above two training purposes are achieved by only one GAN (i.e. generating both approximate data and adaptive data), the modeling is very complicated, the training speed is slow, and the accuracy of the model is difficult to guarantee.
After the vehicle driving decision model is trained, the vehicle driving decision model can be used for determining the driving track of the vehicle. The specific process can be as follows: when the running track of a vehicle is to be determined, the vehicle state information and the environment information monitored by the vehicle are acquired; inputting the monitored vehicle state information and the monitored environment information into a trained vehicle running decision model to obtain a vehicle running decision output by the trained vehicle running decision model; and controlling the vehicle according to the vehicle running decision output by the trained vehicle running decision model.
Optionally, for a specific implementation manner of step S100, the actual environment information in the history data may be scrambled, and then the scrambled actual environment information is input to the first generator in the pre-trained first GAN, and the first generator outputs the virtual environment information according to the scrambled actual environment information. Wherein, actual environment information or virtual environment information all includes: at least one of a road map, a traffic route, a traffic light, a speed of the vehicle, and status information of obstacles in the environment perceived by the vehicle, the virtual driving decision comprising: and (5) vehicle pose. Then, scrambling the actual environment information in the history data may specifically be: at least one of traffic signal lamps, the speed of the vehicle and state information of obstacles in the environment sensed by the vehicle in the actual environment information is scrambled. It should be noted that, when the actual environment information is scrambled, the road map and the traffic route are not scrambled, and the road map and the traffic route are considered as the real road condition, if the actual environment information is scrambled, the scrambled actual environment information may be unavailable, which affects the accuracy of the model training.
The actual environmental information in the historical data includes actual images that may be captured by sensors onboard the vehicle or acquired by a control system of the vehicle. Embodiments of the present description may first convert the actual image into an abstract image, and then input the abstract image to a first generator in a first GAN that is pre-trained. Converting the actual image into an abstract image, wherein the concrete process comprises the following steps: identifying each key element contained in the actual image information; determining the positions of the key elements according to the electronic map and the actual image information; and for each key element, adding a preset model matched with the key element into the abstract image in a specified mode according to the position of the key element. The discrete environmental information is converted into the image and then input into the first generator, so that the training of the model can be facilitated. Furthermore, the mode of converting the actual image into the abstract image can remove some elements which are irrelevant to model training, such as surrounding trees, houses and the like, in the actual image, and can accelerate the training speed of the model. In addition, if the actual image is acquired in a top view mode, the effectiveness of the environmental information is stronger.
Pre-training a first GAN, specifically comprising: inputting actual environment information in the historical data into a first generator in a first GAN to be trained to obtain virtual environment information output by the first generator to be trained; inputting the virtual environment information output by the first generator to be trained into a first discriminator in the first GAN to be trained, so that the first discriminator to be trained judges whether the virtual environment information output by the first generator to be trained is actual environment information (namely real environment information); the virtual environment information output by the first generator to be trained is closer to the actual environment information, the probability that the virtual environment information output by the first generator to be trained is judged to be the actual environment information by the first discriminator to be trained is the minimum, the first GAN is trained as a training target until the first generator to be trained and the first discriminator to be trained reach Nash balance, namely, the virtual environment information output by the first generator to be trained is not different from the actual environment information, and the probability that the first discriminator to be trained judges whether the environment information is true or false is about 50%.
Pre-training the second GAN specifically includes: the virtual environment information which is determined as the actual environment information by the first discriminator is used as sample environment information and is input to a second generator in a second GAN to be trained, and each virtual driving decision which is output by the second generator to be trained and corresponds to the sample environment information is obtained; inputting each virtual driving decision corresponding to the sample environment information and output by a second generator to be trained into a second discriminator in a second GAN to be trained, so that the second discriminator to be trained respectively judges whether each virtual driving decision is matched with the sample environment information; and training a second GAN by taking the maximum probability that each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained is judged to be matched with the sample environment information by the second discriminator to be trained, and taking the maximum accuracy that whether each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained is matched with the sample environment information or not as a training target by the second discriminator to be trained.
Further, the second discriminator is used for determining the matching degree of the virtual driving decision and the sample environment information aiming at each virtual driving decision, and when the matching degree of the virtual driving decision and the sample environment information is greater than a specified threshold value, each virtual driving decision is judged to be matched with the sample environment information, and when the matching degree of the virtual driving decision and the sample environment information is not greater than the specified threshold value, each virtual driving decision is judged to be not matched with the sample environment information; training the second GAN, specifically comprising: and in the process of training the second GAN, when a specified condition is met, the specified threshold is increased until the specified threshold reaches a preset threshold.
It should be noted that the matching degree between each virtual driving decision and the sample environment information can be represented by a score. Specifically, the vehicle can be driven according to each virtual driving decision in the simulation environment, each virtual driving decision is respectively scored according to the driving track of the vehicle and the interaction degree with the obstacle, and the smoother the track and/or the fewer obstacles collide, the higher the corresponding score is, the higher the score is, which indicates that the matching degree of the virtual driving decision and the sample environment information is higher. The simulation environment in the embodiments of the present description may be a real simulation environment or a virtual simulation environment. The specified conditions can be as follows: for a sample environmental information, at least two thirds of the plurality of virtual driving decisions output by the second generator are determined by the second determiner to match the sample environmental information, and then the specified threshold may be increased by 10 points.
The following describes, by way of example, training of the second GAN, where the preset threshold is set to 80 points, the designated threshold is first set to 30 points, and at this time, for one sample environment information, if the second generator outputs 12 virtual driving decisions corresponding to the sample environment information, and the second discriminator determines that the matching degree score between at least 8 virtual driving decisions and the sample environment information is not less than 30 points, the designated threshold is increased to 40 points. And continuing to train a second GAN, wherein aiming at one sample environment information, if the second generator outputs 9 virtual driving decisions corresponding to the sample environment information and the second discriminator judges that the matching degree score between at least 6 virtual driving decisions and the sample environment information is not less than 40 minutes, the assigned threshold value is increased to 50 minutes, and the like until the score of the assigned threshold value is increased to 80 minutes (reaches a preset threshold value), and the second GAN is trained.
In addition, a second GAN can be trained after the first GAN is trained; the first GAN may also be trained simultaneously with the second GAN. That is, when the virtual environment information determined as the actual environment information by the first discriminator is input to the second generator in the second GAN to be trained as the sample environment information during the training of the second GAN, the first discriminator at this time may be the trained first discriminator or the first discriminator during the training. During the training of the first GAN and the second GAN, random Gradient Descent (SGD) or Adaptive ∇ Moment Estimation (ADAM) training may be used.
The first generator in the first GAN may be a Convolutional Neural Network (CNN), the first discriminator in the first GAN may be a Convolutional Neural Network (CNN), the second generator in the second GAN may be a Convolutional Neural Network (CNN), and the second discriminator in the second GAN may be a Responsibility-sensitive safety (RSS) model. Of course, other models can be used, and the embodiments of the present disclosure are not limited thereto.
The embodiment of the present specification may generate virtual environment information (i.e., sample environment information) similar to the actual environment information from the actual environment information through the first GAN, input the sample environment information into the second GAN to generate a virtual driving decision (i.e., a sample driving decision) matching the sample environment information, and train a vehicle driving decision model using the sample environment information and the sample driving decision corresponding to the sample environment information. The training mode reduces the dependence of the vehicle driving decision model on historical data, adopts less historical data to acquire more data similar to the historical data, and enables the data of the vehicle driving decision model to be more diversified.
The training process of the vehicle driving decision model provided by the specification can be particularly applied to training of a driving decision model for an unmanned vehicle. The unmanned vehicle can be an unmanned distribution vehicle, and the unmanned distribution vehicle can be applied to the field of distribution by using the unmanned distribution vehicle, such as the distribution scene of express delivery, takeaway and the like by using the unmanned distribution vehicle. Specifically, in the above-described scenario, delivery may be performed using an autonomous vehicle fleet configured with a plurality of unmanned delivery vehicles.
Based on the same idea, the present specification further provides a corresponding apparatus, a storage medium, and an electronic device.
Fig. 3 is a schematic structural diagram of a training device for a vehicle driving decision model according to an embodiment of the present disclosure, where the training device includes:
a first input module 200, configured to input actual environment information in the historical data to a first generator in a pre-trained first generative confrontation network GAN, so as to obtain virtual environment information output by the first generator;
a first determining module 202, configured to input the virtual environment information to a first discriminator in the first GAN, so that the first discriminator determines whether the virtual environment information is actual environment information;
a second input module 204, configured to use the virtual environment information determined as the actual environment information by the first determiner as sample environment information, and input the sample environment information to a second generator in a second GAN trained in advance, so as to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator;
a second determining module 206, configured to input the virtual driving decisions to a second determiner in the second GAN, so that the second determiner determines whether each virtual driving decision matches the sample environment information;
a determining module 208, configured to use the virtual driving decision determined by the second discriminator to match the sample environmental information as a sample driving decision;
the training module 210 is configured to train a vehicle driving decision model to be trained according to the sample environment information and a sample driving decision matched with the sample environment information.
Optionally, the first input module 200 is specifically configured to scramble actual environment information in the historical data; and inputting the scrambled actual environment information into a first generator in a first pre-trained GAN to obtain the virtual environment information output by the first generator.
Optionally, the apparatus further includes a first pre-training module 212, specifically configured to input actual environment information in the historical data to the first generator in the first GAN to be trained in advance, so as to obtain virtual environment information output by the first generator to be trained; inputting the virtual environment information output by the first generator to be trained into a first discriminator in the first GAN to be trained, so that the first discriminator to be trained judges whether the virtual environment information output by the first generator to be trained is actual environment information; and training the first GAN until the first generator to be trained and the first discriminator to be trained reach Nash balance.
Optionally, the apparatus further includes a second pre-training module 214, specifically configured to take virtual environment information determined as actual environment information by the first discriminator as sample environment information in advance, and input the sample environment information to a second generator in the second GAN to be trained, so as to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained; inputting each virtual driving decision corresponding to the sample environmental information and output by the second generator to be trained into a second discriminator in the second GAN to be trained, so that the second discriminator to be trained respectively judges whether each virtual driving decision is matched with the sample environmental information; and training the second GAN by using the second arbiter to be trained to determine that the probability of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained is the maximum, and determining whether the accuracy of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained with the sample environment information is the maximum by the second arbiter to be trained.
Optionally, the second pre-training module 214 is further configured to enable the second discriminator to determine, for each virtual driving decision, a matching degree between the virtual driving decision and the sample environment information, determine that each virtual driving decision matches with the sample environment information when the matching degree between the virtual driving decision and the sample environment information is greater than a specified threshold, and determine that each virtual driving decision does not match with the sample environment information when the matching degree between the virtual driving decision and the sample environment information is not greater than the specified threshold; training the second GAN, specifically comprising: and in the process of training the second GAN, when a specified condition is met, increasing the specified threshold until the specified threshold reaches a preset threshold.
Optionally, the actual environment information or the virtual environment information both include: at least one of a road map, a traffic route, a traffic light, a speed of the vehicle, and status information of obstacles in the environment perceived by the vehicle; the virtual driving decision comprises: and (5) vehicle pose. The first input module 200 is specifically configured to scramble at least one of the actual environment information, a traffic signal lamp, a speed of a vehicle, and state information of each obstacle in the environment sensed by the vehicle.
The present specification also provides a computer readable storage medium storing a computer program which, when executed by a processor, is operable to perform a method of training a vehicle driving decision model as provided above with respect to fig. 2.
Based on the training method of the vehicle driving decision model shown in fig. 2, the embodiment of the present specification further provides a schematic structural diagram of the electronic device shown in fig. 4. As shown in fig. 4, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, but may also include hardware required for other services. The processor reads a corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to implement the training method of the vehicle driving decision model described in fig. 2.
Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Language Description Language), traffic, pl (core unified Programming Language), HDCal, JHDL (Java Hardware Description Language), langue, Lola, HDL, laspam, hardsradware (Hardware Description Language), vhjhd (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The description has been presented with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the description. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims (10)
1. A method for training a vehicle driving decision model, comprising:
inputting actual environment information in historical data into a first generator in a pre-trained first generative countermeasure network GAN to obtain virtual environment information output by the first generator;
inputting the virtual environment information into a first discriminator in a first GAN, so that the first discriminator judges whether the virtual environment information is actual environment information;
taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputting the sample environment information to a second generator in a second GAN trained in advance to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator;
inputting the virtual driving decisions to a second discriminator in the second GAN, so that the second discriminator respectively judges whether each virtual driving decision matches with the sample environmental information;
taking the virtual driving decision which is determined by the second discriminator to be matched with the sample environment information as a sample driving decision;
and training a vehicle driving decision model to be trained according to the sample environment information and the sample driving decision matched with the sample environment information.
2. The method as claimed in claim 1, wherein inputting the actual environment information in the historical data into a first generator in a pre-trained first generative confrontation network GAN to obtain the virtual environment information output by the first generator, specifically comprises:
scrambling actual environment information in the historical data;
and inputting the scrambled actual environment information into a first generator in a first pre-trained GAN to obtain the virtual environment information output by the first generator.
3. The method of claim 1, wherein pre-training the first GAN comprises:
inputting actual environment information in historical data into a first generator in the first GAN to be trained to obtain virtual environment information output by the first generator to be trained;
inputting the virtual environment information output by the first generator to be trained into a first discriminator in the first GAN to be trained, so that the first discriminator to be trained judges whether the virtual environment information output by the first generator to be trained is actual environment information;
and training the first GAN until the first generator to be trained and the first discriminator to be trained reach Nash balance.
4. The method of claim 1 or 3, wherein pre-training the second GAN specifically comprises:
taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information, and inputting the sample environment information to a second generator in the second GAN to be trained to obtain each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained;
inputting each virtual driving decision corresponding to the sample environmental information and output by the second generator to be trained into a second discriminator in the second GAN to be trained, so that the second discriminator to be trained respectively judges whether each virtual driving decision is matched with the sample environmental information;
and training the second GAN by using the second arbiter to be trained to determine that the probability of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained is the maximum, and determining whether the accuracy of matching each virtual driving decision corresponding to the sample environment information and output by the second generator to be trained with the sample environment information is the maximum by the second arbiter to be trained.
5. The method according to claim 4, wherein the second discriminator is configured to determine, for each virtual driving decision, a degree of matching of the virtual driving decision with the sample environmental information, and determine that each virtual driving decision matches with the sample environmental information when the degree of matching of the virtual driving decision with the sample environmental information is greater than a specified threshold, and determine that each virtual driving decision does not match with the sample environmental information when the degree of matching of the virtual driving decision with the sample environmental information is not greater than the specified threshold;
training the second GAN, specifically comprising:
and in the process of training the second GAN, when a specified condition is met, increasing the specified threshold until the specified threshold reaches a preset threshold.
6. The method of claim 2, wherein the actual environment information or the virtual environment information each comprises: at least one of a road map, a traffic route, a traffic light, a speed of the vehicle, and status information of obstacles in the environment perceived by the vehicle;
the virtual driving decision comprises: and (5) vehicle pose.
7. The method of claim 6, wherein scrambling actual environmental information in the historical data comprises:
and scrambling at least one of traffic signal lamps, the speed of the vehicle and state information of obstacles in the environment sensed by the vehicle in the actual environment information.
8. A training apparatus for a vehicle driving decision model, comprising:
the first input module is used for inputting actual environment information in historical data to a first generator in a pre-trained first generative countermeasure network GAN to obtain virtual environment information output by the first generator;
a first judgment module, configured to input the virtual environment information to a first discriminator in a first GAN, so that the first discriminator judges whether the virtual environment information is actual environment information;
the second input module is used for taking the virtual environment information determined as the actual environment information by the first discriminator as sample environment information and inputting the sample environment information into a second generator in a second GAN trained in advance to obtain each virtual driving decision which is output by the second generator and corresponds to the sample environment information;
a second determining module, configured to input the virtual driving decisions to a second determiner in the second GAN, so that the second determiner determines whether each virtual driving decision matches the sample environment information;
a determination module, configured to take the virtual driving decision determined by the second discriminator to match the sample environmental information as a sample driving decision;
and the training module is used for training a vehicle driving decision model to be trained according to the sample environment information and the sample driving decision matched with the sample environment information.
9. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1-7 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010145000.6A CN110991095B (en) | 2020-03-05 | 2020-03-05 | Training method and device for vehicle driving decision model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010145000.6A CN110991095B (en) | 2020-03-05 | 2020-03-05 | Training method and device for vehicle driving decision model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110991095A true CN110991095A (en) | 2020-04-10 |
| CN110991095B CN110991095B (en) | 2020-07-03 |
Family
ID=70081483
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010145000.6A Active CN110991095B (en) | 2020-03-05 | 2020-03-05 | Training method and device for vehicle driving decision model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110991095B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111795700A (en) * | 2020-06-30 | 2020-10-20 | 浙江大学 | Unmanned vehicle reinforcement learning training environment construction method and training system thereof |
| CN112597217A (en) * | 2021-03-02 | 2021-04-02 | 南栖仙策(南京)科技有限公司 | Intelligent decision platform driven by historical decision data and implementation method thereof |
| CN113077641A (en) * | 2021-03-24 | 2021-07-06 | 中南大学 | Decision mapping method and device for bus on-the-way control and storage medium |
| CN113110526A (en) * | 2021-06-15 | 2021-07-13 | 北京三快在线科技有限公司 | Model training method, unmanned equipment control method and device |
| WO2021232229A1 (en) * | 2020-05-19 | 2021-11-25 | 深圳元戎启行科技有限公司 | Virtual scene generation method and apparatus, computer device and storage medium |
| US20210389776A1 (en) * | 2020-06-12 | 2021-12-16 | Massachusetts Institute Of Technology | Simulation-based training of an autonomous vehicle |
| CN113837272A (en) * | 2021-09-23 | 2021-12-24 | 中汽创智科技有限公司 | Automatic driving long tail data enhancement method |
| GB2598758A (en) * | 2020-09-10 | 2022-03-16 | Toshiba Kk | Task performing agent systems and methods |
| CN114936515A (en) * | 2022-04-25 | 2022-08-23 | 北京宾理信息科技有限公司 | Method and system for generating simulated traffic scene file |
| CN117246345A (en) * | 2023-11-06 | 2023-12-19 | 镁佳(武汉)科技有限公司 | Method, device, equipment and medium for controlling generating type vehicle |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169567A (en) * | 2017-03-30 | 2017-09-15 | 深圳先进技术研究院 | The generation method and device of a kind of decision networks model for Vehicular automatic driving |
| US20170357257A1 (en) * | 2016-06-12 | 2017-12-14 | Baidu Online Network Technology (Beijing) Co., Ltd. | Vehicle control method and apparatus and method and apparatus for acquiring decision-making model |
| CN108944945A (en) * | 2018-07-10 | 2018-12-07 | 深圳地平线机器人科技有限公司 | Trend prediction method, device, electronic equipment and vehicle for assisting driving |
| CN109131348A (en) * | 2018-07-24 | 2019-01-04 | 大连理工大学 | A kind of intelligent vehicle Driving Decision-making method based on production confrontation network |
| CN109816027A (en) * | 2019-01-29 | 2019-05-28 | 北京三快在线科技有限公司 | Training method, device and the unmanned equipment of unmanned decision model |
-
2020
- 2020-03-05 CN CN202010145000.6A patent/CN110991095B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170357257A1 (en) * | 2016-06-12 | 2017-12-14 | Baidu Online Network Technology (Beijing) Co., Ltd. | Vehicle control method and apparatus and method and apparatus for acquiring decision-making model |
| CN107169567A (en) * | 2017-03-30 | 2017-09-15 | 深圳先进技术研究院 | The generation method and device of a kind of decision networks model for Vehicular automatic driving |
| CN108944945A (en) * | 2018-07-10 | 2018-12-07 | 深圳地平线机器人科技有限公司 | Trend prediction method, device, electronic equipment and vehicle for assisting driving |
| CN109131348A (en) * | 2018-07-24 | 2019-01-04 | 大连理工大学 | A kind of intelligent vehicle Driving Decision-making method based on production confrontation network |
| CN109816027A (en) * | 2019-01-29 | 2019-05-28 | 北京三快在线科技有限公司 | Training method, device and the unmanned equipment of unmanned decision model |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021232229A1 (en) * | 2020-05-19 | 2021-11-25 | 深圳元戎启行科技有限公司 | Virtual scene generation method and apparatus, computer device and storage medium |
| US20210389776A1 (en) * | 2020-06-12 | 2021-12-16 | Massachusetts Institute Of Technology | Simulation-based training of an autonomous vehicle |
| CN111795700A (en) * | 2020-06-30 | 2020-10-20 | 浙江大学 | Unmanned vehicle reinforcement learning training environment construction method and training system thereof |
| GB2598758B (en) * | 2020-09-10 | 2023-03-29 | Toshiba Kk | Task performing agent systems and methods |
| GB2598758A (en) * | 2020-09-10 | 2022-03-16 | Toshiba Kk | Task performing agent systems and methods |
| CN112597217A (en) * | 2021-03-02 | 2021-04-02 | 南栖仙策(南京)科技有限公司 | Intelligent decision platform driven by historical decision data and implementation method thereof |
| CN113077641B (en) * | 2021-03-24 | 2022-06-14 | 中南大学 | Decision mapping method and device for bus on-the-way control and storage medium |
| CN113077641A (en) * | 2021-03-24 | 2021-07-06 | 中南大学 | Decision mapping method and device for bus on-the-way control and storage medium |
| CN113110526B (en) * | 2021-06-15 | 2021-09-24 | 北京三快在线科技有限公司 | Model training method, unmanned equipment control method and device |
| CN113110526A (en) * | 2021-06-15 | 2021-07-13 | 北京三快在线科技有限公司 | Model training method, unmanned equipment control method and device |
| CN113837272A (en) * | 2021-09-23 | 2021-12-24 | 中汽创智科技有限公司 | Automatic driving long tail data enhancement method |
| CN113837272B (en) * | 2021-09-23 | 2024-03-26 | 中汽创智科技有限公司 | Automatic driving long tail data enhancement method |
| CN114936515A (en) * | 2022-04-25 | 2022-08-23 | 北京宾理信息科技有限公司 | Method and system for generating simulated traffic scene file |
| CN114936515B (en) * | 2022-04-25 | 2023-09-19 | 北京宾理信息科技有限公司 | Method and system for generating simulated traffic scene file |
| CN117246345A (en) * | 2023-11-06 | 2023-12-19 | 镁佳(武汉)科技有限公司 | Method, device, equipment and medium for controlling generating type vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110991095B (en) | 2020-07-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110991095B (en) | Training method and device for vehicle driving decision model | |
| CN110929431B (en) | Training method and device for vehicle driving decision model | |
| CN111079721B (en) | Method and device for predicting track of obstacle | |
| CN112801229B (en) | Training method and device for recognition model | |
| CN111208838B (en) | Control method and device of unmanned equipment | |
| CN110989636A (en) | Method and device for predicting track of obstacle | |
| KR20180034395A (en) | Transfer learning in neural networks | |
| KR20180028442A (en) | Context-based fryers for object detection in images | |
| CN111912423B (en) | Method and device for predicting obstacle trajectory and training model | |
| US12005922B2 (en) | Toward simulation of driver behavior in driving automation | |
| CN111238523A (en) | Method and device for predicting motion trail | |
| CN111522245B (en) | Method and device for controlling unmanned equipment | |
| CN113110526B (en) | Model training method, unmanned equipment control method and device | |
| CN111126362A (en) | Method and device for predicting obstacle track | |
| CN111007858A (en) | Training method of vehicle driving decision model, and driving decision determining method and device | |
| CN113419547A (en) | Multi-vehicle cooperative control method and device | |
| CN112949756B (en) | Method and device for model training and trajectory planning | |
| KR102260246B1 (en) | Mehod and Device of configuring deep learning algorithm for autonomous driving | |
| CN113033527A (en) | Scene recognition method and device, storage medium and unmanned equipment | |
| CN110895406B (en) | Method and device for testing unmanned equipment based on interferent track planning | |
| CN112859883B (en) | Control method and control device of unmanned equipment | |
| WO2023172835A1 (en) | Single stream multi-level alignment for vision-language pretraining | |
| Zhang et al. | CLAIRE: enabling continual learning for real-time autonomous driving with a dual-head architecture | |
| CN114153207A (en) | Control method and control device of unmanned equipment | |
| CN112849156A (en) | Driving risk identification method and device |
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 |