CN109697875B - Method and device for planning driving track - Google Patents

Abstract

The present application relates to artificial intelligence, and discloses a method and device for planning a driving trajectory, belonging to the fields of data processing technology, automatic driving, and the Internet of Things. The method includes: the server determines a target neural network model corresponding to the road section where the target position of the first vehicle is currently located, and sends the target neural network model to the first vehicle, so that the first vehicle passes through the target neural network model Determine the driving path. Since the target neural network model is obtained by the server training according to the travel trajectories of multiple second vehicles, the multiple second vehicles refer to vehicles passing through the road section where the current target position of the first vehicle is located before the current time. That is, when the first vehicle determines its own driving trajectory at the target position through the target neural network model, it refers to the driving trajectories of a plurality of second vehicles passing through the road section where the target position is located before the current time, which improves the determination of the driving trajectory. the feasibility of the driving trajectory.

Description

Method and device for planning driving trajectory

technical field

The present application relates to the fields of artificial intelligence, data processing technology, automatic driving, and the Internet of Things, and in particular, to a method and device for planning a driving trajectory.

Background technique

Automatic driving, that is, in the process of vehicle driving, the vehicle itself plans the driving trajectory, and drives according to the planned driving trajectory. However, in practical applications, if the planned driving trajectory of the vehicle is not suitable, it will easily affect the safety of other vehicles and itself. Therefore, how to plan the driving trajectory is particularly important.

In the related art, when planning a driving trajectory, the vehicle determines the driving task to be performed and the lane position on the current road, and plans the driving trajectory according to the determined driving task and the lane position on the current road. Among them, the driving tasks include going straight, turning left, turning right, and turning around, and the lane positions on the current road include going straight, turning left, and turning right. For example, if the determined driving task is a right turn, and the current lane position on the current road is the straight lane, the planned driving trajectory at this time can be: starting from the center line of the current lane, along the center of the current lane Drive on the connecting line between the line and the centerline of the right lane until you reach the centerline of the right lane.

However, when the vehicle travels according to the travel trajectory determined by the above method, accidents are likely to occur, that is, the accident rate is high when the vehicle travels according to the travel trajectory determined by the above method. low sex.

SUMMARY OF THE INVENTION

In order to solve the problem of low feasibility of the planned travel trajectory in the related art, the present application provides a method and device for planning a travel trajectory. The technical solution is as follows:

In a first aspect, a method for planning a driving trajectory is provided, applied to a first vehicle, and the method includes:

Receive the target neural network model sent by the server, where the target neural network model refers to the neural network model corresponding to the road section where the target position of the first vehicle is currently located, and the target neural network model is the obtained by training the driving trajectories of a plurality of second vehicles, where the plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time;

According to the driving task and driving information of the first vehicle, and the driving information of the first obstacle vehicle, determine the driving trajectory of the first vehicle through the target neural network model;

Wherein, the driving task includes going straight, turning left, turning right, and making a U-turn, the driving information includes position information of the current location, driving direction and driving speed, and the first obstacle vehicle is the same as the first vehicle. The distance between the vehicles is less than the preset distance threshold.

In this application, since the target neural network model is obtained by the server training according to the driving trajectories of a plurality of second vehicles, and the plurality of second vehicles refer to the road segment where the target position where the first vehicle is currently located before the current time Vehicles. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

Optionally, the road section where the target position is located is an intersection;

According to the driving task and driving information of the first vehicle, and the driving information of the first obstacle vehicle, the driving trajectory of the first vehicle is determined by the target neural network model, including:

Taking the driving task and driving information of the first vehicle, the driving information of the first obstacle vehicle and the signal light state corresponding to the target position at the intersection as the input of the target neural network model, through the target The neural network model determines the travel trajectory of the first vehicle.

Specifically, when the road section where the target position is located is an intersection, when determining the driving trajectory of the first vehicle through the target neural network model, the signal light state corresponding to the target position at the intersection also needs to be considered, so as to further improve the determination of the target position. the feasibility of the driving trajectory.

Optionally, the target neural network model refers to a neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle, and the plurality of second vehicles refers to the number of second vehicles that have passed before the current time. The road section where the target position is located and the vehicle having the same driving task as the driving task of the first vehicle.

Further, at a certain road section, the server can train different neural network models for different driving tasks. At this time, the first vehicle can use the target neural network corresponding to both the road section where the target position is located and the driving task of the first vehicle. The network model determines the driving trajectory.

In a second aspect, another method for planning a driving trajectory is provided, which is applied to a server, and the method includes:

A target neural network model corresponding to the road section where the target position of the first vehicle is currently located is determined from the stored neural network model, and the target neural network model is obtained by training according to the driving trajectories of a plurality of second vehicles. The plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time;

The target neural network model is sent to the first vehicle, so that the first vehicle determines the driving trajectory of the first vehicle through the target neural network model.

In this application, when the first vehicle travels to the target position, the server can directly determine the target neural network model corresponding to the road section where the target position is located, and send the target neural network model to the first vehicle, so that the first vehicle The driving trajectory is determined by the target neural network model. Because the target neural network model is obtained by the server training according to the travel trajectories of a plurality of second vehicles, and the plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time. That is, when the first vehicle determines its own travel trajectory at the target position through the target neural network model, it refers to the travel trajectories of multiple second vehicles passing through the road section where the target position is located before the current time, so as to reduce the speed of the first vehicle. The accident occurrence rate when driving according to the determined driving trajectory, that is, the feasibility of the determined driving trajectory is improved.

Optionally, before determining the target neural network model corresponding to the road section where the target position of the first vehicle is currently located from the stored neural network model, the method further includes:

determining the driving trajectories of all the second vehicles passing through the road section where the target position is located within the preset time period and the score of the driving trajectories of each second vehicle;

Selecting N driving trajectories with a score greater than a preset score from all the acquired driving trajectories, where N is greater than 1 and less than or equal to the total number of the acquired driving trajectories;

The initialized neural network model is trained through the N driving trajectories to obtain the target neural network model.

Since the driving trajectory of the first vehicle is determined according to the target neural network model sent by the server, in this application, the server also needs to predetermine the target neural network model. Further, in order to determine the feasibility of the driving trajectory determined by the target neural network model, the server can select an excellent driving trajectory from a plurality of driving trajectories according to the scores of each driving trajectory, and train the selected excellent driving trajectory through the selected excellent driving trajectory. target neural network model.

Optionally, the initialized neural network model is trained through the N driving trajectories to obtain the target neural network model, including:

determining the driving tasks and driving information of the N second vehicles, and the driving information of the N second obstacle vehicles;

Wherein, the N second vehicles are vehicles corresponding to the N driving trajectories, the N second obstacle vehicles are in one-to-one correspondence with the N second vehicles, and the second obstacle vehicles are vehicles corresponding to the corresponding No. The distance between the two vehicles is less than the preset distance threshold value, the driving task includes going straight, turning left, turning right and U-turn, and the driving information includes the current position, driving direction and driving speed;

According to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the driving trajectories of the N second vehicles, the initialized neural network model is trained to obtain the target neural network model.

The process of training the target neural network model, that is, determining sample data, and then training the initialized neural network model through the determined sample data to obtain the target neural network model.

Optionally, the road section where the target position is located is an intersection;

After determining the driving tasks and driving information of the N second vehicles and the driving information of the N second obstacle vehicles, the method further includes:

Determine N signal light states, the N signal light states are in one-to-one correspondence with the N second vehicles, and each signal light state refers to the signal light state corresponding to the intersection when the corresponding second vehicle passes through the intersection;

Correspondingly, the initialized neural network model is performed according to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the driving trajectories of the N second vehicles. Training to obtain the target neural network model, including:

The driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the N signal light states are used as the input of the initialized neural network model, and the Nth The driving trajectory of the second vehicle is used as the output of the initialized neural network model, and the initialized neural network model is trained to obtain the target neural network model.

Further, when the road section where the target position is located is an intersection, the server determines the sample data for training the target neural network model at this time, and the sample data may also include the signal light state corresponding to the intersection when each second vehicle passes through the intersection. .

Optionally, the determining the score of the driving trajectory of each second vehicle includes:

For any second vehicle among all the second vehicles, determine the driving condition of the second vehicle during the driving process according to the driving track of the second vehicle, where the driving condition includes the number of collisions and whether the traffic rules are complied with. , the number of lane changes, the driving time and whether it is a smooth driving;

The score of the driving trajectory of the second vehicle is determined according to the driving condition of the second vehicle during the driving process.

Wherein, the server may determine the score of the driving trajectory of the second vehicle according to factors such as the number of collisions when the second vehicle passes through the intersection, whether it complies with traffic rules, the number of lane changes, the driving time, and whether it is a smooth driving.

Optionally, the determining of the driving condition of the second vehicle during the driving process according to the driving trajectory of the second vehicle includes:

determining a travel trajectory of a third obstacle vehicle passing through the intersection, where the third obstacle vehicle is a vehicle whose distance from the second vehicle is less than a preset distance threshold when the second vehicle passes through the intersection;

determining the number of collisions between the second vehicle and the third obstacle vehicle according to the travel trajectory of the second vehicle and the third obstacle vehicle;

determining a signal light state corresponding to the intersection during the process of the second vehicle passing through the intersection;

determining whether the second vehicle complies with the traffic rules according to the driving track of the second vehicle and the state of the signal light at the intersection when the second vehicle passes through the intersection;

According to the driving trajectory of the second vehicle, the number of lane changes, the driving time and whether the second vehicle is driving smoothly is determined.

Specifically, the server may use the above method to determine the number of collisions, whether the second vehicle complies with traffic rules, the number of lane changes, the driving time, and whether it is a smooth driving when the second vehicle passes through the intersection.

Optionally, the determining the score of the driving trajectory of the second vehicle according to the driving condition of the second vehicle during driving includes:

If the number of collisions of the second vehicle during driving is greater than or equal to the preset number of collisions, the collision score is determined to be the first score, otherwise, the collision score is determined to be the second score, where the collision score is related to the collision occurrence The number of times is negatively correlated;

If the second vehicle complies with the traffic rules, determine the traffic rule score as the third score, otherwise, determine the traffic rule score as the fourth score;

If the number of lane changes of the second vehicle during driving is greater than or equal to the minimum number of lane changes required for the second vehicle to pass through the intersection, determine the lane change score as the fifth score, otherwise, determine the lane change score The lane score is the sixth score, and the lane change score is negatively correlated with the number of lane changes;

If the driving duration of the second vehicle through the intersection is greater than or equal to the preset driving duration, the duration score is determined to be the seventh score, otherwise, the duration score is determined to be the eighth score, wherein the duration score and the driving duration are in the same negative correlation;

If the second vehicle drives smoothly, determining the driving score as the ninth score, otherwise, determining the driving score as the tenth score;

A sum of the collision score, the traffic rule score, the lane change score, the duration score and the driving score is determined as a score of the driving trajectory of the second vehicle.

Further, when the server determines the number of collisions, whether the second vehicle complies with the traffic rules, the number of lane changes, the driving time and whether it is a smooth driving when the second vehicle passes through the intersection, the collision score and the traffic rule score can be determined respectively. , a lane change score, a duration score, and a driving score to determine a score of the driving trajectory of the second vehicle according to the determined collision score, traffic rule score, lane change score, duration score, and driving score.

Optionally, determining the target neural network model corresponding to the road section where the target position of the first vehicle is currently located from the stored neural network model includes:

According to the position information of the target position and the driving task of the first vehicle, a target neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle is determined from the stored neural network model ;

Correspondingly, the plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle.

Further, at a certain road section, the server can train different neural network models for different driving tasks, and at this time, the server can send to the first vehicle the road section where the target position is located and the driving task of the first vehicle. target neural network model.

Optionally, the determining of travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period includes:

determining a second vehicle passing through the road section where the target position is located at a first moment, where the first moment refers to any moment within the preset time period;

For any determined second vehicle, determining a plurality of second moments during the driving process of the second vehicle passing through the road section where the target position is located;

The driving trajectory of the second vehicle is determined based on the driving information of the second vehicle at each second time.

Wherein, the server determines the driving trajectory of the second vehicle, that is, determines the driving information of the second vehicle at each second moment in the process of passing through the road section where the target position is located.

A third aspect provides an apparatus for planning a driving trajectory, which is applied to a first vehicle, and the apparatus for planning a driving trajectory has the function of implementing the method for planning a driving trajectory in the first aspect. The apparatus for planning a driving trajectory includes at least one unit, and the at least one unit is configured to implement the method for planning a driving trajectory provided in the first aspect.

In a fourth aspect, another apparatus for planning a driving trajectory is provided, which is applied to a server, and the apparatus for planning a driving trajectory has the function of implementing the method for planning a driving trajectory in the second aspect. The apparatus for planning a driving trajectory includes at least one unit, and the at least one unit is configured to implement the method for planning a driving trajectory provided in the second aspect above.

In a fifth aspect, a device for planning a driving trajectory is provided. The structure of the device for planning a driving trajectory includes a processor and a memory, and the memory is used for storing and supporting the beamforming device to perform the planning provided in the first aspect. A program of the method for driving a trajectory, and storing data involved in implementing the method for planning a driving trajectory provided in the first aspect. The processor is configured to execute programs stored in the memory. The operating means of the storage device may further include a communication bus for establishing a connection between the processor and the memory.

In a sixth aspect, another device for planning a driving trajectory is provided. The structure of the device for planning a driving trajectory includes a processor and a memory, and the memory is used for storing and supporting the beamforming device to perform the above-mentioned second aspect. A program of a method for planning a driving trajectory, and storing data involved in implementing the method for planning a driving trajectory provided by the second aspect. The processor is configured to execute programs stored in the memory. The operating means of the storage device may further include a communication bus for establishing a connection between the processor and the memory.

In a seventh aspect, a computer-readable storage medium is provided, where instructions are stored in the computer-readable storage medium, when the computer-readable storage medium runs on a computer, the computer executes the method for planning a driving trajectory described in the first aspect.

In an eighth aspect, another computer-readable storage medium is provided, where instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium runs on a computer, the computer executes the method for planning a travel trajectory described in the second aspect above. .

In a ninth aspect, there is provided a computer program product containing instructions, which, when executed on a computer, cause the computer to execute the method for planning a driving trajectory described in the first aspect above.

In a tenth aspect, there is provided another computer program product containing instructions, which, when executed on a computer, cause the computer to execute the method for planning a travel trajectory described in the second aspect above.

The technical effects obtained by the third aspect, the fifth aspect, the seventh aspect and the ninth aspect are similar to the technical effects obtained by the corresponding technical means in the first aspect, and will not be repeated here. The technical effects obtained by the fourth aspect, the sixth aspect, the eighth aspect and the tenth aspect are similar to the technical effects obtained by the corresponding technical means in the second aspect, and will not be repeated here.

The beneficial effects brought by the technical solution provided by the application are:

In this application, when the first vehicle is currently at the target position, the server may directly determine the target neural network model corresponding to the road section where the target position is located, and send the target neural network model to the first vehicle, so that the first vehicle The driving trajectory is determined by the target neural network model. Because the target neural network model is obtained by the server training according to the travel trajectories of a plurality of second vehicles, and the plurality of second vehicles refer to vehicles passing through the road section where the current target position of the first vehicle is located before the current time. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

Description of drawings

1 is a schematic diagram of a system for planning a driving trajectory according to an embodiment of the present invention;

2 is a schematic diagram of a server according to an embodiment of the present invention;

3 is a schematic structural diagram of a server provided by an embodiment of the present invention;

4 is a schematic diagram of a vehicle provided by an embodiment of the present invention;

5 is a schematic structural diagram of a vehicle according to an embodiment of the present invention;

6A is a flowchart of a method for planning a driving trajectory provided by an embodiment of the present invention;

6B is a schematic diagram of an intersection provided by an embodiment of the present invention;

7 is a flowchart of another method for planning a driving trajectory provided by an embodiment of the present invention;

8 is a block diagram of a device for planning a driving trajectory provided by an embodiment of the present invention;

9A is a block diagram of a device for planning a driving trajectory provided by an embodiment of the present invention;

FIG. 9B is a block diagram of another apparatus for planning a driving trajectory provided by an embodiment of the present invention.

Detailed ways

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

For ease of understanding, firstly, the application scenarios involved in the embodiments of the present invention are briefly introduced.

For self-driving vehicles, the vehicle drives according to the pre-planned driving trajectory, rather than driving according to the instructions issued by the driver in real time like a driver-driven vehicle, so that the self-driving vehicle cannot be driven by a driver. The current environment of the vehicle makes adaptive adjustments to the driving direction or driving speed. For example, the current driving task of the vehicle is to turn right, and the position of the lane on the current road is the straight lane. At this time, the planned driving trajectory is: starting from the center line of the current lane, along the center of the current lane. Drive on the connecting line between the line and the centerline of the right lane until you reach the centerline of the right lane. In fact, when the vehicle is turning right, there may be other vehicles around the vehicle changing lanes. At this time, if you continue to drive according to the above planned driving trajectory, it will easily affect the safety of other vehicles and your own. Therefore, how to plan the driving trajectory is particularly important. However, the embodiment of the present invention is applied to the scenario of how to plan the driving trajectory of the vehicle.

After the application scenarios of the embodiments of the present invention are introduced, the following briefly introduces the system for planning a driving trajectory provided by the embodiments of the present invention.

As shown in FIG. 1 , an embodiment of the present invention provides a system for planning a driving trajectory. The system 100 for planning a driving trajectory includes a server 101 and at least one vehicle 102 , and the server 101 and each vehicle 102 are connected wirelessly. to communicate.

Wherein, the server 101 is used to perform statistics on the historical driving trajectories, so as to train the neural network model through the historical driving trajectories. The vehicle 102 is used to obtain the pre-trained neural network model from the server 101, and plan a driving trajectory through the pre-trained neural network.

Optionally, the server 101 is further configured to push the pre-trained neural network model to at least one vehicle 102 by broadcasting, and the vehicle 102 does not need to actively acquire the pre-trained neural network model from the server 101 at this time.

FIG. 2 is a schematic diagram of a server 200 according to an embodiment of the present invention. 2 , the server 200 includes a data acquisition module 201 , a data storage module 202 , a task analysis module 203 , a time trajectory analysis module 204 , a signal light analysis module 205 , a trajectory scoring module 206 , a model training module 207 and a model push module 208 .

The data collection module 201 is used to collect data, and the collected data may be data reported by multiple fixed cameras installed on a preset road section, or data reported to the server by vehicles passing through the preset road section. If the collected data is data reported by multiple cameras, the collected data should at least include information such as the collected video, the time corresponding to the video, the geographic location of the camera, the height of the camera, and the orientation of the camera. If the collected data is the data reported by a vehicle passing through the preset road section, the collected data should at least include: the position, driving direction, speed and corresponding time information of the vehicle. Optionally, the collected data may also include a video shot by an on-board camera installed on the vehicle, or information after processing the shot video by on-board software installed on the vehicle. Among them, the video captured by the vehicle-mounted camera may include signal light information, as well as information such as the position, driving direction and speed of surrounding dynamic obstacles.

The data storage module 202 is used for storing the data collected by the data collection module 201 .

The task analysis module 203 is used to read the stored data from the data storage module 202, and determine the driving trajectory of each vehicle through the preset video processing technology, and analyze the vehicle in the preset road section according to the driving trajectory of each vehicle. driving task. For example, turn left, go straight, turn right, or make a U-turn.

The time trajectory analysis module 204 reads the stored data from the data storage module 202, and determines the driving trajectory of each vehicle through the preset video processing technology. Information such as position, driving direction, and vehicle speed, that is, the driving trajectory of the vehicle is composed of information such as the position, driving direction, and vehicle speed of the vehicle when the vehicle is driving on the preset road section at different times.

The signal light analysis module 205 reads the stored data from the data storage module 202 , and similarly, analyzes the time-varying information of the signal lights of the preset road section through a preset video processing technology. For example, it is analyzed that in the time period t0 to t1, the straight traffic light in the west to east direction is green.

The trajectory scoring module 206 is used to read the data processed by the above-mentioned task analysis module 203, time trajectory analysis module 204 and signal light analysis module 205, correlate these data with each other according to time, and regard different vehicles within the preset distance threshold as each other. It is an obstacle vehicle to score all driving trajectories, and the scoring criteria may include: whether the traffic rules are obeyed, the time to pass the preset road section, the number of lane changes, the number of collisions with the obstacle vehicle, and the like.

The model training module 207 uses the initialized neural network model to train the historical driving trajectories according to the scoring data of the above-mentioned task analysis module 203, time trajectory analysis module 204, signal light analysis module 205 and trajectory scoring module, so that the neural network model can learn more efficiently. The characteristics of the high-rated driving trajectories are obtained, and the neural network model is obtained, and the trained neural network model is stored.

The model pushing module 208 is used for pushing the stored neural network model to the vehicle when receiving the model request of the vehicle, and the model pushing module 208 is also used for pushing the neural network model to vehicles within a certain range in a broadcast manner.

FIG. 3 is a schematic structural diagram of another server 300 provided by an embodiment of the present invention. The server 300 is used to implement the functions of each module included in the server 200 shown in FIG. 2 . Specifically, as shown in FIG. 3 , the server 300 includes at least one processor 301 , a communication bus 302 , a memory 303 and at least one communication interface 304 .

The processor 301 can be a general-purpose central processing unit (Central Processing Unit, CPU), a microprocessor, an application-specific integrated circuit (application-specific integrated circuit, ASIC), or one or more processors for controlling the execution of the programs of the present application. integrated circuit.

Communication bus 302 may include a path to communicate information between the above-described components.

Memory 303 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, random access memory (RAM)) or other type of static storage device that can store information and instructions type of dynamic storage device, it can also be Electrically Erasable Programmable Read-Only Memory (EEPROM), CompactDisc Read-Only Memory (CD-ROM) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or capable of carrying or storing desired program code in the form of instructions or data structures and capable of being executed by a computer Access any other medium without limitation. The memory 303 can exist independently and is connected to the processor 301 through the communication bus 302 . The memory 303 may also be integrated with the processor 301 .

The communication interface 304, using any device such as a transceiver, is used to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Networks (WLAN) and the like.

In a specific implementation, as an embodiment, the processor 301 may include one or more CPUs.

In a specific implementation, as an embodiment, the server may include multiple processors. Each of these processors can be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and/or processing cores for processing data (eg, computer program instructions). For example, for each module in the task analysis module 203 , the time trajectory analysis module 204 , the signal light analysis module 205 , the trajectory scoring module 206 and the model training module 207 shown in FIG. 2 , the module can be implemented by a processor. function of this module.

In a specific implementation, as an embodiment, the server 300 may further include an output device and an input device. The output device communicates with the processor 301 and can display information in a variety of ways. For example, the output device may be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, or a projector (projector), etc. . The input device communicates with the processor 301 and can receive user input in a variety of ways. For example, the input device may be a mouse, a keyboard, a touch screen device, or a sensor device, or the like.

In a specific implementation, the server may be a desktop computer, a portable computer, a network server, a PDA (Personal Digital Assistant, PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiment of the present invention does not limit the type of the server.

Wherein, the memory 303 is used for storing the program code for executing the solution of the present application, and is controlled and executed by the processor 301, so as to realize the training of the neural network model according to the historical driving trajectory. The processor 301 is used to execute program codes stored in the memory 303 . One or more software modules may be included in the program code.

4 is a schematic diagram of a vehicle 400 according to an embodiment of the present invention. The vehicle 400 includes a positioning module 401, a model request module 402, an in-vehicle communication module 403, a mission planning module 404, a perception module 405, a trajectory calculation module 406, and a vehicle control module Module 407.

The positioning module 401 collects the current position information of the vehicle in real time, and sends the collected position information to the model request module 402 at a certain frequency.

The model request module 402 is used to request the neural network model from the server through the in-vehicle communication module 403, that is, the model request module 402 is used to send a model acquisition request to the server through the in-vehicle communication module 403, and the model acquisition request includes the current target of the vehicle. Location. When the model request module 402 receives the neural network model returned by the server through the in-vehicle communication module 403, the neural network model is stored.

Optionally, when the model push module 208 in the above-mentioned server shown in FIG. 2 pushes the neural network model to vehicles within a certain range by broadcasting, the model request module 402 is used to directly receive the server push through the in-vehicle communication module 403. the neural network model and store the neural network model.

In addition, since different road sections correspond to different neural network models, the model request module 402 is further configured to check whether the local area has a neural network model corresponding to the current driving road section according to the collected position information, and when it is found that the local neural network model does not exist , then obtain the neural network model through the server.

The mission planning module 404 is used to determine the driving mission of the vehicle. The perception module 405 is used to determine information such as the driving information of the vehicle, the driving information of the obstacle vehicle, and the status of the signal lights. The driving information includes information such as the current location, driving direction, and driving speed.

The trajectory calculation module 406 is configured to determine the driving trajectory of the vehicle through the neural network model obtained by the model request module 402 according to the data determined by the task planning module 404 and the perception module 405 . The vehicle control module 407 is configured to control the vehicle to travel according to the travel trajectory determined by the trajectory calculation module 406 .

FIG. 5 is a schematic structural diagram of another vehicle 500 according to an embodiment of the present invention, where the vehicle 500 is used to implement the functions of each module included in the vehicle 400 shown in FIG. 4 . Specifically, as shown in FIG. 5 , the vehicle 500 includes at least one processor 501 , a communication bus 502 , a memory 503 and at least one communication interface 504 .

The structure and functions of the processor 501 and the processor 301 shown in FIG. 3 are basically the same, the structure and functions of the communication bus 502 and the communication bus 502 shown in FIG. 3 are basically the same, and the memory 503 and the memory 503 shown in FIG. 3 are basically the same. The structures and functions of the communication interface 504 are basically the same as those of the communication interface 504 shown in FIG. 3 , and are not described in detail here. ,

The difference is that the memory 503 is used to store the program code for executing the solution of the present application, and is controlled and executed by the processor 501, so as to realize the planning of the driving trajectory for the vehicle according to the trained neural network model.

The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

6A is a flowchart of a method for planning a driving trajectory provided by an embodiment of the present invention. The method is applied to the system shown in FIG. 1 . For the convenience of subsequent description, the vehicle that currently needs to plan a driving trajectory is referred to as the first vehicle. As shown in Figure 6A, the method includes the following steps:

Step 601: The server determines, from the stored neural network model, a target neural network model corresponding to the road section where the target position of the first vehicle is currently located. Yes, the plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time.

In practical applications, the geographical situation of different road sections is very different, for example, the road leveling degree and the fixed obstacles on the road may be different for different road sections. Therefore, in this embodiment of the present invention, the server can be trained for different road sections in advance. Different neural network models. Since different road sections correspond to different neural network models, when the first vehicle is currently at the target position, the server may determine the target neural network model corresponding to the road section where the target position is located.

In particular, since the geographical situation at the intersection is relatively complex, the accident rate of the first vehicle passing through the intersection is relatively high, while the geographical situation of other road sections not at the intersection is relatively simple. Therefore, in this embodiment of the present invention, the server Different neural network models can be trained just for different intersections. That is, when the road section where the target position is located is an intersection, the target neural network model is the neural network model corresponding to the intersection at this time.

It should be noted that, in this embodiment of the present invention, the intersection may be all branch intersections at the intersection of two different roads, or may be any branch intersection of all branch intersections at the intersection of the two different roads. For example, as shown in FIG. 6B , road 1 and road 2 form an intersection at the intersection, and the intersection includes four different branch intersections in four directions of south, east, and northwest respectively.

Wherein, when the intersection in the embodiment of the present invention is the intersection, the server determines the neural network model corresponding to the intersection according to the historical driving trajectories of all branch intersections included in the intersection. When the intersection in the embodiment of the present invention can be a branch intersection in any direction of the intersection, at this time, the server determines the neural network model corresponding to the branch intersection according to the historical driving trajectory passing through the branch intersection, that is, At this intersection, 4 different neural network models can be trained for each branched intersection that faces upwards.

In addition, for the same road segment, the difference between the corresponding driving trajectories of different driving tasks at the road segment is relatively large. In the process of training the neural network, it is necessary to analyze a class of training samples with a certain uniform rule. Therefore, in this embodiment of the present invention, in order to facilitate the neural network model to learn the rules of the data in the training samples, for For the same road section, different neural network models can be trained according to the driving task. That is, at this road segment, a neural network model is trained for each driving task.

At this time, when the first vehicle needs to determine the driving trajectory, the server determines the target neural network model corresponding to both the driving task of the first vehicle and the road section where the target position is located. Correspondingly, when the server trains the target neural network model according to the driving trajectories of a plurality of second vehicles, the plurality of second vehicles refer to the road section where the target position is located before the current time and the driving task is the same as the driving of the first vehicle. Vehicles with the same mission.

The implementation manner of the server obtaining the target neural network model by training according to the driving trajectories of a plurality of second vehicles will be described in detail in the following embodiments, and will not be described here.

Step 602: The server sends the target neural network model to the first vehicle.

After determining the target neural network model in step 601, the server sends the target neural network model to the first vehicle, so that the first vehicle can determine the driving of the first vehicle according to the target neural network model through the following steps 603 and 604. trajectory.

Step 603: The first vehicle receives the target neural network model sent by the server.

When the first vehicle receives the target neural network model sent by the server, the following step 604 can be used to determine the driving trajectory of the first vehicle.

Step 604: The first vehicle determines the driving trajectory of the first vehicle through the target neural network model according to the driving task and driving information of the first vehicle and the driving information of the first obstacle vehicle, wherein the driving tasks include going straight, turning left, Right turn and U-turn, the travel information includes current location, travel direction and travel speed, and the first obstacle vehicle is a vehicle whose distance from the first vehicle is less than a preset distance threshold.

In the embodiment of the present invention, the server may train the initialized neural network model according to the travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period, and obtain the target corresponding to the road section where the target position is located A neural network model, so that the first vehicle can determine the driving trajectory according to the target neural network model. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

FIG. 7 is a flowchart of another method for planning a driving trajectory provided by an embodiment of the present invention, and the method is applied to the system shown in FIG. 1 . The embodiment shown in FIG. 7 is a further detailed description of the embodiment shown in FIG. 6A . As shown in FIG. 7 , the method includes the following steps:

Step 701: The server determines the travel trajectories of all the second vehicles passing through the road section where the target position is located within a preset time period, where the target position is the current position of the first vehicle.

In this embodiment of the present invention, since the target neural network model is obtained by the server through training according to the driving trajectories of multiple second vehicles, before the server sends the target neural network model to the first vehicle, the The driving trajectory of the vehicle is trained to determine the target neural network model. Specifically, the target neural network model can be determined through steps 701 to 705 .

In addition, in order to facilitate the neural network model to learn the law of the data in the training samples, for the same road section, different neural network models can be trained separately according to the driving task. Since the difference between training different neural network models lies in the different training samples, in this embodiment of the present invention, the training of the neural network model corresponding to the road section where the target position is located is taken as an example for description. The training process is no longer detailed.

It is worth noting that the key to training the neural network is to determine the training samples. When training the neural network model corresponding to the road section where the target position is located, the training sample needs to be determined according to the driving trajectory of the road section where the target position is located before the current time. .

In addition, since there may be a large number of driving trajectories passing through the road section where the target position is located before the current time, and the geographical situation of the road section where the target position is located before the current time may also change, that is, the situation that occurs when the distance from the current time is long. The data of the driving trajectories may have become invalid, so it is only necessary to train the neural network model according to the historical driving trajectories within a preset time period.

For example, the preset time period may be one month before the current time, and in this case, the neural network model only needs to be determined according to the driving trajectory of the second vehicle passing through the road section where the target position is located within one month before the current time. .

Specifically, determine the second vehicle that passes through the road section where the target position is located at the first moment, where the first moment refers to any moment within the preset time period, and for any second vehicle that is determined to be obtained, determine the second vehicle. The driving track of the second vehicle is determined based on the driving information of the second vehicle at each of the second moments during the driving process of the two vehicles passing through the road section where the target position is located.

In a possible implementation manner, in order to determine all the driving trajectories passing through the road section where the target position is located within the preset time period, the preset time period may be divided into a plurality of first moments, and each The second vehicles passing through the road section where the target position is located at the first moment, all the second vehicles passing through the road section where the target position is located at the first moment are all vehicles passing through the road section where the target position is located within the preset time period.

Wherein, the driving track of each second vehicle is composed of driving information at each second moment when the second vehicle passes through the intersection, and the driving information includes the current position of the second vehicle at the second moment, and the driving information. information such as direction and driving speed.

Specifically, the video collected by the data acquisition module at the road section where the target position is located within a preset time period is determined by the data storage module shown in FIG. 2, and each second is determined by the time trajectory analysis shown in FIG. 2. The trajectory of the vehicle.

In the embodiment of the present invention, in order to obtain a feasible driving trajectory through the neural network model, the neural network model should be trained by using a relatively excellent driving trajectory in the historical driving trajectory. That is, after determining the driving trajectories of all the second vehicles passing through the road section where the target position is located within the preset time period, it is also necessary to determine the excellent driving trajectories among them.

Wherein, when the road section where the target position is located is an intersection, the server can specifically determine the score of each driving track through the following steps 702 and 703, so that the following step 704 can be used to determine the best among the driving tracks according to the score of each driving track. 's driving trajectory.

Optionally, when the road segment where the target position is located is another type of road segment, the following steps 702 to 704 may also be referred to to determine the excellent driving trajectory, which will not be described in detail here.

Step 702: For any second vehicle among all the second vehicles, the server determines the driving condition of the second vehicle during the driving process according to the driving track of the second vehicle, and the driving condition includes the number of collisions and whether the traffic is obeyed. The rules, the number of lane changes, the duration of the drive, and whether it was a smooth ride.

Wherein, determining the number of collisions may specifically include: determining a driving trajectory of a third obstacle vehicle when passing through the intersection, where the third obstacle vehicle is a vehicle whose distance from the second vehicle is less than a preset distance threshold. The number of collisions between the second vehicle and the third obstacle vehicle is determined according to the travel trajectory of the second vehicle and the travel trajectory of the third obstacle vehicle.

Since the driving track of the second vehicle is composed of the driving information of the second vehicle at each second time when the second vehicle passes the intersection, the driving track of the third obstacle vehicle is also composed of the driving information of the third obstacle vehicle at each second time. The composition, that is, the travel trajectory of the second vehicle and the time instants in the travel trajectory of the third obstacle vehicle are in one-to-one correspondence. At this time, an implementation manner of determining the number of collisions between the second vehicle and the third obstacle vehicle may be: for each second moment, determining the current position of the second vehicle and the current position of the third obstacle vehicle If the distance between the current position of the second vehicle and the current position of the third obstacle vehicle is less than the preset collision distance, it is determined that at the second moment, the distance between the second vehicle and the third obstacle vehicle is collision occurs.

The preset collision distance is a preset distance, and the preset collision distance may be 0.1m, 0.15m, or 0.2m.

Secondly, determining whether the second vehicle complies with the traffic rules may specifically include: determining the state of the signal light corresponding to the intersection during the process of the second vehicle passing through the intersection, and according to the driving track of the second vehicle and the process of the second vehicle passing through the intersection It is determined whether the second vehicle complies with the traffic rules according to the state of the corresponding signal light at the intersection.

That is, in the process of the second vehicle traveling at each second time, the state of the signal light corresponding to the intersection at each second time is determined, and for each second time, at the second time, if it is the same as the intersection If the state of the corresponding signal light is red, it means that the second vehicle is not complying with the traffic rules. If the state of the signal light corresponding to the intersection is green, it means that the second vehicle is complying with the traffic rules.

In addition, according to the driving track of the second vehicle, determine the number of lane changes, the driving time and whether the second vehicle is driving smoothly.

Wherein, in order for the driving duration to accurately describe the duration of the second vehicle passing through the intersection, determining the driving duration may specifically include: determining the time for the second vehicle to pass through the intersection, and determining that the signal light is red according to the corresponding signal light status during the time. For the prohibition time of traffic when the light is on, the time for prohibition of communication is subtracted from the time when the second vehicle passes through the intersection, and the obtained time is the driving time.

Determining whether the second vehicle drives smoothly may specifically include: determining the traveling speed of the second vehicle at each second moment, and performing variance calculation on the traveling speed of the second vehicle at each second moment to obtain the speed variance. Since the variance can describe the degree of dispersion of a set of data, if the speed variance is greater than the preset variance, it indicates that the speed of the second vehicle is unstable, that is, the second vehicle is not driving smoothly. If the speed variance is less than the preset variance, it indicates that The speed of the second vehicle is stable, ie the second vehicle is driving smoothly.

Step 703: The server determines the score of the driving track of the second vehicle according to the driving condition of the second vehicle during the driving process.

In order to determine the best travel trajectory among the travel trajectories of all the second vehicles, the travel trajectory of each second vehicle may be scored. Wherein, the scoring standard is the driving condition of the second vehicle during the driving process.

Specifically, if the number of collisions of the second vehicle during driving is greater than or equal to the preset number of collisions, the collision score is determined to be the first score; otherwise, the collision score is determined to be the second score, where the collision score and the occurrence The number of collisions is negatively correlated.

If the second vehicle complies with the traffic rules, the traffic rule score is determined to be the third score, otherwise, the traffic rule score is determined to be the fourth score.

If the number of lane changes of the second vehicle during driving is greater than or equal to the minimum number of lane changes required for the second vehicle to pass through the intersection, the lane change score is determined to be the fifth score; otherwise, the lane change score is determined to be the fifth score. Six scores, of which the lane change score was negatively correlated with the number of lane changes.

If the driving duration of the second vehicle passing through the intersection is greater than or equal to the preset driving duration, the duration score is determined to be the seventh score, otherwise, the duration score is determined to be the eighth score, wherein the duration score is negatively correlated with the driving duration .

If the second vehicle is driving smoothly, the driving score is determined to be the ninth score, otherwise, the driving score is determined to be the tenth score.

The sum of the collision score, the traffic rule score, the lane change score, the duration score and the driving score is determined as the score of the driving trajectory of the second vehicle.

Among them, the first score, the second score, the third score, the fourth score, the fifth score, the sixth score, the seventh score, the eighth score, the ninth score and the tenth score are preset scores. The set score can be any score, as long as the above conditions are met.

For example, the second score, the third score, the sixth score, the eighth score and the ninth score are set as +5 points, and the first score, the fourth score, the fifth score and the tenth score are set as -5 points. If the number of collisions of the second vehicle during driving is greater than or equal to the preset number of collisions, the collision score is determined to be +5 points, that is, the score of the driving track of the second vehicle will be increased by 5 points at this time. If the second vehicle does not comply with the traffic rules, it is determined that the traffic rule score is -5 points, that is, the score of the driving track of the second vehicle will be reduced by 5 points at this time.

That is, in this embodiment of the present invention, the above steps 701 to 703 may be used to determine the travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period and the score of each travel trajectory.

For example, a plurality of first moments in a preset time period are divided and marked as t0, t1, t2, ..., tm, and starting from the first first moment t0, determine the second vehicle passing through the intersection at the first moment t0 and the driving track of the second vehicle, and determine the score of the driving track through the above steps 802 and 803 . Continue to the next time t1, repeat the above process, and determine the driving track score of the second vehicle passing through the intersection at the first time t1, ..., and so on, until the last first time tm is determined. Driving track score.

Step 704: The server selects N driving trajectories with a score greater than a preset score from all the acquired driving trajectories, where N is greater than 1 and less than or equal to the total number of the acquired driving trajectories.

Since the score of the driving track is determined according to the driving condition of the second vehicle during the driving process, the score of the driving track can be used to describe the excellent degree of the driving track, that is, the higher the score of the driving track, the higher the score of the driving track is. The better the track. Therefore, through step 704, a relatively excellent driving trajectory among the historical driving trajectories can be obtained.

The preset score is a preset score, and the preset score may be 80 points, 90 points, or 95 points.

Step 705: The server trains the initialized neural network model through the N driving trajectories to obtain the target neural network model.

For the training sample of the neural network model in the embodiment of the present invention, the training sample includes multiple independent variables and multiple dependent variables corresponding to the multiple independent variables one-to-one. For the convenience of description, the multiple independent variables are marked as x1, x2,...,xn, mark multiple dependent variables corresponding to multiple independent variables as y1, y2,...,yn. Train the neural network model, that is, make the initialized neural network model learn the mapping relationship between the multiple independent variables and the multiple dependent variables corresponding to the multiple independent variables one-to-one to obtain y=f(x), the y =f(x) is the neural network model after training.

Therefore, the above step 705 may specifically be: determining the driving tasks and driving information of the N second vehicles, and the driving information of the N third obstacle vehicles, where the N second vehicles are the vehicles corresponding to the N driving trajectories, The N second obstacle vehicles are in one-to-one correspondence with the N second vehicles, and the second obstacle vehicle is a vehicle whose distance from the corresponding second vehicle is smaller than a preset distance threshold.

The driving tasks and driving information of the N second vehicles and the driving information of the N second obstacle vehicles are used as the input of the initialized neural network model, and the driving trajectories of the N second vehicles are used as the initialized neural network model. The output of the network model is used to train the initialized neural network model to obtain a neural network model corresponding to the road section where the target position is located.

Further, when the road section where the target position is located is an intersection, the above step 705 may specifically be: determining the driving tasks and driving information of the N second vehicles, and the N second vehicles at the intersection corresponding to the N second vehicles one-to-one. The status of the signal lights, and the driving information of the N second obstacle vehicles

The driving tasks and driving information of the N second vehicles, the state of the N signal lights, and the driving information of the N second obstacle vehicles are used as the input of the initialized neural network model, and the driving of the N second vehicles The trajectory is used as the output of the initialized neural network model, and the initialized neural network model is trained to obtain the neural network model corresponding to the intersection.

That is, the driving tasks and driving information of the N third vehicles, the signal light states at the intersection and the N second vehicles corresponding to each other, and the driving information of the N second obstacle vehicles are used as the training samples. As independent variables, the driving trajectories of the N second vehicles are used as the dependent variables of the training samples to determine the mapping relationship y=f(x) between the independent variables and the dependent variables, that is, to determine the neural network model.

Since the data of the training samples are determined from the travel trajectories of all the second vehicles passing through the intersection within the preset time period, the obtained neural network model is the neural network model corresponding to the intersection.

It should be noted that after the server determines the neural network model corresponding to the road section where the target position is located, the neural network model can be stored, that is, the server stores the neural network model and the road section where the target position is located. The corresponding relationship between, that is, a road segment corresponds to a neural network model.

Afterwards, the server may send the target neural network model corresponding to the target position where the first vehicle is currently located to the first vehicle according to the model acquisition request sent by the first vehicle. Alternatively, the server pushes the neural network model corresponding to the road segment to vehicles passing through the road segment in a broadcast manner.

As can be seen from the above steps 701 to 705, in order to obtain the neural network model corresponding to the road section where the target position is located, the training samples can be determined according to the travel trajectory passing through the road section where the target position is located within the preset time period. Therefore, for a certain road section, if you want to obtain the neural network model corresponding to a certain driving task, you can determine the training sample according to the driving trajectory that passes through the road section and the driving task is the driving task. The process of the corresponding neural network model is basically the same as the above-mentioned process of training the neural network model, which is not described in detail in this embodiment of the present invention.

That is, through the above steps 701 to 705, the server can pre-train different neural network models for different road sections, so that the first vehicle can determine the driving trajectory through the target neural network model corresponding to the road section where the current target position is located. Specifically, the driving trajectory of the first vehicle may be determined through the following steps 706 to 707 .

Step 706: The server determines, from the stored neural network model, a target neural network model corresponding to the road section where the target position of the first vehicle is currently located.

In a possible implementation manner, when the server receives a model acquisition request sent by the first vehicle, the server stores the neural network corresponding to different road sections according to the current target position of the first vehicle carried in the model acquisition request. The target neural network model corresponding to the road section where the target position is located is determined in the model, and the target neural network model is sent to the first vehicle through the following step 707 .

In another possible implementation manner, for the road section where the target position is located, the server determines the target neural network model corresponding to the road section where the target position is located from the stored neural network models corresponding to different road sections, and passes the following step 707 The target neural network model is pushed to vehicles within the range of the road segment in a broadcast manner. Accordingly, the first vehicle currently at the target position can directly receive the target neural network model.

Step 707: The server sends the target neural network model to the first vehicle, so that the first vehicle can determine the travel trajectory of the first vehicle through the target neural network model.

Wherein, the first vehicle may specifically determine the driving trajectory according to the target neural network model through the following steps 708 and 709 .

Step 708: The first vehicle receives the target neural network model sent by the server.

It can be known from step 707 that, in this embodiment of the present invention, the first vehicle may receive the target neural network model from the server through two different implementation manners. It will not be elaborated here.

Step 709: The first vehicle determines the driving trajectory of the first vehicle through the target neural network model according to the driving task and driving information of the first vehicle and the driving information of the first obstacle vehicle, wherein the driving tasks include going straight, turning left, Right turn and U-turn, the travel information includes current location, travel direction and travel speed, and the first obstacle vehicle is a vehicle whose distance from the first vehicle is less than a preset distance threshold.

Specifically, step 709 can be implemented by the following two steps:

(1) Acquire the driving task of the first vehicle, the driving information of the first vehicle, and the driving information of the first obstacle vehicle.

Specifically, the first vehicle may determine its current driving task through the task planning module shown in FIG. 4 to obtain the driving task of the first vehicle. Determine your current position through the positioning module, and determine your own driving direction and driving speed through the perception module, as well as the current position, driving direction and driving speed of the first obstacle vehicle, and obtain the first vehicle and the first vehicle. Driving information of obstacle vehicles.

Among them, the task planning module determines whether the current driving task is going straight, turning left, turning right or turning around according to the predetermined navigation path and the current position. The positioning module may determine the current position of the first vehicle through GPS (Global Positioning System, global positioning system) technology. The perception module may determine the driving direction, the driving speed of the first vehicle and the driving information of the first obstacle vehicle through the video collected by the camera installed on the first vehicle.

In addition, the preset distance threshold is a preset distance. When the distance between two vehicles is less than the preset distance threshold, the driving situation of one vehicle may affect the driving situation of the other vehicle. Each vehicle is an obstacle vehicle. The preset distance threshold may be 1 meter, 0.75 meters, or 0.5 meters.

Optionally, when the road segment where the target position of the first vehicle is currently located is an intersection, in order to reduce the probability of a traffic accident when the first vehicle is driving at the intersection, when determining the driving trajectory of the first vehicle, it is also necessary to Consider the signal light state at the intersection corresponding to the target location.

Therefore, after acquiring the driving task of the first vehicle, the driving information of the first vehicle, and the driving information of the first obstacle vehicle, the first vehicle also needs to acquire the signal light state corresponding to the target position at the intersection.

(2) According to the driving task and driving information of the first vehicle, and the driving information of the first obstacle vehicle, determine the driving trajectory of the first vehicle through the target neural network model.

Specifically, the first vehicle may use the driving task of the first vehicle, the driving information of the first vehicle, and the driving information of the first obstacle vehicle as the input of the neural network model, and determine the first vehicle through the neural network model corresponding to the road section. 's driving trajectory.

Since the target neural network model is trained according to the driving trajectories of a plurality of second vehicles passing through the road section where the target position is located before the current time, the target neural network model has learned the characteristics of the historical driving trajectories, so when When the driving task of the first vehicle, the driving information of the first vehicle and the driving information of the first obstacle vehicle are used as the input of the target neural network model, the target neural network model can determine the characteristics of the historical driving trajectory that have been learned. The travel trajectory of the first vehicle.

Further, when the road segment where the target position of the first vehicle is currently located is an intersection, at this time, the first vehicle can combine the driving task of the first vehicle, the driving information of the first vehicle, the driving information of the first obstacle vehicle, and the The signal light state corresponding to the target position at the intersection is used as the input of the neural network model, and the driving trajectory of the first vehicle is determined through the neural network model corresponding to the road section.

Wherein, the neural network model corresponding to the road section where the target position is located is obtained by the server through training in advance according to the historical travel trajectories of historical vehicles passing through the road section. The neural network model trained by the server corresponding to the road section where the target position is located will be described in detail in the following embodiments, and will not be explained here.

In the embodiment of the present invention, the server may train the initialized neural network model according to the travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period, and obtain the target corresponding to the road section where the target position is located A neural network model, so that the first vehicle can determine the driving trajectory according to the target neural network model. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

In addition to providing the method for planning a driving trajectory described in the foregoing embodiments, the embodiments of the present invention also provide a device for planning a driving trajectory, which will be introduced in the following embodiments.

FIG. 8 is an apparatus 800 for planning a driving trajectory provided by an embodiment of the present invention, which is applied to the vehicle shown in FIG. 1 . As shown in FIG. 8 , the apparatus 800 includes a receiving unit 801 and a first determining unit 802:

a receiving unit 801, configured to perform step 603 in the embodiment shown in FIG. 6A or step 708 in the embodiment shown in FIG. 7 ;

a first determining unit 802, configured to perform step 604 in the embodiment shown in FIG. 6A or step 709 in the embodiment shown in FIG. 7 ;

Wherein, the driving task includes going straight, turning left, turning right and U-turn, the driving information includes the position information of the current location, the driving direction and the driving speed, and the distance between the first obstacle vehicle and the first vehicle is less than the preset distance Threshold vehicles.

Optionally, the road section where the target position is located is an intersection;

The first determining unit 802 is specifically configured to:

The driving task and driving information of the first vehicle, the driving information of the first obstacle vehicle, and the signal light state corresponding to the target position at the intersection are used as the input of the target neural network model, and the target neural network model is used to determine the first vehicle. driving track.

Optionally, the target neural network model refers to a neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle, and the plurality of second vehicles refer to vehicles passing through the target position before the current time. A vehicle with the same driving task as the driving task of the first vehicle.

In the embodiment of the present invention, the server may train the initialized neural network model according to the travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period, and obtain the target corresponding to the road section where the target position is located A neural network model, so that the first vehicle can determine the driving trajectory according to the target neural network model. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

It should be noted that: when planning the driving trajectory of the first vehicle, the device for planning a driving trajectory provided in the above-mentioned embodiment only takes the division of the above-mentioned functional modules as an example. In practical applications, the above-mentioned functions may be allocated by Different functional modules are completed, that is, the internal structure of the first vehicle is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the device for planning a driving trajectory provided in the above embodiment and the method for planning a driving trajectory in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which is not repeated here.

9A is another apparatus 900 for planning a driving trajectory provided by an embodiment of the present invention, which is applied to the server shown in FIG. 1 . As shown in FIG. 9A , the apparatus 900 includes a second determining unit 901 and a sending unit 902:

The second determination unit 901 is configured to perform step 601 in the embodiment shown in FIG. 6A or step 706 in the embodiment shown in FIG. 7 ;

The sending unit 902 is configured to perform step 602 in the embodiment shown in FIG. 6A or step 707 in the embodiment shown in FIG. 7 .

Optionally, referring to FIG. 9B , the apparatus 900 further includes a third determination unit 903, a selection unit 904 and a training unit 905:

a third determining unit 903, configured to perform steps 701 to 703 in the embodiment shown in FIG. 7;

a selection unit 904, configured to perform step 704 in the embodiment shown in FIG. 7;

The training unit 905 is configured to perform step 705 in the embodiment shown in FIG. 7 .

Optionally, the training unit 905 includes a first determination subunit and a training subunit:

a first determination subunit, configured to determine the driving tasks and driving information of the N second vehicles, and the driving information of the N second obstacle vehicles;

The N second vehicles are vehicles corresponding to the N driving trajectories, the N second obstacle vehicles are in one-to-one correspondence with the N second vehicles, and the second obstacle vehicles are between the corresponding second vehicles The distance of the vehicle is less than the preset distance threshold, the driving task includes going straight, turning left, turning right and U-turn, and the driving information includes the current position, driving direction and driving speed;

a training subunit, configured to train the initialized neural network model according to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles and the driving trajectories of the N second vehicles, Get the target neural network model.

Optionally, the road section where the target position is located is an intersection;

The training unit 905 also includes a second determination subunit:

The second determination subunit is used to determine N signal light states, the N signal light states are in one-to-one correspondence with the N second vehicles, and each signal light state refers to the corresponding second vehicle at the intersection when passing through the intersection. the status of the signal light;

Correspondingly, the training subunit is specifically used for:

The driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the N signal light states are used as the input of the initialized neural network model, and the driving trajectories of the N second vehicles As the output of the initialized neural network model, the initialized neural network model is trained to obtain the target neural network model.

Optionally, the third determining unit 903 is specifically configured to perform steps 702 and 703 in the embodiment shown in FIG. 7 .

Optionally, the second determining unit 901 is specifically configured to:

According to the position information of the target position and the driving task of the first vehicle, determine the target neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle from the stored neural network model;

Correspondingly, the plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle.

In the embodiment of the present invention, the server may train the initialized neural network model according to the travel trajectories of all second vehicles passing through the road section where the target position is located within a preset time period, and obtain the target corresponding to the road section where the target position is located A neural network model, so that the first vehicle can determine the driving trajectory according to the target neural network model. That is, when the first vehicle determines its own driving trajectory through the target neural network model at the target position, it not only considers the driving information of the first vehicle and the first obstacle vehicle, but also refers to the location where the target position passed before the current time. The driving trajectories of a plurality of second vehicles on the road segment are determined, so as to reduce the accident rate when the first vehicle travels according to the determined driving trajectory, that is, to improve the feasibility of the determined driving trajectory.

It should be noted that: when planning the driving trajectory of the first vehicle, the device for planning a driving trajectory provided in the above-mentioned embodiment only takes the division of the above-mentioned functional modules as an example. In practical applications, the above-mentioned functions may be allocated by Different functional modules are completed, that is, the internal structure of the server is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the device for planning a driving trajectory provided in the above embodiment and the method for planning a driving trajectory in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which is not repeated here.

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented in software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer may be a general purpose computer, special purpose computer, computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server or data center Transmission is performed to another website site, computer, server or data center by wire (eg coaxial cable, optical fiber, Digital Subscriber Line, DSL) or wireless (eg infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes an integration of one or more available media. The usable media may be magnetic media (eg: floppy disk, hard disk, magnetic tape), optical media (eg: Digital Versatile Disc (DVD)), or semiconductor media (eg: Solid State Disk (SSD)) )Wait.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, etc.

The above-mentioned examples provided for this application are not intended to limit this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the protection scope of this application. Inside.

Claims (18)

1. A method for planning a driving trajectory, applied to a first vehicle, wherein the method comprises: Receive the target neural network model sent by the server, where the target neural network model refers to the neural network model corresponding to the road section where the target position of the first vehicle is currently located, and the target neural network model is the obtained by training the driving trajectories of a plurality of second vehicles, where the plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time; According to the driving task and driving information of the first vehicle, and the driving information of the first obstacle vehicle, determine the driving trajectory of the first vehicle through the target neural network model; Wherein, the driving task includes going straight, turning left, turning right, and making a U-turn, the driving information includes position information of the current location, driving direction and driving speed, and the first obstacle vehicle is the same as the first vehicle. The distance between the vehicles is less than the preset distance threshold. 2. The method according to claim 1, wherein the road section where the target position is located is an intersection; According to the driving task and driving information of the first vehicle, and the driving information of the first obstacle vehicle, the driving trajectory of the first vehicle is determined by the target neural network model, including: Taking the driving task and driving information of the first vehicle, the driving information of the first obstacle vehicle and the signal light state corresponding to the target position at the intersection as the input of the target neural network model, through the target The neural network model determines the travel trajectory of the first vehicle. 3. The method according to claim 1 or 2, wherein the target neural network model refers to a neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle, and The plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle. 4. A method for planning a driving trajectory, applied to a server, wherein the method comprises: A target neural network model corresponding to the road section where the target position of the first vehicle is currently located is determined from the stored neural network model, and the target neural network model is obtained by training according to the driving trajectories of a plurality of second vehicles. The plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time; The target neural network model is sent to the first vehicle, so that the first vehicle determines the driving trajectory of the first vehicle through the target neural network model. 5. The method according to claim 4, characterized in that, before determining the target neural network model corresponding to the road section where the target position of the first vehicle is currently located from the stored neural network model, the method further comprises: determining the driving trajectories of all the second vehicles passing through the road section where the target position is located within the preset time period and the score of the driving trajectories of each second vehicle; Selecting N driving trajectories with a score greater than a preset score from all the acquired driving trajectories, where N is greater than 1 and less than or equal to the total number of the acquired driving trajectories; The initialized neural network model is trained through the N driving trajectories to obtain the target neural network model. 6. The method according to claim 5, wherein the training of the initialized neural network model through the N driving trajectories to obtain the target neural network model comprises: determining the driving tasks and driving information of the N second vehicles, and the driving information of the N second obstacle vehicles; Wherein, the N second vehicles are vehicles corresponding to the N driving trajectories, the N second obstacle vehicles are in one-to-one correspondence with the N second vehicles, and the second obstacle vehicles are vehicles corresponding to the corresponding No. The distance between the two vehicles is less than the preset distance threshold value, the driving task includes going straight, turning left, turning right and U-turn, and the driving information includes the current position, driving direction and driving speed; According to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the driving trajectories of the N second vehicles, the initialized neural network model is trained to obtain the target neural network model. 7. The method according to claim 6, wherein the road section where the target position is located is an intersection; After determining the driving tasks and driving information of the N second vehicles and the driving information of the N second obstacle vehicles, the method further includes: Determine N signal light states, the N signal light states are in one-to-one correspondence with the N second vehicles, and each signal light state refers to the signal light state corresponding to the intersection when the corresponding second vehicle passes through the intersection; Correspondingly, the initialized neural network model is performed according to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the driving trajectories of the N second vehicles. Training to obtain the target neural network model, including: The driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the N signal light states are used as the input of the initialized neural network model, and the Nth The driving trajectory of the second vehicle is used as the output of the initialized neural network model, and the initialized neural network model is trained to obtain the target neural network model. 8. The method according to claim 7, wherein the determining the score of the driving trajectory of each second vehicle comprises: For any second vehicle among all the second vehicles, determine the driving condition of the second vehicle during the driving process according to the driving track of the second vehicle, where the driving condition includes the number of collisions and whether the traffic rules are complied with. , the number of lane changes, the driving time and whether it is a smooth driving; The score of the driving trajectory of the second vehicle is determined according to the driving condition of the second vehicle during the driving process. 9. The method according to any one of claims 4 to 8, wherein the determining, from the stored neural network model, the target neural network model corresponding to the road section where the target position of the first vehicle is currently located, comprising: : According to the position information of the target position and the driving task of the first vehicle, a target neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle is determined from the stored neural network model ; Correspondingly, the plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle. 10. A device for planning a driving trajectory, applied to a first vehicle, wherein the device comprises: a receiving unit, configured to receive a target neural network model sent by a server, where the target neural network model refers to a neural network model corresponding to the road section where the target position of the first vehicle is currently located, and the target neural network model is obtained by the server through training according to the travel trajectories of multiple second vehicles, where the multiple second vehicles refer to vehicles passing through the road section where the target position is located before the current time; a first determining unit, configured to determine the driving trajectory of the first vehicle through the target neural network model according to the driving task and driving information of the first vehicle and the driving information of the first obstacle vehicle; Wherein, the driving task includes going straight, turning left, turning right, and making a U-turn, the driving information includes position information of the current location, driving direction and driving speed, and the first obstacle vehicle is the same as the first vehicle. The distance between the vehicles is less than the preset distance threshold. 11. The device according to claim 10, wherein the road section where the target position is located is an intersection; The first determining unit is specifically used for: Taking the driving task and driving information of the first vehicle, the driving information of the first obstacle vehicle and the signal light state corresponding to the target position at the intersection as the input of the target neural network model, through the target The neural network model determines the travel trajectory of the first vehicle. 12. The device according to claim 10 or 11, wherein the target neural network model refers to a neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle, and The plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle. 13. A device for planning a driving trajectory, applied to a server, wherein the device comprises: The second determination unit is configured to determine, from the stored neural network model, a target neural network model corresponding to the road section where the target position of the first vehicle is currently located, where the target neural network model is based on the driving of a plurality of second vehicles. Obtained from trajectory training, the plurality of second vehicles refer to vehicles passing through the road section where the target position is located before the current time; A sending unit, configured to send the target neural network model to the first vehicle, so that the first vehicle can determine the driving trajectory of the first vehicle through the target neural network model. 14. The apparatus of claim 13, wherein the apparatus further comprises: a third determining unit, configured to determine the driving trajectories of all second vehicles passing through the road section where the target position is located within a preset time period and the score of the driving trajectories of each second vehicle; a selection unit, configured to select, from all the acquired driving trajectories, N driving trajectories with a score greater than a preset score, where N is greater than 1 and less than or equal to the total number of the acquired driving trajectories; A training unit, configured to train the initialized neural network model through the N driving trajectories to obtain the target neural network model. 15. The apparatus according to claim 14, wherein the training unit comprises: a first determination subunit, configured to determine the driving tasks and driving information of the N second vehicles, and the driving information of the N second obstacle vehicles; Wherein, the N second vehicles are vehicles corresponding to the N driving trajectories, the N second obstacle vehicles are in one-to-one correspondence with the N second vehicles, and the second obstacle vehicles are vehicles corresponding to the corresponding No. The distance between the two vehicles is less than the preset distance threshold value, the driving task includes going straight, turning left, turning right and U-turn, and the driving information includes the current position, driving direction and driving speed; A training subunit, configured to perform an initialized neural network model according to the driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the driving trajectories of the N second vehicles Perform training to obtain the target neural network model. 16. The device according to claim 15, wherein the road section where the target position is located is an intersection; The training unit also includes: The second determination subunit is configured to determine N signal light states, the N signal light states are in one-to-one correspondence with the N second vehicles, and each signal light state refers to the state of the corresponding second vehicle when it passes through the intersection. The state of the corresponding signal light at the intersection; Correspondingly, the training subunit is specifically used for: The driving tasks and driving information of the N second vehicles, the driving information of the N second obstacle vehicles, and the N signal light states are used as the input of the initialized neural network model, and the Nth The driving trajectory of the second vehicle is used as the output of the initialized neural network model, and the initialized neural network model is trained to obtain the target neural network model. 17. The apparatus according to claim 16, wherein the third determining unit is specifically configured to: For any second vehicle among all the second vehicles, determine the driving condition of the second vehicle during the driving process according to the driving track of the second vehicle, where the driving condition includes the number of collisions and whether the traffic rules are complied with. , the number of lane changes, the driving time and whether it is a smooth driving; The score of the driving trajectory of the second vehicle is determined according to the driving condition of the second vehicle during the driving process. 18. The apparatus according to any one of claims 13 to 17, wherein the second determining unit is specifically configured to: According to the position information of the target position and the driving task of the first vehicle, a target neural network model corresponding to both the road section where the target position is located and the driving task of the first vehicle is determined from the stored neural network model ; Correspondingly, the plurality of second vehicles refer to vehicles that have passed through the road section where the target position is located before the current time and have the same driving task as the driving task of the first vehicle.

Images