Vehicle control method and system

Abstract

The embodiment of the invention provides a vehicle control method which is executed by a vehicle and comprises the steps that environment data collected by a sensor of the vehicle is obtained; extracting road elements according to the environmental data; determining road change information based on the road elements and the map data; the road change information is input into a strategy determination model, a control strategy of the vehicle is determined, and the strategy determination model is a machine learning model. The automatic driving vehicle is helped to deal with the situation that the actual road is inconsistent with the road in the map, and the experience of automatic driving and the road safety are improved.

Description

Technical Field

[0001] This specification relates to the field of mobile control, and particularly to a vehicle control method and system. Background Art

[0002] In the field of mobile control, for example, in autonomous driving, traditional autonomous driving solutions rely relatively heavily on environmental information provided by offline high-precision maps, etc. When there are changes in the actual road, such as construction or re-drawing of lane lines, it may lead to a situation where the actual environment in this area is inconsistent with the environmental information in the offline map. Or when an autonomous driving vehicle travels to a mapless area outside the autonomous driving area, it may cause the autonomous driving vehicle to fail to timely perceive potential risks in the environment, and abnormal situations are likely to occur, such as emergency braking, deviation of the driving direction, etc. At this time, the autonomous driving vehicle cannot guarantee the personal safety of the user, and manual takeover of the driving function is required to ensure the normal driving of the vehicle.

[0003] Therefore, providing a vehicle control method and system helps to improve the environmental perception ability of autonomous driving vehicles and improve the safety of autonomous driving. Summary of the Invention

[0004] One or more embodiments of this specification provide a vehicle control method, which is executed by the vehicle and includes: obtaining environmental data collected by sensors of the vehicle; extracting road elements according to the environmental data; determining road change information based on the road elements and map data; inputting the road change information into a policy determination model to determine the control policy of the vehicle, and the policy determination model is a machine learning model.

[0005] One or more embodiments of this specification provide a vehicle control system, and the system includes: an obtaining module configured to obtain environmental data collected by sensors of the vehicle; an extracting module configured to extract road elements according to the environmental data; a first determination module configured to determine road change information based on the road elements and map data; a second determination module configured to input the road change information into a policy determination model to determine the control policy of the vehicle, and the policy determination model is a machine learning model.

[0006] One or more embodiments of this specification provide a computer-readable storage medium, and the storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the vehicle control method of any one of the above. Brief Description of the Drawings

[0007] This specification will be further described in the form of exemplary embodiments, and these exemplary embodiments will be described in detail through the drawings. These embodiments are not restrictive. In these embodiments, the same numbers represent the same structures, where:

[0008] Figure 1 is an application scenario diagram of a vehicle control system shown in some embodiments of this specification;

[0009] Figure 2 is an exemplary flowchart of a vehicle control method shown in some embodiments of this specification;

[0010] Figure 3 is an exemplary schematic diagram of extracting road elements shown in some embodiments of this specification;

[0011] Figure 4 is an exemplary schematic diagram of an element extraction model shown in some embodiments of this specification;

[0012] Figure 5 is an exemplary schematic diagram of a policy determination model shown in some embodiments of this specification;

[0013] Figure 6 is a module diagram of a vehicle control system shown in some embodiments of this specification. Detailed implementation manners

[0014] To more clearly illustrate the technical solutions of the embodiments of this specification, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some examples or embodiments of this specification. For those of ordinary skill in the art, without creative efforts, this specification can also be applied to other similar scenarios based on these drawings. Unless obvious from the language context or otherwise stated, the same reference numerals in the figures represent the same structure or operation.

[0015] It should be understood that the "system", "device", "unit" and / or "module" used herein is a way to distinguish different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the said words can be replaced by other expressions.

[0016] As shown in this specification and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0017] Flowcharts are used in this specification to illustrate the operations performed by the systems according to the embodiments of this specification. It should be understood that the operations before or after do not necessarily have to be executed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Also, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0018] Generally, during the process of autonomous driving, if the road environment changes, it may lead to traffic accidents. Therefore, during the process of autonomous driving, it is necessary to take control measures in a timely manner according to the change information of the current road. Currently, most vehicles rely only on offline maps for autonomous driving and cannot change the driving strategy in real time in combination with the actual road conditions, resulting in a low road passing ability for autonomous driving and being unable to handle unfamiliar road conditions and changing road conditions, and relying on the driver for manual takeover; while the rule-based autonomous driving strategy can handle most road conditions, but the rule-based autonomous driving strategy has poor information perception ability for the environment, making the autonomous driving strategy unable to fully adapt to the changing environment.

[0019] In view of this, in some embodiments of this specification, based on the environmental data sensed by the vehicle in real time, the existing offline map is fused, and the control strategy of the vehicle is determined through the fused result, solving the problem that autonomous driving vehicles cannot pass smoothly when encountering road changes, and controlling the vehicle to drive according to the control strategy, improving the intelligence of autonomous driving.

[0020] Figure 1 It is a schematic diagram of the application scenario of the vehicle control system shown in some embodiments of this specification.

[0021] In some embodiments, the application scenario 100 can be applied to autonomous driving vehicles in various fields. For example, driverless taxis, unmanned road sweepers, unmanned delivery vehicles, etc. It should be noted that the autonomous driving vehicles mentioned here can refer to fully autonomous driverless vehicles or semi-autonomous driverless vehicles. The vehicle can include taxis, private cars, hitchhiking cars, shared vehicles, buses, trains, bullet trains, high-speed rails, subways, ships, airplanes, spaceships, hot air balloons, bicycles, tricycles, motorcycles, robots, etc., or any combination thereof. The systems and methods of this application can also be applied to scenarios of autonomous mobile robots, such as scenarios of food delivery robots, patrol robots, etc.

[0022] In semi-autonomous driverless vehicles, some functions can be optionally manually controlled by the user (e.g., passengers or operators of the vehicle) some or all of the time. In addition, semi-autonomous driverless vehicles can be configured to be able to switch between a fully manual operation mode, a semi-automatic operation mode, and / or a fully automatic operation mode.

[0023] In some embodiments, the application scenario 100 can be applied to navigation path planning and the like. For example, the application scenario 100 can be applied to path navigation for autonomous driving. An autonomous driving vehicle can travel based on a control strategy, thereby improving the safety, intelligence, and comfort of the vehicle. In some embodiments, the application scenario 100 can be applied to online car-hailing services and the like. In some embodiments, the application scenario 100 can also be applied to driving services, express delivery, food delivery, and the like.

[0024] In some embodiments, as Figure 1 shown, the application scenario 100 can include a server 110, a vehicle 120, a network 130, and a memory 140.

[0025] In some embodiments, the server 110 can be a single server or a server group. The server group can be centralized or distributed (for example, the server 110 can be a distributed system). In some embodiments, the server 110 can be local or remote. For example, the server 110 can access information and / or data stored in the vehicle 120 and / or the memory 140 via the network 130. Alternatively, the server 110 can be directly connected to the vehicle 120 and / or the memory 140 to access the stored information and / or data. In some embodiments, the server 110 can be implemented on a cloud platform or an in-vehicle computer. By way of example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, or any combination thereof.

[0026] In some embodiments, the server 110 can include a processing device 112. The processing device 112 can aggregate information uploaded by multiple vehicles, such as road change information determined by each vehicle and the corresponding control strategy, etc. The processing device 112 can integrate the collected information to obtain aggregated information. The processing device 112 can also send the collected aggregated information to the processing units of each vehicle as reference information to be used as auxiliary information for each vehicle to determine its control strategy.

[0027] In some embodiments, the processing device 112 can include one or more processing devices (for example, a single-chip processing device or a multi-chip processing device). By way of example only, the processing device 112 can include a central processing unit (CPU), an application-specific integrated circuit (ASIC), a graphics processing unit (GPU), a digital signal processor (DSP), a field-programmable gate array (FPGA), a microprocessor, or any combination thereof.

[0028] In some embodiments, the server 110 may be connected to the network 130 to communicate with one or more components of the application scenario 100 (e.g., the vehicle 120, the memory 140). In some embodiments, the server 110 may be directly connected to or communicate with one or more components of the application scenario 100 (e.g., the vehicle 120, the memory 140).

[0029] The network 130 may facilitate the exchange of information and / or data. In some embodiments, one or more components of the application scenario 100 (e.g., the server 110, the vehicle 120, or the memory 140) may send information and / or data to other components of the application scenario 100 via the network 130. For example, the server 110 may obtain environmental data related to the vehicle 120 and / or the state of the vehicle 120 via the network 130.

[0030] In some embodiments, the network 130 may be a wired network, a wireless network, or any combination thereof. By way of example only, the network 130 may include a cable network, a wired network, an optical fiber network, a telecommunications network, an internal network, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, a near field communication (NFC) network, etc., or any combination thereof.

[0031] In some embodiments, the network 130 may include one or more network access points. For example, the network 130 may include a wired or wireless network access point through which one or more components of the application scenario 100 may be connected to the network 130 to exchange data and / or information.

[0032] The vehicle 120 may be a device with an autonomous driving function, such as an autonomous vehicle. The autonomous vehicle is capable of sensing environmental information and performing driving planning without human manipulation. The vehicle 120 may include the structure of a traditional vehicle. For example, the vehicle 120 may include a number of control components configured to control the operation of the vehicle 120. The control components may control components including a steering device (e.g., a steering wheel), a braking device (e.g., a brake pedal), an accelerator, etc. The steering device may be configured to adjust the orientation and / or direction of the vehicle 120. The braking device may be configured to perform a braking operation to stop the vehicle 120. The accelerator may be configured to control the speed and / or acceleration of the vehicle 120.

[0033] In some embodiments, the vehicle 120 is equipped with a processing unit 121, which can process information and / or data related to environmental data and / or the state of the vehicle 120 to perform one or more functions described in this application. For example, the processing unit 121 can obtain environmental data related to the vehicle 120 (e.g., road information, obstacle information) and / or the state of the vehicle 120 (e.g., current location, current speed). The processing unit 121 can determine the control strategy of the vehicle 120 based on the environmental data, the state of the vehicle 120, and / or the reference information obtained from the processing device 112, and control the operation of the corresponding control components based on the control strategy. The processing unit 121 can be configured as a planning component of the vehicle 120. For example, the processing unit 121 can input road change information into a strategy determination model to determine the control strategy of the vehicle 120.

[0034] The processing unit 121 can include a central processing unit (CPU), an application specific integrated circuit (ASIC), a graphics processing unit (GPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a microprocessor, etc., or any combination thereof.

[0035] In some embodiments, the vehicle 120 can be configured to be operated by a user using the vehicle 120, remotely controlled, and / or autonomously.

[0036] In some embodiments, the vehicle 120 can be equipped with at least one sensor. In some embodiments, the sensor can include an imaging device, a LIDAR (Light Detection and Ranging) device, a RADAR (Radio Detection and Ranging) device, an ultrasonic sensor, an environmental sensor, etc.

[0037] In some embodiments, the imaging device can detect visible light, infrared light, or ultraviolet light. For example, the imaging device can be a camera. The camera can be any type of camera, such as a still camera, a video camera, etc. The imaging device can capture images and / or videos of the environment.

[0038] The LIDAR device can use laser beams to scan objects (e.g., other vehicles, pedestrians and / or obstacles on the road, traffic signals and signs, roads, intersections, lane information, lane boundaries, buildings, etc.) and obtain LIDAR measurement results. The LIDAR measurement results include three-dimensional (3D) point cloud data of the scanned objects, and the 3D point cloud data can provide a basis for establishing a three-dimensional model of the scanned objects. The 3D point cloud data can include 3D spatial information (e.g., 3D coordinates) of the scanned objects, intensity information (e.g., measurement of the retroreflectivity characteristics of the scanned objects), the distance between the scanned objects and the vehicle, etc.

[0039] A RADAR device can use radio waves to obtain RADAR measurement results. The RADAR measurement results can include the distance between the scanned object and the vehicle, the speed of the scanned object, the direction of the scanned object, the height of the scanned object, etc.

[0040] An ultrasonic sensor can obtain ultrasonic measurement results based on the propagation of sound. The ultrasonic measurement results can include the distance between the scanned object and the vehicle, the speed of the scanned object, the direction of the scanned object, the height of the scanned object, etc.

[0041] An environmental sensor can include one or more sensors for obtaining environmental information. The environmental information can include information related to the state of the vehicle (e.g., remaining battery power, remaining gasoline quantity, etc.), weather conditions (e.g., sunny, rainy, snowy), traffic conditions (e.g., smooth or congested roads), visibility conditions, wind speed, temperature, humidity, air pollution index (e.g., PM2.5 index), etc., or a combination of one or more of them.

[0042] In some embodiments, the vehicle 120 can be equipped with a positioning device for detecting the position of the vehicle 120. The positioning device can include a Global Positioning System (GPS) device, a Global Navigation Satellite System (GLONASS) device, a Compass Navigation System (COMPASS) device, a BeiDou Navigation Satellite System device, a Galileo positioning system device, a Quasi-Zenith Satellite System (QZSS) device, etc. The positioning device can provide real-time position information of the vehicle 120 when the vehicle 120 is traveling; for example, providing the position and driving direction of the vehicle 120 at each time point.

[0043] In some embodiments, the environmental data related to the vehicle 120 can include perception information (e.g., road information, obstacle information) within a certain range from the vehicle 120, map information of the location where the vehicle 120 is located, etc.

[0044] The memory 140 can store data and / or instructions. In some embodiments, the memory 140 can store data obtained from the vehicle 120, such as environmental data and / or the state of the vehicle 120 acquired by at least two detection units. In some embodiments, the memory 140 can store the data and / or instructions used by the server 110 to execute or use to complete the exemplary methods described in this application.

[0045] In some embodiments, the memory 140 can include a mass storage device, a removable storage device, a volatile read-write memory, a read-only memory (ROM), etc., or any combination thereof. In some embodiments, the memory 140 can be implemented on a cloud platform. By way of example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

[0046] In some embodiments, the memory 140 may be connected to the network 130 to communicate with one or more components of the application scenario 100 (e.g., the server 110, the vehicle 120). One or more components of the application scenario 100 may access the data or instructions stored in the memory 140 via the network 130. In some embodiments, the memory 140 may be directly connected to or communicate with one or more components of the application scenario 100 (e.g., the server 110, the vehicle 120). In some embodiments, the memory 140 may be a part of the server 110. In some embodiments, the memory 140 may be integrated in the vehicle 120.

[0047] It should be noted that the application scenario 100 of the vehicle control system is provided for illustrative purposes only and is not intended to limit the scope of this specification. Those of ordinary skill in the art can make various changes and modifications according to the description of this specification. For example, the application scenario 100 may further include a database, an information source, etc. Again, for example, the application scenario 100 may be implemented on other devices to achieve similar or different functions. However, these changes and modifications will not depart from the scope of this specification.

[0048] Figure 2 is an exemplary flowchart of a vehicle control method shown in some embodiments of this specification. In some embodiments, the process 200 may be executed based on the processing device of the vehicle control system. As Figure 2 shown, the process 200 includes the following steps.

[0049] Step 210, obtaining environmental data collected by sensors of the vehicle.

[0050] The environmental data is information related to the environment in which the vehicle is located. For example, the environmental data may include information about objects around the vehicle, such as dynamic object information (e.g., other vehicles, pedestrians, animals, etc.) and static object information (e.g., buildings, roads, street lights, traffic signs, traffic lights, etc.). In some embodiments, the static objects may include planar objects (e.g., lane boundary lines, lane center lines, etc.) or non-planar objects (e.g., buildings, traffic lights, street lights, signs, flower beds, road dividers, etc.). In some embodiments, the environmental data may include various forms of data. For example, the environmental data may include images and / or videos of the environment, lidar (LIDAR) measurement results of the environment, radar (RADAR) measurement results of the environment, ultrasonic measurement results of the environment, environmental information, and other data.

[0051] In some embodiments, the processing device may collect environmental data through sensors installed on the vehicle. For example, the processing device may capture images and / or videos of the environment where the vehicle is located through imaging devices installed on the vehicle.

[0052] In some embodiments, the processing device may obtain environmental data over time to track the environment where the vehicle is located. For example, the sensor may communicate with the server through a network to send sensor data continuously (in real time) or periodically or intermittently.

[0053] Step 220, extract road elements according to the environmental data.

[0054] Road elements refer to information related to the driving road of the vehicle. For example, road elements may include planar elements and three-dimensional elements within and around the road on which the vehicle travels.

[0055] Among them, planar elements refer to two-dimensional elements corresponding to planar objects in the vehicle's surrounding environment, such as line elements, surface elements, etc. Line elements can represent various forms of boundaries in the surrounding environment, such as road boundaries, lane centerlines, left and right lane boundaries, lane directions, etc. Surface elements can represent various forms of regions in the surrounding environment, such as lanes, virtual lanes, etc.

[0056] Three-dimensional elements refer to three-dimensional elements corresponding to non-planar objects in the surrounding environment, such as volume elements. Volume elements can represent various three-dimensional shaped objects in the surrounding environment, such as traffic signs, buildings, etc.

[0057] The geographical area where the vehicle is located may refer to the location area corresponding to the vehicle's location. The shape of the geographical area can be triangular, rectangular, square, hexagonal, circular, etc. For example, the shape of the geographical area can be a rectangle with dimensions of M meters × N meters, where M and N can be preset values.

[0058] In some embodiments, the road elements of the driving road can be represented in various forms such as sequences or matrices. For example, the road elements can be represented in the form of a sequence {A, B}, where the A and B elements in the sequence represent planar elements and three-dimensional elements respectively.

[0059] In some embodiments, each element in the sequence may include multiple sub-elements, and different sub-elements correspond to different characteristic information. Only as an example, element A represents the planar element of the driving road, and element A can be represented as (A1, A2,...), where A1 represents the lane line, A2 represents the stop line,...

[0060] Regarding the representation form of road elements, only as an example here, road elements can also be other representation forms.

[0061] In some embodiments, the processing device may extract road elements in various ways based on environmental data. For example, the processing device may identify objects (such as traffic lights, guiding lines, roadblocks, buildings, or vehicles, etc.) in the environmental data through various image processing methods and machine learning methods to obtain road elements.

[0062] For example, the processing device may analyze and process the acquired image based on image processing algorithms (such as support vector machines, K-nearest neighbors, decision trees, neural networks, etc.) and extract road elements in the image.

[0063] In some embodiments, the processing device may determine road elements based on the bird's-eye view data and map data converted from environmental data through an element extraction model. For more content on how to determine road elements through the element extraction model, reference can be made to Figure 3 、 Figure 4 the relevant description.

[0064] Step 230, determine road change information based on road elements and map data.

[0065] The map data may be an offline map pre-loaded in the vehicle or an online map obtained online, etc. In some embodiments, the map data may be a high-precision map containing centimeter-level or millimeter-level accuracy information. The map data may present driving assistance information related to the geographical area where the vehicle is located, such as the direction of the road, the location of intersections, the location of traffic lights, lane rules, etc.

[0066] In some embodiments, the map data may include map elements within the geographical area where the vehicle is located. Map elements refer to road elements presented on the map.

[0067] Road change information is information related to changes in road elements. In some embodiments, the road change information may be determined based on the inconsistent element information between the road elements extracted from the environmental data and the map elements on the map data.

[0068] For example, for a certain intersection, if the map data shows that there is no traffic light at this intersection, while the road elements collected by the vehicle reflect that there is actually a traffic light set at this intersection, it is considered that the road elements at this intersection have changed, and the road change information is the element information corresponding to the newly added traffic light in the road elements.

[0069] In some embodiments, the road change information may further include other information related to the change of road information, such as the update information of road information. By way of example only, the road change information may further include the change range of road information, the update interval of road information, the road change comparison information, etc. or any combination thereof. Among them, the change range of road information refers to the change amount of the currently extracted road elements compared with the corresponding road map elements in the map data. The update interval of road information refers to the time interval between the update time of the map elements of the corresponding road in the map data and the current time.

[0070] The road change comparison information refers to the change information of each road element of the actual road compared with each corresponding road map element in the map data. For example, the road change information may include the position of the changed road element, the size of the changed road element, the shape of the changed road element, etc. or any combination thereof.

[0071] In some embodiments, the processing device may determine the road change information in various ways based on the road elements and the map data. For example, the processing device may compare based on the road elements in the map data, and regard the elements newly added, missing or with changed appearance compared with the map elements as the road change information. Among them, the appearance change may include but is not limited to size change, shape change, etc.

[0072] In some embodiments, the road change information includes newly added traffic light information, and the process 200 may further include: obtaining road structure information; determining the binding relationship between the newly added traffic light and the road based on the road structure information and the newly added traffic light information.

[0073] The newly added traffic light information may include the type, structure, position, size, orientation, etc. of the newly installed traffic light. Among them, the traffic light type can be determined based on various methods. For example, based on the indication form of the traffic light, the traffic light type may include circular traffic lights, arrow traffic lights, and based on whether the traffic light has a countdown prompt, the traffic light type may include timed traffic lights, non-timed traffic lights, etc.

[0074] The structure of the traffic light can represent the distribution of each indicator light in the traffic light, such as the distribution of straight-ahead lights, left-turn lights, right-turn lights, etc.

[0075] In some embodiments, the processing device may obtain the newly added traffic light information in various ways. For example, the processing device may obtain the newly added traffic light information based on environmental data through an image recognition algorithm (such as, deep neural network, YOLO, etc.).

[0076] In some embodiments, the processing device may determine whether the traffic light element in the current geographical area of the road elements is consistent with the traffic light element in the map data. If there is a traffic light element at the same location in the road elements but no traffic light element in the map data, the processing device may determine that the road change information includes newly added traffic light information; and the processing device may determine the newly added traffic light information based on the type, structure, location, size, and orientation of the added traffic light.

[0077] Road structure information refers to information related to the architecture of the road on which the vehicle is currently traveling. For example, road structure information may include the geometric forms of multiple roads and the road connection relationships, the location information and lane connection relationships of the lanes on each road, lane information, etc. Among them, road structure information can be used to characterize the relationships between different roads and intersections, as well as the relationships between roads and lanes, roads and appendages, etc. The multiple roads corresponding to the road structure information may include the road on which the vehicle is currently traveling and the roads that have intersection points with the road on which the vehicle is currently traveling. Lane information is information related to lanes. For example, lane information may include the number of lanes and the traffic directions of the lanes.

[0078] Merely by way of example, road structure information may include that the road on which the vehicle is currently traveling forms a Y-shaped road with the first road and the second road, etc. The first road is the road located in the front left of the road on which the vehicle is currently traveling, and the second road is the road located in the front right of the road on which the vehicle is currently traveling. In some embodiments, the processing device may obtain road structure information in various ways. For example, the processing device may obtain the road structure information stored in a storage device (such as, memory 140). Also, for example, the processing device may obtain the road structure information provided by an information source through a network, and the information source may be a map database (such as, Amap, Google Maps, etc.), a traffic management bureau database, a road network database, etc.

[0079] In some embodiments, the processing device may obtain road structure information based on the video information collected by an imaging device (such as, a driving recorder) on the vehicle.

[0080] In some embodiments, the road structure information may include one or any combination of the current latest road structure information, the road structure information corresponding to a historical time, etc.

[0081] The binding relationship refers to the corresponding relationship between each newly added traffic light and each lane. That a certain newly added traffic light has a binding relationship with a certain road can be used to indicate that the newly added traffic light controls the vehicles on that road. Exemplarily, for a newly added traffic light in a certain orientation, if the newly added traffic light is an arrow traffic light, including a left-turn light and a straight-ahead light, the binding relationship includes the binding relationship between the left-turn light and the corresponding road, and the binding relationship between the straight-ahead light and the corresponding road.

[0082] In some embodiments, the binding relationship can be represented based on establishing a mapping relationship between the newly added traffic light and the road. For example, the processing device can create a mapping relationship between the identifier corresponding to the newly added traffic light and the identifier of the road to which it is bound. The identifier can refer to specific marker information used to distinguish different objects. In some embodiments, the identifier can include, but is not limited to, one or more of ID, digital coding, etc.

[0083] In some embodiments, the processing device can determine the binding relationship between the newly added traffic light and the road based on the newly added traffic light information and the road structure information according to a preset rule. Exemplarily, the preset rule is: determine a first vector based on the orientation of the newly added traffic light, determine a second vector based on the driving direction of each road, and make a judgment based on the angle between the first vector and the second vector. When the angle is within a preset range, it is determined that the newly added traffic light and the corresponding road have a binding relationship. Among them, the orientation of the newly added traffic light refers to the orientation of the traffic light arrow. The preset range can be a manually preset value, a system default value, etc.

[0084] In some embodiments, the processing device can determine the binding relationship based on a binding relationship model. The binding relationship model is a machine learning model. The input of the binding relationship model can include road structure information and newly added traffic light information, and the output can include the binding relationship. In some embodiments, the binding relationship model can be obtained by training based on a large number of second training samples with a second label in a manner similar to the training of the element extraction model. For more content about the training of the element extraction model, reference can be made to Figure 4 the relevant description.

[0085] In some embodiments, the second training samples can include multiple groups of training samples. Each group of training samples includes at least the sample road structure information where the sample vehicle is located and the sample newly added traffic light information. The second training samples can be obtained from historical data.

[0086] In some embodiments, the second label can be the binding relationship between the sample newly added traffic light and the sample road. The second label can be obtained by manual or processing device annotation. For example, the sample newly added traffic light and the sample road are annotated based on the control relationship to obtain the second label.

[0087] In some embodiments of this specification, by using the road structure information and the newly added traffic light information to determine the binding relationship between the newly added traffic light and the road, the road information corresponding to the road change information can be determined in a timely manner, which is beneficial to accurately determining the vehicle control strategy based on the road change information. At the same time, the map data can be updated based on the road change information, which can play an auxiliary role in making an accurate control strategy for other vehicles.

[0088] Step 240, input the road change information into the strategy determination model to determine the vehicle control strategy.

[0089] The control strategy is the relevant control instruction for controlling the vehicle when driving on a road with road change information. For example, the control strategy includes but is not limited to stopping, decelerating, accelerating, changing the moving direction of the vehicle, or reversing, etc., or any combination thereof.

[0090] In some embodiments, the policy determination model is a machine learning model. For example, the policy determination model may include a deep neural network model (Deep Neural Network, DNN), and the policy determination model may also be other neural networks, etc., or any combination thereof.

[0091] In some embodiments, the processing device can obtain the trained policy determination model from a storage device (such as, memory 140). In some embodiments, the policy determination model may include a Deep Deterministic Policy Gradient (DDPG) network model. The DDPG network model can be a reinforcement learning model, which can be used in machine learning related to continuous signal problems. By using the trained policy determination model, the multi-factor analysis process for making decisions can be bypassed, and then the control strategy can be determined to control the operation of the vehicle.

[0092] In some embodiments, the input of the policy determination model further includes at least one of: map data, the original control policy corresponding to the map data, and reference road change information. For more content about the policy determination model and its training, reference can be made to Figure 5 the relevant description.

[0093] In some embodiments, the processing device can guide the vehicle to drive based on the control strategy. For example, the processing device inputs the road change information into the policy determination model, determines the control strategy of the vehicle, generates corresponding control instructions based on the control strategy, and sends them to the control component of the vehicle to guide the vehicle to execute the driving actions corresponding to the control instructions. Among them, different control strategies correspond to different control instructions.

[0094] In some embodiments, the control instruction can refer to an instruction configured to control the operation of the vehicle. In some embodiments, the control instruction may include one or more control parameters. For example, control parameters related to the accelerator or braking device, steering control parameters, etc.

[0095] Merely by way of example, the control instruction can be expressed as S(X, Y), where X represents the control parameter related to the accelerator or braking device (for example, the opening degree), and Y represents the steering control parameter (for example, the steering wheel angle). As used herein, for the "opening degree", a positive value represents accelerator operation, a negative value represents braking device operation, and 0 represents maintaining the current speed; for the "steering wheel angle", a positive value represents "turning right", a negative value represents "turning left", and 0 represents maintaining the current direction.

[0096] In some embodiments of the present specification, different control strategies of a vehicle can be determined based on different road change information. Different types of driving actions can be controlled for the autonomous vehicle according to the road change information, so that the generated control strategies are more adaptable to the current actual road conditions. That is, in the case of actual road changes or no map, a control strategy with high accuracy can be obtained based on the road change information, which helps to improve the control efficiency of the autonomous vehicle, making the autonomous vehicle more adaptable to various mapless areas or roads with changes and enabling the control strategy to have a more accurate control effect, thus contributing to improving the driving safety.

[0097] It should be noted that the above description of process 200 is only for illustration and explanation, and does not limit the scope of application of the present specification. Those skilled in the art can make various corrections and changes to process 200 under the guidance of the present specification. However, these corrections and changes are still within the scope of the present specification.

[0098] Figure 3 It is an exemplary schematic diagram of extracting road elements shown in some embodiments of the present specification.

[0099] In some embodiments, the processing device can convert the environmental data 310 into the bird's-eye view space 320 to obtain the bird's-eye view data 330; based on the bird's-eye view data 330, the road elements 350 can be extracted through the element extraction model 340.

[0100] For more content about the environmental data, reference can be made to Figure 2 the relevant description.

[0101] The bird's-eye view refers to the perspective of looking down at the ground from a high altitude. The bird's-eye view space can include a spatial coordinate system for describing the coordinates of the bird's-eye view data in the bird's-eye view.

[0102] The bird's-eye view data refers to the three-dimensional data obtained by observing the environment of the geographical area where the vehicle is located from the bird's-eye view. For example, the bird's-eye view data can include the height information of objects in the environment (such as the height of obstacles, uphill and downhill height information), the longitude and latitude information of objects, etc.

[0103] In a vehicle, different sensors can be deployed, such as cameras for collecting image data or lidar for collecting point cloud data. During the process of autonomous driving, image data or point cloud data can be collected to assist in autonomous driving. In practical applications, the image data collected by cameras is two-dimensional image data, and the features in the two-dimensional image perspective have certain limitations. For example, the two-dimensional image data cannot reflect the spatial position of the features relative to the surrounding environment, which may affect some perception tasks during the process of autonomous driving, such as affecting the accuracy of semantic segmentation tasks and 3D object detection tasks. Therefore, it may be necessary to convert the two-dimensional environmental data into bird's-eye view data.

[0104] In some embodiments, the processing device can convert the environmental data into the bird's-eye view space in various ways to obtain the bird's-eye view data.

[0105] For example, the processing device can convert the environmental data into the bird's-eye view space based on inverse perspective transformation to obtain the bird's-eye view data.

[0106] In some embodiments, the processing device can obtain the bird's-eye view data based on the processing of the environmental data by the backbone network layer and the perspective transformation layer. For example, the processing device can input the environmental data into the backbone network layer and output multiple first feature maps. Among them, the first feature map is the feature data in the two-dimensional image perspective. The multiple first feature maps can be feature maps with the same or different sizes.

[0107] The backbone network layer is a model or algorithm for extracting image features. In some embodiments, the backbone network layer can include the Visual Geometry Group network (VGG), the Residual Network (ResNet), etc.

[0108] In some embodiments, the processing device can perform feature fusion on multiple first feature maps with the same or different sizes to obtain the image features corresponding to the two-dimensional image data, and input the image features into the perspective transformation layer to obtain the bird's-eye view data corresponding to the two-dimensional image data. Feature fusion refers to the process of integrating the first feature maps of the surrounding environment collected by different cameras into the same image. In some embodiments, feature fusion can be achieved in various ways, such as feature stitching, feature summation, element-wise multiplication, etc.

[0109] Among them, the perspective conversion layer is used to convert the feature data in the two-dimensional image perspective to the bird's-eye view space. In some embodiments, the perspective conversion layer may include BEVDET (Bounding Box Estimation using 3D Edgeboxes), BEVDet4D (Bounding Box Estimation using 4D Edge boxes), etc.

[0110] Although vehicles are equipped with different sensors, such as cameras, LiDARs, and radars, each sensor has its advantages and disadvantages. For example, camera data contains dense color and texture information but cannot capture depth information. LiDAR provides accurate depth and structural information but is limited by a limited range and sparsity. Millimeter-wave radar is sparser than LiDAR but has a longer sensing range and can capture information about moving objects. Therefore, in some embodiments of this specification, the data sensed by multiple sensors are fused so that the processing device can obtain more useful information and improve the accuracy and fault tolerance of sensing.

[0111] In some embodiments, the processing device can, based on the sensing data of each sensor, in the foregoing manner, sequentially pass through the backbone network layer and the perspective conversion layer to convert the sensing data of each sensor to the bird's-eye view space, and obtain the bird's-eye view data corresponding to each sensing data.

[0112] For example, the processing device can, based on the image data and the point cloud data, respectively obtain the first bird's-eye view data and the second bird's-eye view data in a similar manner as above, and perform cross-modal fusion on the first bird's-eye view data and the second bird's-eye view data to obtain the final bird's-eye view data. Among them, the first bird's-eye view data is the result of converting the image data to the bird's-eye view space. The second bird's-eye view data is the result of converting the point cloud data to the bird's-eye view space.

[0113] Cross-modal fusion is the process of integrating the information of data from different sensors. In some embodiments, the processing device can perform cross-modal fusion in various ways. For example, it can obtain the final bird's-eye view data through pre-fusion, deep fusion, and post-fusion methods.

[0114] In some embodiments, the element extraction model is a machine learning model. For example, the element extraction model may include a deep neural network model (Deep Neural Network, DNN), and the element extraction model may also be other neural networks, etc., or any combination thereof.

[0115] In some embodiments, the processing device can input the bird's-eye view data into the element extraction model and output road elements.

[0116] For more information about road elements, refer to Figure 2 the relevant description.

[0117] In some embodiments, the element extraction model can be trained based on a large number of first training samples with first labels through various feasible methods. For example, parameter updates can be performed based on the gradient descent method. An exemplary training process includes: inputting multiple first training samples with first labels into the initial element extraction model, constructing a loss function based on the first labels and the results of the initial element extraction model, and iteratively updating the parameters of the initial element extraction model based on the loss function through gradient descent or other methods. When the preset conditions are met, the model training is completed, and the trained element extraction model is obtained. Among them, the preset conditions can be the convergence of the loss function, the number of iterations reaching the threshold, etc.

[0118] In some embodiments, the first training samples can include multiple groups of training samples, and each group of training samples includes at least the sample bird's-eye view data corresponding to the sample environment data. The first training samples can be obtained based on historical data.

[0119] In some embodiments, the first label can be the sample road element corresponding to the sample environment data. For example, the actual road elements in the sample bird's-eye view data corresponding to the sample environment data are labeled to obtain the first label.

[0120] In some embodiments, the processing device can input the bird's-eye view data and map data into the element extraction model to output road elements. For more information about the element extraction model, refer to Figure 4 the relevant description.

[0121] In some embodiments, the road elements include current road elements and historical road elements, and the processing device can fuse the current road elements and historical road elements to generate four-dimensional road elements.

[0122] The current road element refers to the road element in the geographical area within the current time interval. The current time interval refers to a period of time range including the current time. The time interval can be determined based on experiments or experience. For example, the time interval can be 1 second, 5 seconds, 10 seconds, 1 minute, 10 minutes, 1 hour, one day, etc.

[0123] The historical road element refers to the road element in the geographical area within each past time interval.

[0124] In some embodiments, during the vehicle driving process, the sensor can continuously collect continuous frame data containing environment data. The current road element refers to the road element extracted according to the environment data of the current frame. The historical road element can refer to the road element extracted according to the environment data of the previous frame or the previous few frames of the current frame.

[0125] The four-dimensional road element is used to describe the change of the road element over time. For example, the four-dimensional road element of a certain geographical area can be represented by a vector as [(T1, D1), (T2, D2),...], where T1 represents the first time interval, and D1 represents the road element corresponding to the geographical area in the first time interval T1; T2 represents the second time interval, and D2 represents the road element corresponding to the geographical area in the second time interval T2.

[0126] In some embodiments, the processing device can perform statistical analysis based on the current road element and the historical road element to determine the change of the road element over time and obtain the four-dimensional road element.

[0127] In some embodiments of this specification, by performing time-series fusion on the current road element and the historical road element to determine the four-dimensional road element, the road change information can be determined by combining the spatial dimension and the time dimension, improving the accuracy of determining the road change information. At the same time, by perceiving the influence of time on the obstacles on the road, the monitoring ability of the vehicle for obstacles (such as pedestrians, vehicles, etc.) is improved, which is conducive to obtaining a more accurate control strategy, improving the accuracy of autonomous vehicle control, and helping to enhance road driving safety.

[0128] In some embodiments of this specification, by converting the environmental data into the bird's-eye view space to obtain three-dimensional bird's-eye view data, the information about the orientation and distance of the obstacle relative to the vehicle can be obtained, enabling the vehicle to not only perceive the relative orientation of the obstacle in the environment but also perceive the distance from itself, thus achieving flexible obstacle avoidance.

[0129] Figure 4 It is an exemplary schematic diagram of an element extraction model shown in some embodiments of this specification.

[0130] In some embodiments, the processing device can further extract road elements based on the map data through the element extraction model.

[0131] In some embodiments, as Figure 4 shown, the input of the element extraction model 340 can include the bird's-eye view data 330 and the map data 420, and the output can include the road element 350.

[0132] For more content about the map data, the bird's-eye view data, and the road element, reference can be made to the relevant description in Figure 2 . In some embodiments, the bird's-eye view data 330 and the map data 420 input into the element extraction model 340, and the road element 350 output by the element extraction model 340 can be data for the same location area.

[0133] In some embodiments of the present specification, by inputting map data as auxiliary information into the element extraction model, the perception effect of occluded areas in the environment can be enhanced, and the accuracy of the road elements output by the element extraction model can be improved.

[0134] In some embodiments, the element extraction model can be trained based on a large number of second training samples with second labels through various feasible methods. For example, parameter updates can be performed based on the gradient descent method. The specific training process can refer to the description of training the element extraction model based on the first training samples with first labels mentioned above.

[0135] In some embodiments, the second training samples can include multiple groups of training samples, and each group of second training samples includes at least the sample map data corresponding to the sample environmental data and the sample bird's-eye view data. The second training samples can be obtained based on historical data.

[0136] In some embodiments, the second label can be the sample road elements corresponding to the sample environmental data. For example, by combining the map data, the actual road elements in the sample bird's-eye view data corresponding to the sample environmental data are labeled to obtain the second label.

[0137] In some embodiments of the present specification, the road elements can be obtained efficiently and accurately through the element extraction model, which is beneficial to subsequent determination of road change information and improvement of the accuracy of vehicle control.

[0138] Figure 5 It is an exemplary schematic diagram of the policy determination model shown according to some embodiments of the present specification.

[0139] In some embodiments, as Figure 5 shown, the input of the policy determination model includes at least one of the road change information 510, the map data 520, the original control policy 530 corresponding to the map data, the reference road change information 540, and the reference control policy 580. The output of the policy determination model 550 can include the control policy 560 of the vehicle. For more content about the map data, reference can be made to Figure 2 the relevant description.

[0140] Among them, the original control policy corresponding to the map data refers to the control policy of the vehicle before the road change information is perceived. In some embodiments, the processing device can obtain the original control policy of the vehicle corresponding to the map data from a storage device (such as, the memory 140). In some embodiments, the original control policy can also be read through an interface, and the interface includes but is not limited to a program interface, a data interface, a transmission interface, etc. In some embodiments, the original control policy can also be obtained in any manner well-known to those skilled in the art, and the present specification does not limit this.

[0141] The reference road change information refers to the road change information determined by other vehicles. In some embodiments, the processing device can communicate with other vehicles through a network to obtain the reference road change information determined by other vehicles.

[0142] The reference control strategy 580 is a control strategy used by the vehicle for reference.

[0143] The reference control strategy 580 can be obtained in various ways. In some embodiments, other vehicles can determine control strategies according to their respective corresponding road change information. For example, other vehicles can determine control strategies based on the method of manual takeover or autonomous control. Other vehicles can upload the control strategies to a cloud server (such as server 110), and the cloud server integrates all the control strategies and distributes them to each vehicle as the reference control strategy.

[0144] In some embodiments, the policy determination model can be trained based on a large number of sample data with labeled data through various feasible methods. For example, parameter updates can be based on the gradient descent method. In some embodiments, the initial model can be trained in a manner similar to the training of the element extraction model to obtain the policy determination model. For more content about the training of the element extraction model, reference can be made to Figure 4 the relevant description.

[0145] In some embodiments, the training data of the policy determination model can include multiple groups of training samples, and each group of training samples includes at least sample road change information. The training data can be obtained based on historical data.

[0146] In some embodiments, when the input of the policy determination model includes at least one of the road change information 510, map data 520, the original control policy 530 corresponding to the map data, reference road change information 540, and reference control strategy 580, each group of training samples includes at least one of the sample road change information of the sample road, sample map data, sample original control policy corresponding to the sample map data, sample reference road change information, and sample reference control strategy.

[0147] In some embodiments, the labeled data can be the actual control policy corresponding to when the sample vehicle successfully passes the sample road.

[0148] In some embodiments of this specification, the control strategy can be efficiently and accurately obtained through the policy control model, which is beneficial to accurately controlling the vehicle to perform different driving operations for different road change information and improving the accuracy of vehicle control.

[0149] In some embodiments, such as Figure 5As shown, the training data of the policy determination model 550 includes first training data 573. The first training data 573 corresponds to a first sample vehicle 571. The movement control of the first sample vehicle 571 includes autonomous control. The first training data 573 includes: sample road change information 573-1 of the first sample road; a first sample control strategy 573-2 for manually taking over the first sample vehicle 571 to make the first sample vehicle 571 pass through the first sample road.

[0150] The first training data 573 is used to train an initial model 575 to obtain at least a partial sample set of the policy determination model. In some embodiments, the processing device may obtain the initial model from a storage device (such as, memory 140) and / or an external data source via a network. The initial parameters of the initial model may be based on the default settings of the processing device.

[0151] The first sample vehicle 571 refers to a sample terminal that can switch between autonomous control and manual takeover. In some embodiments, the first sample vehicle 571 may be a vehicle with the same or different terminal attributes as the vehicle. The terminal attributes may include one or more of terminal model, service life, terminal weight, engine model, etc. For example, the first sample vehicle 571 may be a vehicle with the same model as the vehicle.

[0152] In some embodiments, the movement control of the first sample mobile terminal includes modes such as full autonomous control and semi-autonomous control.

[0153] Autonomous control means that the vehicle automatically detects and / or receives environmental data, processes the environmental data to independently determine a control strategy, and executes the control strategy without driver manipulation to control the realization of automatic driving. For example, the vehicle can receive map data and route information from a positioning device, detect the traffic conditions and weather conditions around itself, and determine and independently execute a control strategy without driver manipulation. Among them, the driver's manipulation may include the driver's control of the steering wheel, accelerator, brake pedal, etc.

[0154] Manual takeover means a control mode in which the vehicle travels under the manual manipulation of the driver. In some embodiments, manual takeover can also be implemented by a server in the cloud (such as, server 110) in a remote control manner.

[0155] The first sample road refers to the road on which the first sample vehicle 571 travels when collecting the first training data 573 based on the first sample vehicle 571. In some embodiments, the first sample road may be an actual road or a virtual road generated by software.

[0156] The sample road change information 573-1 of the first sample road refers to the information related to the change of the road elements of the first sample road collected based on the first sample vehicle 571.

[0157] The first sample control strategy 573-2 is the control strategy adopted when the first sample vehicle 571 travels to the first sample road with sample road change information and is controlled manually.

[0158] In some embodiments, the first training data 573 can be obtained in various ways. In some embodiments, the processing device can obtain a large amount of first historical driving data of the first sample vehicle 571 from a third-party platform, and determine the sample road change information 573-1 of the first sample road and the first sample control strategy 573-2. The first historical driving data includes the environmental data of the first sample road collected by the first sample vehicle 571, the sample road change information 573-1 of the first sample road determined based on the environmental data, and the actual control strategy for manually taking over and driving the first sample vehicle through the first sample road.

[0159] Among them, the environmental data can be an image sequence collected when the first sample vehicle 571 travels to the first sample road. See the corresponding description in Figure 2 for the sample road change information determined based on the environmental data.

[0160] The actual control strategy includes one or more control parameters. For example, control parameters related to the accelerator or braking device, steering control parameters, etc.

[0161] Exemplarily, assuming that the sample road change information 573-1 of the first sample road indicates that there is an obstacle in front of the sample vehicle, the driver can control the first sample vehicle 571 to decelerate and steer to avoid the obstacle after taking over the first sample vehicle 571, and the processing device can determine the first sample control strategy 573-2 based on the operation signal of the driver on the first sample vehicle 571.

[0162] In some embodiments, the first training data further includes at least one of map data, original control strategy, reference road change information, and reference control strategy.

[0163] In some embodiments, when the input of the policy determination model includes at least one of road change information 510, map data 520, the original control policy 530 corresponding to the map data, reference road change information 540, and reference control policy 580, the first training data 573 may further include at least one of the sample map data when the first sample vehicle 571 travels to the first sample road with sample road change information, the sample original control policy corresponding to the sample map data, sample reference road change information, and sample reference control policy.

[0164] In some embodiments, the first sample control policy 573-2 when the first sample vehicle 571 passes through the first sample road in the first training data 573 may be used as a sample label, and the remaining data in the first training data 573 may be used as training samples. The initial model 575 is trained based on the training samples and the sample label to obtain the policy determination model 550. For the description of training the initial model based on the training samples and the sample label, refer to the foregoing description of training the element extraction model based on the first training samples with the first label.

[0165] In some embodiments of the present specification, by obtaining the first training data, it is beneficial to learn the correlation between the sample road change information of the first sample road and the control policy of the first sample vehicle, can enrich the set of training samples, improve the training efficiency of the initial model, and improve the accuracy of the control policy output by the policy determination model.

[0166] In some embodiments, as Figure 5 shown, the training data of the policy determination model 550 includes second training data 574, which corresponds to the second sample vehicle 572. The movement control of the second sample vehicle 572 is pure manual control. The second training data 574 includes: sample road information 574-1 of the second sample road; the second sample control policy 574-2 for manually controlling the second sample vehicle 572 to pass through the second sample road.

[0167] The second training data 574 is used to train the initial model 575 to obtain another part of the sample set of the policy determination model.

[0168] The second sample vehicle 572 refers to a sample terminal controlled purely manually. In some embodiments, the second sample vehicle 572 may be a vehicle with the same or different terminal attributes as the vehicle. For example, the second sample vehicle 572 may be a vehicle with the same external dimensions as the vehicle.

[0169] Pure manual control means a control mode in which the vehicle travels completely under the control of the driver. For example, the driver can independently control the vehicle's steering wheel, accelerator, brake pedal, etc.

[0170] The second sample road refers to the road on which the second sample vehicle 572 travels when collecting the second training data 574. In some embodiments, the second sample road can be an actual road or a virtual road generated by software.

[0171] The sample road change information 574-1 of the second sample road refers to the information related to the change of the road elements of the second sample road collected based on the second sample vehicle 572.

[0172] The second sample control strategy 574-2 is a control strategy for controlling the second sample vehicle 572 in a fully manual driving manner when the second sample vehicle 572 travels to the second sample road with sample road change information.

[0173] In some embodiments, the second training data 574 can be obtained in various ways. In some embodiments, the processing device can obtain a large amount of second historical driving data of the second sample vehicle 572 from a third-party platform, and determine the sample road change information 574-1 and the second sample control strategy 574-2 of the second sample road. The second historical driving data includes the environmental data of the second sample road collected by the second sample vehicle 572, the sample road change information 574-1 of the second sample road determined based on the environmental data, and the actual control strategy for the second sample vehicle 572 when the second sample vehicle is manually driven through the second sample road.

[0174] Among them, the environmental data can be an image sequence collected when the second sample vehicle 572 travels to the second sample road. For the sample road change information determined based on the environmental data, see Figure 2 the corresponding description in

[0175] Exemplarily, assuming that the sample road change information 574-1 of the second sample road indicates that the road ahead of the second sample vehicle 572 is blocked, the driver can control the second sample vehicle 572 to make a U-turn, and the processing device can determine the second sample control strategy 574-2 based on the operation signal of the driver for the second sample vehicle 572.

[0176] In some embodiments, when the input of the policy determination model 550 includes at least one of the road change information 510, the map data 520, the original control policy 530 corresponding to the map data, the reference road change information 540, and the reference control policy 580, the second training data 574 can also include at least one of the sample map data, the sample original control policy corresponding to the sample map data, the sample reference road change information, and the sample reference control strategy when the second sample vehicle 572 travels to the second sample road with sample road change information.

[0177] In some embodiments, the second sample control strategy 574-2 when the second sample vehicle 572 passes through the second sample road in the second training data 574 can be used as the sample label, and the remaining data in the second training data 574 can be used as the training samples. The initial model 575 is trained based on the training samples and the sample label to obtain the policy determination model 550. For the description of training the initial model based on the training samples and the sample label, refer to the aforementioned description of training the element extraction model based on the first training samples with the first label.

[0178] In some embodiments of the present specification, by obtaining the second training data, it is beneficial to learn the correlation between the sample road change information of the second sample road and the control strategy of the second sample vehicle, which can enrich the set of training samples, improve the training efficiency of the initial model, and improve the accuracy of the control strategy output by the policy determination model.

[0179] In some embodiments, the processing device can also use the first training data and the second training data as the training data for two training stages of the initial model 575, and use the initial model after completing the two-stage training as the final policy determination model. The policy determination model obtained through two-stage training can be applicable to more driving scenarios and can more flexibly determine more accurate control strategies.

[0180] Figure 6 It is a schematic structural diagram of a vehicle control system 600 shown according to some embodiments of the present specification.

[0181] As Figure 6 shown, the vehicle control system 600 may include an acquisition module 610, an extraction module 620, a first determination module 630, and a second determination module 640.

[0182] The acquisition module 610 is configured to acquire the environmental data collected by the sensors of the vehicle.

[0183] The extraction module 620 is configured to extract road elements according to the environmental data.

[0184] In some embodiments, the extraction module 620 is further configured to: convert the environmental data to the bird's-eye view space to obtain the bird's-eye view data; based on the bird's-eye view data, extract road elements through the element extraction model; the element extraction model is a machine learning model.

[0185] In some embodiments, the extraction module 620 is further configured to: further extract road elements through the element extraction model based on the map data.

[0186] In some embodiments, the road elements may include current road elements and historical road elements, and the extraction module 620 is further configured to: fuse the current road elements and the historical road elements to generate four-dimensional road elements.

[0187] The first determination module 630 is configured to determine road change information based on the road elements and the map data.

[0188] The second determination module 640 is configured to input the road change information into a policy determination model to determine the control policy of the vehicle, and the policy determination model is a machine learning model.

[0189] In some embodiments, the road change information includes new traffic light information, and the vehicle control system 600 may further include a binding module (not shown in the figure). The binding module is configured to: obtain road structure information; determine the binding relationship between the new traffic light and the road based on the road structure information and the new traffic light information.

[0190] In some embodiments, the input of the policy determination model further includes at least one of: map data, the original control policy corresponding to the map data, and reference road change information; the reference road change information is obtained based on other vehicles.

[0191] In some embodiments, the training data of the policy determination model includes first training data, which corresponds to a first sample vehicle. The movement control of the first sample vehicle includes autonomous control. The first training data includes: sample road change information of a first sample road; a first sample control policy for manually taking over the first sample vehicle to make the first sample vehicle pass through the first sample road.

[0192] In some embodiments, the training data of the policy determination model includes second training data, which corresponds to a second sample vehicle. The movement control of the second sample vehicle is pure manual control. The second training data includes: sample road information of a second sample road; a second sample control policy for manually controlling the second sample vehicle to pass through the second sample road.

[0193] For more descriptions of the acquisition module 610, the extraction module 620, the first determination module 630, and the second determination module 640, reference may be made to the relevant descriptions above.

[0194] One or more embodiments of this specification also provide a vehicle control device, including a processing device, and the processing device is used to execute a vehicle control method as described in any of the above embodiments.

[0195] One or more embodiments of this specification also provide a computer-readable storage medium, and the storage medium stores computer instructions. When the computer reads the computer instructions in the storage medium, the computer runs a vehicle control method as described in any of the above embodiments.

[0196] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only an example and does not constitute a limitation to this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification, so such modifications, improvements, and corrections still fall within the spirit and scope of the exemplary embodiments of this specification.

[0197] Meanwhile, this specification uses specific terms to describe the embodiments of this specification. Such as "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that the "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more at different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this specification can be appropriately combined.

[0198] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numerical letters, or the use of other names in this specification is not used to limit the order of the processes and methods in this specification. Although some currently considered useful invention embodiments are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration. The appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only through software solutions, such as installing the described system on existing servers or mobile devices.

[0199] Similarly, it should be noted that, in order to simplify the expression of the disclosure of this specification and thus help the understanding of one or more invention embodiments, in the previous description of the embodiments of this specification, sometimes multiple features are merged into one embodiment, drawing, or its description. However, this disclosure method does not mean that the features required by the object of this specification are more than those mentioned in the claims. In fact, the features of the embodiments are fewer than all the features of the individual embodiments disclosed above.

[0200] In some embodiments, numbers are used to describe components and the quantity of attributes. It should be understood that such numbers used in the description of embodiments are, in some examples, modified by the modifiers "about", "approximate" or "substantially". Unless otherwise stated, "about", "approximate" or "substantially" indicate that the said numbers allow a variation of ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may vary according to the characteristics required by individual embodiments. In some embodiments, the numerical parameters should consider the specified significant digits and adopt the method of retaining the general number of digits. Although the numerical ranges and parameters used in some embodiments of this specification to confirm the breadth of their scope are approximate values, in specific embodiments, such numerical settings are as precise as possible within the feasible range.

[0201] For each patent, patent application, patent application publication and other materials cited in this specification, such as articles, books, specifications, publications, documents, etc., their entire contents are hereby incorporated into this specification by reference. Except for the application history documents that are inconsistent with or conflict with the content of this specification, and also except for the documents that limit the broadest scope of the claims of this specification (currently or subsequently appended to this specification). It should be noted that if there are inconsistencies or conflicts between the descriptions, definitions, and / or uses of terms in the supplementary materials of this specification and the content described in this specification, the descriptions, definitions, and / or uses of terms in this specification shall prevail.

[0202] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered to be consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A vehicle control method, executed by the vehicle, comprising: Obtaining environmental data collected by sensors of the vehicle; Extracting road elements according to the environmental data; Determining road change information based on the road elements and map data; Inputting the road change information into a policy determination model to determine the control policy of the vehicle, where the policy determination model is a machine learning model.

2. The method according to claim 1, wherein the extracting road elements according to the environmental data comprises: Converting the environmental data into a bird's-eye view space to obtain bird's-eye view data; Extracting the road elements based on the bird's-eye view data through an element extraction model; The element extraction model is a machine learning model.

3. The method according to claim 2, wherein the extracting road elements according to the environmental data comprises: Further extracting the road elements based on the map data through the element extraction model.

4. The method according to claim 2, wherein the road elements comprise current road elements and historical road elements; The extracting road elements according to the environmental data further comprises: Fusing the current road elements and the historical road elements to generate four-dimensional road elements.

5. The method according to claim 1, wherein the road change information comprises newly added traffic light information; The method further comprises: Obtaining road structure information; Determining the binding relationship between the newly added traffic light and the road based on the road structure information and the newly added traffic light information.

6. According to the method described in claim 1, the input of the policy determination model further includes: At least one of the map data, the original control policy corresponding to the map data, reference road change information, and reference control policy; The reference road change information and the reference control policy are obtained based on other vehicles.

7. The method according to claim 1, wherein the training data of the policy determination model comprises first training data, the first training data corresponds to a first sample vehicle, the movement control of the first sample vehicle comprises autonomous control, and the first training data comprises: Sample road change information of a first sample road; A first sample control policy for manually taking over the first sample vehicle to make the first sample vehicle pass through the first sample road.

8. The method according to claim 1, wherein the training data of the policy determination model comprises second training data, the second training data corresponds to a second sample vehicle, the movement control of the second sample vehicle is pure manual control, and the second training data comprises: Sample road change information of a second sample road; A second sample control policy for manually controlling the second sample vehicle to pass through the second sample road.

9. A vehicle control system, comprising: An acquisition module configured to acquire environmental data collected by sensors of the vehicle; An extraction module configured to extract road elements according to the environmental data; A first determination module configured to determine road change information based on the road elements and map data; A second determination module configured to input the road change information into a policy determination model to determine the control policy of the vehicle, where the policy determination model is a machine learning model.

10. A computer-readable storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the vehicle control method according to any one of claims 1-8.

Images