Automatic driving map updating method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide an automatic driving map updating method and device, electronic equipment and storage medium, relating to the technical field of map updating. The method generates a normalized map updating task set according to the confidence of each grid in the automatic driving map, publishes the normalized map updating task set to multiple opportunistic vehicles, receives the cost quotes of each opportunistic vehicle for a single grid updating task, selects a winning vehicle set from the opportunistic vehicles based on the system utility of each opportunistic vehicle, receives the perception data uploaded by the winning vehicles in the winning vehicle set, and updates the map data of the corresponding grid. Thus, the automatic driving map can be updated at a relatively small cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of map updating, in particular to an automatic driving map updating method and device, electronic equipment and a storage medium. BACKGROUND

[0002] In recent years, the continuous progress of artificial intelligence technology has strongly driven the development of automatic driving technology, making it the core engine of the transformation and upgrading of the automobile industry. At present, China is in a key period of transition from assisted driving to conditional automatic driving, and has begun to layout the research and application of high-level automatic driving technology, with extremely broad market and technology prospects.

[0003] In this development process, the construction and updating of high-precision automatic driving maps are essential and unavoidable basic links. However, the existing map updating method relying on professional collection vehicles not only has high cost, but also causes deviation between map data and the real environment, which brings potential risks to the automatic driving system. SUMMARY

[0004] Therefore, the purpose of the embodiments of the present application is to provide an automatic driving map updating method, device, electronic equipment and storage medium to at least partially improve the above problems.

[0005] In order to achieve the above purpose, the technical solutions adopted by the embodiments of the present application are as follows:

[0006] In a first aspect, the embodiments of the present application provide an automatic driving map updating method, comprising:

[0007] generating a normalized map updating task set according to the confidence of each grid in the automatic driving map; the normalized map updating task set includes at least one grid updating task that needs to be updated;

[0008] publishing the normalized map updating task set to a plurality of opportunistic vehicles, and receiving the cost bid of each opportunistic vehicle for a single grid updating task;

[0009] selecting a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle; wherein the system utility is the ratio of the number of grid updating tasks that the opportunistic vehicle can complete to the cost bid;

[0010] receiving the perception data uploaded by the winning vehicles in the set of winning vehicles, and updating the map data of the corresponding grid.

[0011] Optionally, the step of selecting a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle comprises:

[0012] selecting an initial set of winning vehicles from the opportunistic vehicles based on system utility of each of the opportunistic vehicles using a greedy algorithm;

[0013] optimizing the initial set of winning vehicles using a variable neighborhood search algorithm to obtain a globally optimal set of winning vehicles.

[0014] Optionally, the selecting an initial set of winning vehicles from the opportunistic vehicles based on system utility of each of the opportunistic vehicles using a greedy algorithm comprises:

[0015] initializing the initial set of winning vehicles as empty, the set of tasks to be completed as the set of regular map update tasks, and the set of available vehicles as all the opportunistic vehicles;

[0016] calculating system utility of each of the opportunistic vehicles in the set of available vehicles;

[0017] selecting an opportunistic vehicle with the highest system utility from the set of available vehicles as a target vehicle, adding the target vehicle to the initial set of winning vehicles, and removing the target vehicle from the set of available vehicles;

[0018] adding one to the cumulative coverage count of each grid update task covered by the target vehicle, and removing any grid update task from the set of tasks to be completed if the cumulative coverage count of the grid update task plus its original confidence reaches a preset threshold;

[0019] determining whether the set of tasks to be completed is empty or the set of available vehicles is empty, and if not, returning to the step of calculating system utility of each of the opportunistic vehicles in the set of available vehicles; otherwise, outputting the current initial set of winning vehicles.

[0020] Optionally, the optimizing the initial set of winning vehicles using a variable neighborhood search algorithm to obtain a globally optimal set of winning vehicles comprises:

[0021] setting d neighborhood structures, and letting a current neighborhood index j = 1;

[0022] performing local search on the initial set of winning vehicles based on the jth neighborhood structure, updating the initial set of winning vehicles to a better solution if a better solution is found, and resetting the neighborhood index j = 1, otherwise, adding one to the neighborhood index j;

[0023] determining whether the neighborhood index j is less than or equal to d, and if so, performing the step of performing local search on the initial set of winning vehicles based on the jth neighborhood structure, otherwise, outputting the initial set of winning vehicles as the final set of winning vehicles.

[0024] Optionally, the method further comprises a step of calculating the reward income of the vehicles in the set of winning vehicles, which comprises:

[0025] calculating a first total number of grid update tasks for all winning vehicles in the set of winning vehicles;

[0026] removing one of the winning vehicles from the set of opportunistic vehicles, and selecting a second set of winning vehicles from the set of opportunistic vehicles based on the system utility of the remaining opportunistic vehicles;

[0027] calculating a second total number of grid update tasks for all winning vehicles in the second set of winning vehicles;

[0028] calculating the reward income of the winning vehicle removed from the set of opportunistic vehicles based on the difference between the first total number and the second total number.

[0029] Optionally, the method further comprises:

[0030] generating a set of emergency map update tasks in response to the received emergency event information, and sending the set of emergency map update tasks to each participating vehicle;

[0031] each participating vehicle independently generates a scheduling action based on the set of emergency map update tasks through the trained distributed decision network to obtain a scheduled vehicle;

[0032] receiving the perception data uploaded by each scheduled vehicle and updating the map data of the corresponding grid.

[0033] Optionally, the method further comprises a training step of the distributed decision network, which comprises:

[0034] each distributed decision network generates a first path planning decision of itself according to the location of the participating vehicle in which it is located, the set of emergency map update tasks, and the set of normal map update tasks;

[0035] each participating vehicle updates the location of the participating vehicle, the set of emergency map update tasks, and the set of normal map update tasks after executing the first path planning decision of itself;

[0036] generating a new second path planning decision of itself according to the updated location of the participating vehicle, the set of emergency map update tasks, and the set of normal map update tasks;

[0037] According to the positions of the respective participatory vehicles, the values before and after the set of sudden map update tasks and the set of normalized map update tasks are updated, and the first path planning decision and the second path planning decision, loss information of a value network is calculated, and the value network is updated.

[0038] Each distributed decision network is updated using the gradient provided by the value network.

[0039] In a second aspect, an embodiment of the present application provides an automatic driving map updating device, comprising:

[0040] A task generation unit is configured to generate a set of normalized map update tasks according to the confidence of each grid in the automatic driving map, wherein the set of normalized map update tasks comprises at least one grid update task that needs to be updated.

[0041] A task publishing unit is configured to publish the set of normalized map update tasks to a plurality of opportunistic vehicles, and receive cost quotes of each of the opportunistic vehicles for a single grid update task.

[0042] A vehicle selection unit is configured to select a set of winning vehicles from the opportunistic vehicles based on the system utility of each of the opportunistic vehicles, wherein the system utility is a ratio of the number of grid update tasks that can be completed by the opportunistic vehicle to the cost quote.

[0043] A map updating unit is configured to receive perception data uploaded by the winning vehicles in the set of winning vehicles, and update map data of the corresponding grid.

[0044] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the method of any one of the above aspects when executing the computer program.

[0045] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method of any one of the above aspects.

[0046] The automatic driving map updating method, device, electronic device and storage medium provided by the embodiments of the present application can publish map update tasks to a plurality of opportunistic vehicles, receive cost quotes of the vehicles for a single task, and select an optimal vehicle set based on system utility. Thus, the high cost of relying on professional collection vehicles is avoided, and the incidental collection capability of ordinary vehicles is utilized to complete the map update task at a lower cost, thereby significantly reducing the overall map maintenance cost.

[0047] In order to make the above objectives, characteristics and advantages of the present application more apparent, more comprehensible, the following preferred embodiments are specifically described in detail below, and the accompanying drawings are referred to, as follows. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without paying creative labor on the basis of these drawings.

[0049] Figure 1 A schematic structural block diagram of an electronic device provided by the embodiments of the present application is shown in the figure.

[0050] Figure 2 A flowchart of an automatic driving map updating method provided by the embodiments of the present application is shown in the figure.

[0051] Figure 3 Another flowchart of an automatic driving map updating method provided by the embodiments of the present application is shown in the figure.

[0052] Figure 4 A flowchart of step S231 provided by the embodiments of the present application is shown in the figure.

[0053] Figure 5 A flowchart of a reward income calculation provided by the embodiments of the present application is shown in the figure.

[0054] Figure 6 Another flowchart of an automatic driving map updating method provided by the embodiments of the present application is shown in the figure.

[0055] Figure 7 A schematic structural block diagram of an automatic driving map updating device provided by the embodiments of the present application is shown in the figure.

[0056] Icon: 100-electronic device; 101-memory; 102-communication interface; 103-processor; 104-communication bus; 600-automatic driving map updating device; 610-task generation unit; 620-task issuing unit; 630-vehicle selection unit; 640-map updating unit. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more apparent, the following will combine the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0058] Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the application.

[0059] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0060] It should be noted that, in this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the sentence "including a…" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0061] With the rapid development of automatic driving technology, high-precision maps have become an important basis for realizing environmental perception, path planning and automatic driving decision. The accuracy and real-time performance of map data directly affect the safety and reliability of the automatic driving system. However, the traditional map updating method mainly relies on professional collection vehicles for periodic road data collection and updating. These professional vehicles are usually equipped with high-precision laser radars, cameras and positioning systems, and can obtain high-quality map data. However, the deployment and operation cost of professional collection vehicles is high.

[0062] Based on the above situation, the embodiments of the application provide an automatic driving map updating method and device, electronic equipment and storage medium, which publish map updating tasks to a plurality of opportunistic vehicles, receive cost quotes of the vehicles for a single task, and select an optimal vehicle set based on system utility. Thus, the high cost of relying on professional collection vehicles is avoided, and the opportunistic collection capability of ordinary vehicles is utilized to complete the map updating task at a lower cost, thereby significantly reducing the overall map maintenance cost.

[0063] To implement the flow steps and functions of various examples of the present application, please refer to Figure 1 , Figure 1 An exemplary structural block diagram of an electronic device is provided for the embodiments of the present application. The electronic device 100 includes a memory 101 and a processor 103, which are directly or indirectly electrically connected to each other to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more communication buses 104 or signal lines. The memory 101 can be used to store software programs and modules, and the processor 103 executes the software programs and modules stored in the memory 101, thereby performing various functional applications and data processing.

[0064] The electronic device 100 can be, but is not limited to, a personal computer (PC), a server, a distributed computer, etc. It can be understood that the electronic device 100 is also not limited to a physical server, but can also be a virtual machine on a physical server, a virtual machine based on a cloud platform, etc. A computer that can provide the same functions as the server or virtual machine. The operating system of the electronic device 100 can be, but is not limited to, a Windows system, a Linux system, etc.

[0065] Among them, the memory 101 can be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0066] The communication connection between the electronic device 100 and the external device is realized through at least one communication interface 102 (which can be wired or wireless).

[0067] The processor 103 can be an integrated circuit chip with processing capability. In implementation process, each step of the embodiments of the present application can be completed by integrated logic circuit of hardware in the processor 103 or by instructions in the form of software. The processor 103 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; also can be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0068] It can be understood that, Figure 1 The structure shown is only schematic, and the electronic device 100 can further include more or less components than those shown, or have a different configuration of components than those shown. Figure 1 The components shown in the figures can be implemented in hardware, software, or a combination thereof. Figure 1 The components shown in the figures can be implemented in hardware, software, or a combination thereof. Figure 1 The components shown in the figures can be implemented in hardware, software, or a combination thereof.

[0069] The automatic driving map updating method provided by the present application will be described below. Referring to Figure 2 , the method execution subject can be the electronic device 100 shown in the above Figure 1 , which can be a cloud platform, and the method includes the following steps as shown in Figure 2 .

[0070] S210: generating a set of normalized map updating tasks according to the confidence of each grid in the automatic driving map; the set of normalized map updating tasks includes at least one grid updating task that needs to be updated.

[0071] S220: issuing the set of normalized map updating tasks to a plurality of opportunistic vehicles, and receiving cost quotes of each opportunistic vehicle for a single grid updating task.

[0072] Among them, the opportunistic vehicle, that is, the vehicle that keeps its original trajectory unchanged when participating in crowd sensing, completes the map updating task along the way in the form of opportunism, can complete the normalized map updating task.

[0073] S230: selecting a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle; wherein the system utility is the ratio of the number of grid updating tasks that can be completed by the opportunistic vehicle to the cost quote.

[0074] S240: receiving the perception data uploaded by the winning vehicles in the winning vehicle set, and updating the map data of the corresponding grid.

[0075] In the autonomous driving map, a large number of grids are divided, and each grid is associated with a confidence degree. Since the confidence degree of each grid of the autonomous driving map is in a state of decay over time, the decay amount of the map confidence degree is fixed as λ in a time period (e.g., one day), if there are λ vehicles going to the grid Collecting data, the confidence degree ρ of the grid remains unchanged, if the number of vehicles m passing through the grid is less than λ, the confidence degree is marked as (ρ+m-λ).

[0076] The time interval for generating the set of normalized map update tasks can be the above-mentioned time period (e.g., one day), at which time, the set of normalized map update tasks needs to be generated according to the confidence degrees of the grids in the autonomous driving map. For example, the grids with confidence degrees less than or equal to ρ-λ need to be updated, and these grids are added to the set of normalized map update tasks . Each grid corresponds to a grid update task.

[0077] At this time, the set of normalized map update tasks is published to each opportunistic vehicle , and each opportunistic vehicle sends a cost bid for a single grid update task to the cloud platform .

[0078] After receiving the cost bids of each opportunistic vehicle , the system utility of each is calculated, which is the ratio of the number of grid update tasks that the opportunistic vehicle can complete to the cost bid. It can be represented by the following formula:

[0079]

[0080] wherein is the set of grids that the vehicle can update, i.e. the grids that can be collected in the process of normal driving. is the number of map update tasks completed by the vehicle , i.e. the intersection of the published and the grids that can be updated.

[0081] According to the system utility of each ​the system utility, a set of winning vehicles can be selected Since it is preferred to use lower cost to complete more tasks as much as possible, the vehicles with the highest system utility in turn can be added to the initial set of winning vehicles until the tasks are completed.

[0082] After the set of winning vehicles is selected, the information is informed to the corresponding winning vehicles, and each winning vehicle collects corresponding grid perception data during driving and uploads the perception data to the cloud platform.

[0083] After receiving the perception data uploaded by each winning vehicle, the map data of the corresponding grid can be updated.

[0084] The method publishes a map update task to a plurality of opportunistic vehicles, receives bid information of each vehicle for a single task, and then selects a set of winning vehicles based on the system utility. Compared with the traditional way of relying on professional collection vehicles for map updating, this scheme fully utilizes the on-the-way collection capability of ordinary vehicles during daily driving, effectively avoids the high deployment and operation cost of professional vehicles, thereby realizing continuous map updating at a lower cost and significantly reducing the overall maintenance cost of the autonomous driving map.

[0085] In order to use lower cost to complete more tasks as much as possible, in a possible implementation manner, referring to Figure 3 Step S230 can include the following steps:

[0086] S231: based on the system utility of each opportunistic vehicle, an initial set of winning vehicles is selected from the opportunistic vehicles by using a greedy algorithm.

[0087] S232: the initial set of winning vehicles is optimized by using a variable neighborhood search algorithm to obtain a globally optimal set of winning vehicles.

[0088] An initial set of winning vehicles is first selected from a large number of opportunistic vehicles by using a greedy algorithm, so that the set can cover as many grid update tasks as possible. The initial set is not necessarily globally optimal, but can serve as a high-quality "starting point" to provide a basis for subsequent optimization. Based on the obtained initial set of winning vehicles, the variable neighborhood search algorithm is used to further improve the overall system utility. The goal is to find a globally more optimal set of winning vehicles to reduce the total cost or improve the task completion rate as much as possible under the premise of ensuring task coverage.

[0089] In order to efficiently obtain an initial set of winning vehicles as optimally as possible, referring to Figure 4 Step S231 can include the following steps:

[0090] S2311: initialize the initial winning vehicle set as empty, the to-be-completed task set as the set of normalizing map updating tasks, and the available vehicle set as all opportunistic vehicles.

[0091] S2312: calculate the system utility of each opportunistic vehicle in the available vehicle set.

[0092] S2313: select the opportunistic vehicle with the highest system utility from the available vehicle set as the target vehicle, add the target vehicle to the initial winning vehicle set, and remove the target vehicle from the available vehicle set.

[0093] S2314: add one to the cumulative coverage count of each grid updating task covered by the target vehicle; if the cumulative coverage count of any grid updating task plus its original confidence reaches a preset threshold, remove the grid updating task from the to-be-completed task set.

[0094] S2315: determine whether the to-be-completed task set is empty or the available vehicle set is empty; if not, return to execute steps S2312 to S2314; otherwise, output the current initial winning vehicle set.

[0095] First, initialize the initial winning vehicle set as empty, the to-be-completed task set as the set of normalizing map updating tasks, and the available vehicle set as all opportunistic vehicles. Among them, the initial winning vehicle set is initialized as empty, which will be continuously added with vehicles in subsequent processing. The to-be-completed task set is initialized as the set of normalizing map updating tasks, that is, all tasks are initially included, and after any task is completely allocated, it is removed from the to-be-completed task set. The available vehicle set is initialized as all opportunistic vehicles, that is, all opportunistic vehicles are initially available, and after a vehicle wins, it is removed from the available vehicle set.

[0096] Then, the system utility of each opportunistic vehicle in the available vehicle set is calculated. The calculation method has been described in the above formula, and it needs to be explained that, here, the to-be-completed task set is taken.

[0097] After obtaining the system utility of all available vehicles, the available vehicle with the highest system utility is added to the initial winning vehicle set, and the vehicle needs to be removed from the available vehicle set.

[0098] After selecting a winning vehicle, the grid set that can be covered by the winning vehicle is known The intersection with the set of tasks to be completed is the grid update task that the winning vehicle needs to handle, and the cumulative coverage count of these grid update tasks is added by 1 (which can be understood as the confidence of the corresponding grid being added by 1). At the same time, an action needs to be handled. When the cumulative coverage count of a grid update task plus its current confidence reaches a preset threshold (such as the above p), that is, the confidence of the grid reaches p, it indicates that the grid no longer needs to collect data, and the grid update task is removed from the set of tasks to be completed.

[0099] Finally, it is judged whether the set of tasks to be completed is empty or the set of available vehicles is empty. When the set of tasks to be completed is empty or the set of available vehicles is empty, that is, as long as one of them is empty, the allocation cannot continue, the current initial winning vehicle set is output. If neither of them is empty, steps S2312 to S2314 are returned to be executed. It needs to be explained why the system utility needs to be recalculated in step S2312. Because the set of tasks to be completed is used to calculate the system utility here, not the set of normalized map update tasks published at the beginning.

[0100] Through the above processing, a relatively optimal initial winning vehicle set can be obtained. The initial winning vehicle set can be understood as a local optimal solution, and therefore, it does not calculate all available vehicles, but only selects them in turn after sorting. In order to achieve the selection of a globally optimal winning vehicle, in an optional implementation manner, the above step S232 can include:

[0101] S2321: Set d neighborhood structures, and let the current neighborhood index j = 1.

[0102] S2322: Perform local search on the initial winning vehicle set based on the jth neighborhood structure. If a better solution is found, update the initial winning vehicle set to the better solution, and reset the neighborhood index j = 1. Otherwise, increase the neighborhood index j by 1.

[0103] S2323: Judge whether the neighborhood index j is less than or equal to d. If yes, perform the step of performing local search on the initial winning vehicle set based on the jth neighborhood structure. Otherwise, take the initial winning vehicle set as the final winning vehicle set.

[0104] Set d neighborhood structures (for example, 4), search from the first one, based on the current neighborhood structure j, perform local search on the current winning vehicle set, if a better solution (such as higher utility, lower cost) is found: update the winning vehicle set to this better solution; reset j = 1, and search from the first neighborhood again; if no better solution is found: increase the neighborhood index j by 1, and try the next neighborhood structure.

[0105] If j≤d (there are still untried neighborhood structures): continue the search using the next neighborhood; if j>d (all neighborhoods have been tried, still no better solution is found): end the search, output the current winning vehicle set as the final result.

[0106] For example, the 4 neighborhoods are .

[0107] Definition: element removal, i.e. removing a winning vehicle from the initial winning vehicle set. For example, after removing a winning vehicle, the winning vehicle set can still cover the same grid update task, then the winning vehicle set is a better solution.

[0108] Definition: element exchange, i.e. exchanging two vehicles randomly selected from the initial winning vehicle set and the non-winning vehicle set respectively. If the exchanged winning vehicle set can still cover the same grid update task and is lower in cost, then the winning vehicle set is a better solution.

[0109] Definition: symmetric set exchange, i.e. exchanging multiple vehicles randomly selected from the initial winning vehicle set and the non-winning vehicle set respectively.

[0110] Definition: asymmetric set exchange, i.e. exchanging multiple vehicles randomly selected from the initial winning vehicle set and the non-winning vehicle set respectively, with different numbers of vehicles.

[0111] Considering the individual rationality of the opportunistic vehicles, they often hope to get higher returns. In order to encourage vehicles that can make greater contributions to participate more actively in the map update task, the cloud platform can give them additional rewards. Therefore, referring to Figure 5 , the method can further include a step of calculating the reward income of the vehicles in the winning vehicle set, which includes:

[0112] S310: Calculate the first total number of grid update tasks of all winning vehicles in the winning vehicle set.

[0113] S320: Remove a winning vehicle from the opportunistic vehicles, and select a second winning vehicle set from the opportunistic vehicles based on the system utility of the remaining opportunistic vehicles.

[0114] S330: Calculate the second total number of grid update tasks of all winning vehicles in the second winning vehicle set.

[0115] S340: Calculate the reward income of the winning vehicle removed from the opportunistic vehicles according to the difference between the first total number and the second total number.

[0116] First, calculate the first total number of grid update tasks for all vehicles that bid wherein is the number of vehicles that complete tasks, is the set of vehicles that bid. Then, the bid is removed from the initial set of optional vehicles V, and a second set of bid vehicles is obtained by the above step S230 , the second total number of grid update tasks at this time is . Then the improvement in the amount of tasks completed by the platform can be expressed as , the reward income of the vehicle can be expressed as , and the actual income of the vehicle can be expressed as , wherein is the fixed cost of the vehicle to pass through a grid and upload data, which is less than or equal to .

[0117] Due to the complex and changeable road environment and frequent traffic accidents, the semantic information contained in the automatic driving map is extremely easy to change in a short period of time, bringing sudden map update needs. Based on this, referring to Figure 6 , the method can further include the following steps:

[0118] S410: In response to the received sudden event information, a set of sudden map update tasks is generated, and the set of sudden map update tasks is sent to each participating vehicle.

[0119] Among them, the participating vehicle, that is, the vehicle that changes its trajectory to the specified grid according to the cloud platform dispatching instruction to complete the perception task, can complete the sudden map update task.

[0120] S420: Each participating vehicle independently generates a dispatching action based on the set of sudden map update tasks through the trained distributed decision network to obtain a dispatched vehicle.

[0121] S430: Receive the perception data uploaded by each dispatched vehicle, and update the map data of the corresponding grid.

[0122] Receive sudden event information (such as road closure, traffic accident, temporary construction, etc.) from traffic monitoring, user reporting or other sensors, generate a set of sudden map update tasks (i.e. grid areas that need to be updated) according to the event location and influence range, and send the task set to all participating vehicles (i.e. vehicles with perception capabilities and can participate in sudden tasks).

[0123] Each participatory vehicle independently runs a trained distributed decision network; the network can output a scheduling action (such as "whether to go to collect" and "which grid to collect") according to the current location of the vehicle, the current set of emergent tasks, and the normalizing map update tasks (if any). It can be known which participatory vehicles will become scheduling vehicles. The scheduled vehicles collect perception data (such as cameras, radars, etc.) during the execution of the task, upload the data to the cloud platform, and the platform updates the map data of the corresponding grid, improving its confidence.

[0124] This method can quickly generate update tasks when road emergencies (such as accidents, construction, obstacles, etc.) occur, and schedule vehicles to complete perception tasks through an intelligent scheduling mechanism (based on a distributed decision network) to achieve rapid and efficient map updates.

[0125] The training of the distributed decision network can use a centralized training and decentralized execution strategy, and each distributed decision network shares the same value network. The distributed decision network can be an Actor network, and the value network can be a Critic network. The method also includes the training steps of the distributed decision network, which can include:

[0126] S510: Each distributed decision network generates its own first path planning decision according to the location of the participatory vehicle it is in, the set of emergent map update tasks, and the set of normalizing map update tasks.

[0127] S520: After each participatory vehicle executes its own first path planning decision, update the location of each participatory vehicle, the set of emergent map update tasks, and the set of normalizing map update tasks.

[0128] S530: According to the updated location of each participatory vehicle, the set of emergent map update tasks, and the set of normalizing map update tasks, generate a new second path planning decision.

[0129] S540: According to the values before and after updating the location of each participatory vehicle, the set of emergent map update tasks, and the set of normalizing map update tasks, and the first path planning decision and the second path planning decision, calculate the loss information of the value network and update the value network.

[0130] S550: Update each distributed decision network using the gradient provided by the value network.

[0131] In the offline centralized training phase, the Critic network acts as a central coordinator, calculates a centralized action-value function through global state information, evaluates the action of each Actor from a global perspective, and guides it to select a better action.

[0132] Steps S510 to S550 are a complete training activity in one time step, the core of which is to update the network using the data generated by one interaction.

[0133] First, the distributed decision-making network (Actor network) on each participating vehicle generates an action (i.e., path planning decision) according to its local observation. The local observation includes the self position, the set of emergent map updating tasks, and the set of normalized map updating tasks. Each vehicle makes an independent decision based on its own observations (local observation) without real-time communication with other vehicles.

[0134] Each vehicle executes its "first path planning decision" in parallel. Then, the cloud platform aggregates the action results of all vehicles to update the global environment state. Understandably, in the training process, this process can be performed to simulate the first path planning decision. The specific update content can be: updating the vehicle positions: updating the positions of the vehicles to the new grid according to their actions; updating the task set: if a vehicle passes through a grid that needs to be updated during its movement and uploads data, the task of that grid should be removed from the corresponding set. If an emergent task is completed, it is removed from the set of emergent map updating tasks. If a normalized task is completed, it is removed from the set of normalized map updating tasks.

[0135] Based on the updated global state of step S520, each vehicle's Actor network generates a new path planning decision (second path planning decision) again.

[0136] Then, based on the above obtained data, the loss of the Critic network is calculated and its parameters are updated. The formula for loss calculation is:

[0137]

[0138] wherein, represents the expectation, represents the action-value function, is the parameter of the Critic network, and S is the global state at the beginning of step S510, i.e., the positions of all participating vehicles, the set of emergent map updating tasks, and the set of normalized map updating tasks, is the first path planning decision of each vehicle, represents the updated global state after the execution of step S520, including: the updated positions of each vehicle, the updated set of emergent tasks (removed completed ones), and the updated set of normalized tasks (removed completed ones), is the second path planning decision, represents the discount coefficient, the value of the Critic network for the tuple (S, the first path planning decision) at the current time, the value of the target Critic network for the tuple (S', the second path planning decision) at the current time, the immediate reward at time step t.

[0139] participating vehicles at time t performing actions The reward obtained is:

[0140] ,

[0141] wherein, represents the set of all tasks at time t, is the difference set of the two sets, indicating the set of sudden map update tasks, represents the vehicle the number of sudden map update tasks completed by the vehicle within the time interval t, represents the uncompleted normalization map update task, represents the vehicle the number of normalization map update tasks completed by the vehicle within the time interval t, and are weight coefficients, since it is hoped that the participating vehicles will give priority to completing the sudden map update tasks, a higher reward will be allocated to them, i.e. .

[0142] Finally, the Critic network just updated is used to guide the update of each Actor network. For each participating vehicle, the gradient is calculated, and the gradient direction is to improve the value of its first path planning decision evaluated by the Critic network. The calculation formula of the gradient is:

[0143]

[0144] wherein, is the update direction of each Actor network parameter, is the value of the Critic network for S and , S' is the state of the vehicle at the current time, represents the path planning decision generated by the Actor network according to , and ▽ is the gradient operator, representing differentiation.

[0145] Through the gradient ascent method, the parameters of the distributed decision network (Actor) of each vehicle are updated along the direction of .

[0146] Further, the embodiment of the present application also provides an automatic driving map updating device, which is described with reference to Figure 7The automatic driving map updating device 600 comprises:

[0147] A task generation unit 610 is configured to generate a set of normalization map updating tasks according to the confidence of each grid in the automatic driving map; the set of normalization map updating tasks comprises at least one grid updating task that needs to be updated.

[0148] A task publishing unit 620 is configured to publish the set of normalization map updating tasks to a plurality of opportunistic vehicles and receive cost quotes of each opportunistic vehicle for a single grid updating task.

[0149] A vehicle selection unit 630 is configured to select a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle; wherein the system utility is the ratio of the number of grid updating tasks that can be completed by the opportunistic vehicle to the cost quote.

[0150] A map updating unit 640 is configured to receive perception data uploaded by the winning vehicles in the set of winning vehicles and update the map data of the corresponding grid.

[0151] In summary, the automatic driving map updating method, device, electronic equipment and storage medium provided by the embodiment of the present application generate normalization updating tasks based on the confidence of the map grid, combine the system utility evaluation mechanism, and optimally select the set of winning vehicles with the highest cost performance from the opportunistic vehicles, thereby significantly reducing the map updating cost and improving the updating efficiency; the greedy algorithm and the variable neighborhood search algorithm are combined to optimize the vehicle scheduling scheme, ensuring the local and global optimality of the solution; the reward mechanism based on the contribution evaluation is introduced to encourage the vehicles to actively participate in the map updating task; for sudden events, the opportunistic vehicles can be quickly dispatched to respond to the sudden map updating tasks, thereby improving the real-time performance and safety of the map.

[0152] In the embodiments of the present application, it should be understood that the disclosed apparatus and method can also be implemented in other manners. The embodiments described above are merely exemplary for describing the present application. For example, the flowcharts and block diagrams in the accompanying drawings show the possible implementation architectures, functions and operations of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowcharts and block diagrams can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that, in some alternative implementations, the functions noted in the blocks can occur in different orders from those described in the accompanying drawings. For example, two consecutive blocks can actually be executed in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and the combination of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system for executing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0153] In addition, each functional module in the various embodiments of the present application can be integrated together to form a separate part, or each module can exist independently, or two or more modules can be integrated to form a separate part.

[0154] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the present application or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a computer readable storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0155] The above merely provides preferred embodiments of the present application, but is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0156] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics of the present application. The embodiments should therefore be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims should be considered as limiting the scope of the claims with respect to the figures of the patent document.

Claims

1. An automatic driving map update method characterized by comprising: The method comprises the following steps: generating a set of normalized map update tasks according to the confidence of each grid in the autonomous driving map; the set of normalized map update tasks comprises at least one grid update task that needs to be updated; publishing the set of normalized map update tasks to a plurality of opportunistic vehicles and receiving cost quotes of each opportunistic vehicle for a single grid update task; selecting a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle; wherein the system utility is the ratio of the number of grid update tasks that the opportunistic vehicle can complete to the cost quote; receiving perception data uploaded by the winning vehicles in the set of winning vehicles and updating the map data of the corresponding grid.

2. The method of claim 1, wherein, The method further comprises the following steps of selecting a set of winning vehicles from the opportunistic vehicles based on the system utility of each opportunistic vehicle: selecting an initial set of winning vehicles from the opportunistic vehicles using a greedy algorithm based on the system utility of each opportunistic vehicle; optimizing the initial set of winning vehicles using a variable neighborhood search algorithm to obtain a globally optimal set of winning vehicles.

3. The method of claim 2, wherein, The method further comprises the following steps of selecting an initial set of winning vehicles from the opportunistic vehicles using a greedy algorithm based on the system utility of each opportunistic vehicle: initializing the initial set of winning vehicles as empty, the set of tasks to be completed as the set of normalized map update tasks, and the set of available vehicles as all the opportunistic vehicles; calculating the system utility of each opportunistic vehicle in the set of available vehicles; selecting the opportunistic vehicle with the highest system utility from the set of available vehicles as the target vehicle, adding the target vehicle to the initial set of winning vehicles, and removing the target vehicle from the set of available vehicles; adding one to the cumulative coverage count of each grid update task covered by the target vehicle; if the cumulative coverage count of any grid update task plus its original confidence reaches a preset threshold, removing the grid update task from the set of tasks to be completed; determining whether the set of tasks to be completed is empty or the set of available vehicles is empty; if not, returning to the step of calculating the system utility of each opportunistic vehicle in the set of available vehicles; otherwise, outputting the current initial set of winning vehicles.

4. The method of claim 2, wherein, The method further comprises the following steps of optimizing the initial set of winning vehicles using a variable neighborhood search algorithm to obtain a globally optimal set of winning vehicles: setting d neighborhood structures and letting the current neighborhood index j = 1; performing local search on the initial set of winning vehicles based on the jth neighborhood structure; if a better solution is found, updating the initial set of winning vehicles to the better solution and resetting the neighborhood index j = 1; otherwise, increasing the neighborhood index j by one; determining whether the neighborhood index j is less than or equal to d; if yes, performing the step of local search on the initial set of winning vehicles based on the jth neighborhood structure; otherwise, taking the initial set of winning vehicles as the final set of winning vehicles.

5. The method of claim 1, wherein, The method further comprises the following steps of calculating the reward income of the vehicles in the set of winning vehicles, which comprises: calculating a first total number of grid update tasks of all the winning vehicles in the winning vehicle set; removing one of the winning vehicles from the opportunistic vehicles, and selecting a second winning vehicle set from the opportunistic vehicles based on system utilities of the remaining opportunistic vehicles; calculating a second total number of grid update tasks of all the winning vehicles in the second winning vehicle set; calculating a reward income of the removed winning vehicle based on a difference between the first total number and the second total number.

6. The method of claim 1, wherein, The method further comprises: generating a set of emergency map update tasks in response to the received emergency information, and sending the set of emergency map update tasks to each of the participating vehicles; each of the participating vehicles independently generates a scheduling action based on the set of emergency map update tasks through the trained distributed decision network to obtain a scheduled vehicle; receiving perception data uploaded by each of the scheduled vehicles, and updating map data of a corresponding grid.

7. The method of claim 6, wherein, The method further comprises a training step of the distributed decision network, which comprises: each of the distributed decision networks generates a first path planning decision based on a location of the participating vehicle where the distributed decision network is located, the set of emergency map update tasks, and a set of normal map update tasks; each of the participating vehicles updates the location of the participating vehicle, the set of emergency map update tasks, and the set of normal map update tasks after executing the first path planning decision; each of the participating vehicles generates a second path planning decision based on the updated location of the participating vehicle, the set of emergency map update tasks, and the set of normal map update tasks; each of the participating vehicles calculates loss information of a value network based on values before and after the location of the participating vehicle, the set of emergency map update tasks, and the set of normal map update tasks are updated, and the first path planning decision and the second path planning decision, and updates the value network; each of the distributed decision networks is updated using a gradient provided by the value network.

8. An automatic driving map updating apparatus characterized by comprising: comprises: a task generation unit configured to generate a set of normal map update tasks based on a confidence level of each grid in an autonomous driving map; the set of normal map update tasks comprises at least one grid update task that needs to be updated; a task publishing unit configured to publish the set of normal map update tasks to a plurality of opportunistic vehicles, and receive cost quotes of each of the opportunistic vehicles for a single grid update task; a vehicle selection unit configured to select a winning vehicle set from the opportunistic vehicles based on system utilities of each of the opportunistic vehicles; the system utility is a ratio of a number of grid update tasks that can be completed by the opportunistic vehicle to the cost quote; a map updating unit configured to receive perception data uploaded by a winning vehicle in the winning vehicle set, and update map data of a corresponding grid.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor implements the method of any one of claims 1 to 7 when executing the computer program.

10. A storage medium having stored thereon a computer program, characterized in that: The computer program, which is executed by a processor, implements the method of any one of claims 1 to 7.

Images