Optimization method, device, equipment and storage medium for autonomous driving model

Abstract

The present application discloses an optimization method for an autonomous driving model. The method includes: obtaining vehicle data of a target vehicle and environmental data of the target vehicle, inputting the vehicle data and environmental data into the autonomous driving model to be optimized, determining whether to change lanes based on the vehicle data and environmental data, and using the result of whether to change lanes as the output result of the autonomous driving model to be optimized, comparing the output result with the actual lane change result to obtain a comparison result, and optimizing the autonomous driving model to be optimized based on the comparison result. In this way, by comparing the output result of whether to change lanes obtained by the autonomous driving model with the actual lane change result, it is possible to determine whether the decision of the autonomous driving model is consistent with the decision of the actual driver, and optimizing the autonomous driving model based on the comparison result, so as to achieve the optimization of the autonomous driving model in the lane change scenario.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to an optimization method, device, equipment and storage medium for an autonomous driving model. Background Art

[0002] With the development of science and technology, autonomous driving has become one of the hottest research projects. Autonomous driving means that the vehicle can automatically drive to the designated destination without human operation or control. In autonomous driving, the autonomous driving model is the key to autonomous driving vehicles. The autonomous driving model can provide autonomous driving vehicles with the same decision-making, operation or control as human drivers in actual scenarios.

[0003] In the prior art, the autonomous driving model is to acquire a large amount of lane change scene data for reinforcement learning, so as to make corresponding decisions when facing lane change scenarios. However, the autonomous driving model that has learned the lane change scene data may not necessarily be able to make the correct lane change behavior decision in different scenarios. Once the autonomous driving model provides the autonomous driving vehicle with an incorrect lane change behavior decision, the consequences will be disastrous. However, in the prior art, there is no method to optimize the autonomous driving model in the lane change scenario.

[0004] Based on this, how to optimize the autonomous driving model in the lane change scenario is an urgent problem to be solved by technical personnel in this field. Summary of the invention

[0005] Based on the above problems, the present application provides an optimization method, device, equipment and storage medium for the autonomous driving model to achieve optimization of the black box in the lane change scenario.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides an optimization method for an autonomous driving model, the method comprising:

[0008] Acquire vehicle data of a target vehicle and environmental data where the target vehicle is located;

[0009] Inputting the vehicle data and the environment data into an autonomous driving model to be optimized;

[0010] The autonomous driving model determines whether to change lanes according to the vehicle data and the environmental data, and uses the result of whether to change lanes as the output result of the autonomous driving model to be optimized;

[0011] Comparing the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle;

[0012] The automatic driving model to be optimized is optimized according to the comparison result.

[0013] Optionally, optimizing the automatic driving model to be optimized according to the comparison result includes:

[0014] Accumulate and obtain a preset number of comparison results;

[0015] Determine the ratio of the comparison results that the output result is the same as the actual lane change result in the preset number of comparison results;

[0016] Determine whether the proportion is greater than or equal to a preset proportion value. If so, complete the optimization of the autonomous driving model; if not, adjust the parameters of the autonomous driving model to be optimized, and continue to use the vehicle data and the environmental data to optimize the autonomous driving model to be optimized.

[0017] Optionally, the method further comprises:

[0018] Acquire actual driving data of the target vehicle, wherein the actual driving data includes at least two of the steering information of the target vehicle, the vehicle orientation, the target distance, and the lane where the target vehicle is located, and the target distance data is obtained by the distance between the target vehicle and the lane where the target vehicle is located;

[0019] The lane change data of the target vehicle is obtained according to the actual driving data.

[0020] Optionally, acquiring lane change data of the target vehicle according to the target distance and the lane where the target vehicle is located in the actual driving data includes:

[0021] When the target vehicle moves, the relative size of the target distance and the preset distance is detected in real time;

[0022] When it is detected that the target distance is greater than the preset distance, the current time is recorded as the starting lane change time of the target vehicle, and the lane in which the target vehicle is located at the starting lane change time is recorded;

[0023] After recording the starting lane change time of the target vehicle, continuously detecting the relative size of the target distance and the preset distance, and determining whether the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the starting lane change time;

[0024] When it is detected that the target distance is less than the preset distance, and the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the start lane change time, the current time is recorded as the end lane change time of the target vehicle, and the lane where the target vehicle is located at the end lane change time is recorded;

[0025] The lane change data of the target vehicle is obtained according to the starting lane change time, the ending lane change time, the lane where the target vehicle is located at the starting lane change time, and the lane where the target vehicle is located at the ending lane change time.

[0026] Optionally, obtaining lane change data of the target vehicle according to the actual driving data includes:

[0027] Acquiring actual driving data of the target vehicle, wherein the actual driving data includes steering information and vehicle orientation;

[0028] Dividing the actual driving data into multiple time windows to obtain driving data corresponding to the multiple time windows respectively;

[0029] The actual driving data corresponding to each time window having the wheel turning information that meets the preset wheel turning rule is used as the target data;

[0030] It is determined whether the vehicle orientation in each target data meets the preset orientation rule. If so, the target data is marked as lane change data.

[0031] In a second aspect, an embodiment of the present application provides an optimization device for an autonomous driving model, the device comprising: a data acquisition unit, an input unit, an output result determination unit, a result comparison unit, and an autonomous driving model optimization unit;

[0032] The data acquisition unit is used to acquire vehicle data of the target vehicle and environmental data where the target vehicle is located;

[0033] The input unit is used to input the vehicle data and the environment data into the automatic driving model to be optimized;

[0034] The output result determination unit is used to determine whether to change lanes according to the vehicle data and the environmental data, and use the result of whether to change lanes as the output result of the automatic driving model to be optimized;

[0035] The result comparison unit is used to compare the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle;

[0036] The autonomous driving model optimization unit is used to optimize the autonomous driving model to be optimized according to the comparison result.

[0037] Optionally, the autonomous driving model optimization unit includes:

[0038] A comparison result accumulation unit, used for accumulating a preset number of comparison results;

[0039] A ratio determination unit, used to determine the ratio of the comparison results in which the output result is the same as the actual lane change result in the preset number of comparison results;

[0040] The autonomous driving model optimization subunit is used to determine whether the proportion is greater than or equal to a preset proportion value. If so, the optimization of the autonomous driving model is completed; if not, the parameters of the autonomous driving model to be optimized are adjusted, and the vehicle data and the environmental data are continued to be used to optimize the autonomous driving model to be optimized.

[0041] Optionally, the device further comprises:

[0042] an actual driving data acquisition unit, configured to acquire actual driving data of a target vehicle, wherein the actual driving data includes at least two of the steering information of the target vehicle, the vehicle orientation, the target distance, and the lane where the target vehicle is located, wherein the target distance data is obtained by the distance between the target vehicle and the lane where the target vehicle is located;

[0043] A lane change data determination unit is used to obtain the lane change data of the target vehicle according to the actual driving data.

[0044] In a third aspect, an embodiment of the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the optimization method for the autonomous driving model as described in the first aspect is implemented.

[0045] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes the optimization method for the autonomous driving model as described in the first aspect.

[0046] Compared with the prior art, this application has the following beneficial effects:

[0047] This application obtains the vehicle data of the target vehicle and the environmental data where the target vehicle is located, inputs the vehicle data and the environmental data into the autonomous driving model to be optimized, determines whether to change lanes based on the vehicle data and the environmental data, and uses the result of whether to change lanes as the output result of the autonomous driving model to be optimized, compares the output result with the actual lane change result, obtains a comparison result, and optimizes the autonomous driving model to be optimized based on the comparison result. Among them, by comparing the output result of whether to change lanes obtained by the autonomous driving model with the actual lane change result, it can be determined whether the decision of the autonomous driving model is consistent with the decision of the actual driver, and further optimizes the autonomous driving model based on the comparison result, so as to achieve the optimization of the autonomous driving model in the lane change scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 A flowchart of an optimization method for an autonomous driving model provided in an embodiment of the present application;

[0050] Figure 2 A flowchart for extracting lane change data in an optimization method for an autonomous driving model provided in an embodiment of the present application;

[0051] Figure 3 A flowchart of the specific process of S205 provided in the embodiment of the present application;

[0052] Figure 4 A flowchart of positioning lane change data in an optimization method for an autonomous driving model provided in an embodiment of the present application;

[0053] Figure 5 A schematic diagram of dividing actual driving data into multiple time windows provided in an embodiment of the present application;

[0054] Figure 6 A schematic diagram of the lateral displacement of two adjacent time nodes provided in an embodiment of the present application;

[0055] Figure 7 A schematic diagram of the lateral displacement between the first time node and the last time node provided in an embodiment of the present application;

[0056] Figure 8A schematic diagram of the structure of an optimization device for an autonomous driving model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] As described above, the inventors found in their research on the autonomous driving model that in the prior art, the autonomous driving model is to make corresponding decisions when facing lane changing scenarios by performing reinforcement learning on a large amount of lane changing scene data. However, the autonomous driving model that has learned the lane changing scene data may not necessarily be able to make the correct lane changing behavior decisions in different scenarios. Once the autonomous driving model provides the autonomous driving vehicle with an incorrect lane changing behavior decision, the consequences will be disastrous. However, in the prior art, there is no method for optimizing the autonomous driving model in the lane changing scenario.

[0058] In order to solve the above problems, an embodiment of the present application provides an optimization method for an autonomous driving model. The method includes: obtaining vehicle data of a target vehicle and environmental data of the target vehicle, inputting the vehicle data and environmental data into the autonomous driving model to be optimized, determining whether to change lanes based on the vehicle data and environmental data, and using the result of whether to change lanes as the output result of the autonomous driving model to be optimized, comparing the output result with the actual lane change result to obtain a comparison result, and optimizing the autonomous driving model to be optimized based on the comparison result.

[0059] In this way, by comparing the output result of whether to change lanes obtained by the autonomous driving model with the actual lane changing result, it is possible to determine whether the decision of the autonomous driving model is consistent with the decision of the actual driver. By further optimizing the autonomous driving model based on the comparison result, the optimization of the autonomous driving model can be achieved in the lane changing scenario.

[0060] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0061] Based on the above content, the present application embodiment provides an optimization method for an autonomous driving model, see Figure 1 , Figure 1 A flowchart of an optimization method for an autonomous driving model provided in an embodiment of the present application, combined with Figure 1 As shown, the optimization method for the autonomous driving model provided in the embodiment of the present application may include:

[0062] S101: Acquire vehicle data of a target vehicle and environmental data where the target vehicle is located.

[0063] Vehicle data may include the vehicle's own hardware information, software information, network information, and data information generated by the target vehicle during driving, which is not specifically limited here.

[0064] Environmental data may include the lane where the target vehicle is located, the lanes on both sides of the lane where the target vehicle is located (if any), the distance between the target vehicle and the vehicle in front, the distance between the target vehicle and the vehicle behind, the turning lane in front of the target vehicle, construction locations, traffic lights, pedestrians, etc., including but not limited to the several types of environmental data mentioned above, which are only used as examples here and are not specifically limited.

[0065] S102: Inputting the vehicle data and the environmental data into the autonomous driving model to be optimized.

[0066] Autonomous driving models are generally used to provide decision-making, operation or control for autonomous driving vehicles, thereby achieving autonomous driving.

[0067] Among them, vehicle data and environmental data are input into the autonomous driving model, and the autonomous driving model will make corresponding decisions, operations or controls based on the vehicle data and environmental data.

[0068] S103: The autonomous driving model determines whether to change lanes based on the vehicle data and the environmental data, and uses the result of whether to change lanes as the output result of the autonomous driving model to be optimized.

[0069] Based on the input vehicle data and environmental data, the autonomous driving model will decide whether to perform a lane change operation or a non-lane change operation, where non-lane change operations refer to turning, parking, uphill and other operations, that is, all operations except lane change operations are considered lane change operations. The above are just a few examples of non-lane change operations and are not specifically limited here.

[0070] S104: Compare the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle.

[0071] Among them, the output result refers to the decision result made by the autonomous driving model based on vehicle data and environmental data.

[0072] Among them, the actual lane change result refers to the lane change result of the above-mentioned vehicle data and environmental data in the actual scenario, that is, the lane change decision and operation made by a real driver in the actual scenario.

[0073] S105: Optimizing the autonomous driving model to be optimized according to the comparison result.

[0074] The decision results of the autonomous driving model are compared with those of the real driver to obtain a comparison result, and the autonomous driving model is optimized according to the comparison result.

[0075] As a feasible implementation, S105 may specifically include:

[0076] Step 1: Accumulate and obtain a preset number of comparison results.

[0077] Step 2: Determine the proportion of the comparison results in which the output result is the same as the actual lane change result among the preset number of comparison results.

[0078] Step 3: Determine whether the proportion is greater than or equal to the preset proportion value. If so, complete the optimization of the autonomous driving model; if not, adjust the parameters of the autonomous driving model to be optimized, and continue to use the vehicle data and the environmental data to optimize the autonomous driving model to be optimized.

[0079] Since adjusting the autonomous driving model based on a single comparison result cannot reflect the objective reality of the autonomous driving model, the comparison results can be accumulated and the proportion of the comparison results whose output results are the same as the actual lane change results in the preset number of comparison results can be determined. The proportion is compared with a preset proportion value. If it is greater than or equal to the preset proportion value, it is considered that the correctness or accuracy of the autonomous driving model output can be used for lane change decisions. If the proportion is less than the preset proportion value, it is considered that the autonomous driving model needs to be optimized, and the parameters of the autonomous driving model can be adjusted, and the vehicle data and the environmental data can be used to optimize the autonomous driving model.

[0080] Among them, by using the same set of vehicle data and environmental data to continuously adjust parameters and optimize the autonomous driving model, the accuracy of decisions made by the optimized autonomous driving model is higher. This is different from the method of using new data every time the parameters are adjusted. The embodiment of the present application can better reflect the iterative process of autonomous driving model optimization.

[0081] The preset ratio value is set according to the actual optimization scenario and is not specifically limited here.

[0082] As an optional implementation, in order to better optimize the autonomous driving model, the method may further include:

[0083] Step 1: Acquire actual driving data of the target vehicle, wherein the actual driving data includes steering information data, vehicle orientation data, target distance data and preset distance data of the target vehicle, and the target distance data is obtained by the distance between the target vehicle and the lane where the target vehicle is located;

[0084] Step 2: Obtain lane change data of the target vehicle based on at least two types of data in the actual driving data.

[0085] In the embodiment of the present application, the lane change data used to determine the actual lane change result can be determined using the actual driving data of the target vehicle.

[0086] The optimization method for the autonomous driving model provided in the embodiment of the present application can determine whether the decision of the autonomous driving model is consistent with the decision of the actual driver by comparing the output result of whether to change lanes obtained by the autonomous driving model with the actual lane changing result. The autonomous driving model can be further optimized according to the comparison result, thereby realizing the optimization of the autonomous driving model in the lane changing scenario.

[0087] Based on the content of the above embodiment, the embodiment of the present application also provides a process for extracting lane change data in the optimization method for the automatic driving model, which can obtain lane change data according to the target distance data and the lane where the target vehicle is located in the actual driving data, see Figure 2 , Figure 2 A flowchart of extracting lane change data in an optimization method for an autonomous driving model provided in an embodiment of the present application. The process of extracting lane change data in an optimization method for an autonomous driving model provided in an embodiment of the present application may include:

[0088] S201: When the target vehicle moves, the relative size between the target distance and the preset distance is detected in real time.

[0089] As a feasible implementation, step S201 may specifically include:

[0090] Step 1: When the target vehicle moves, the positioning point of the target vehicle is detected in real time.

[0091] Among them, the positioning point can be installed at the center position of the target vehicle, at the center position of the front of the vehicle, or at the center position of the rear axle of the vehicle, including but not limited to the above three positions, which are not specifically limited here.

[0092] Step 2: Determine the shortest distance between the positioning point and the center line of the lane where the target vehicle is located.

[0093] The shortest distance is the shortest straight line segment between the positioning point and the center line of the lane where the target vehicle is located.

[0094] Step 3: Detect the relative size of the shortest distance and the preset distance in real time.

[0095] The preset distance refers to the difference between half the width of the lane where the target vehicle is located and half the width of the target vehicle. Since the vehicle widths of different models are different, the preset distance is not specifically limited. In addition, the conditions for determining the preset distance, including but not limited to the above, are not specifically limited here.

[0096] S202: When it is detected that the target distance is greater than the preset distance, the current time is recorded as the starting lane change time of the target vehicle, and the lane where the target vehicle is located at the starting lane change time is recorded.

[0097] Among them, during normal driving, the target distance will always be less than the preset distance. When the target vehicle changes lanes, the target vehicle will change from the original lane to the changed lane. During this process, the target distance will change from less than the preset distance to greater than the preset distance, thereby determining that when the target distance is greater than the preset distance, the target vehicle changes lanes, and this moment is recorded as the starting lane change moment.

[0098] S203: After recording the starting lane change time of the target vehicle, continuously detecting the relative size of the target distance and the preset distance, and determining whether the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the starting lane change time.

[0099] Among them, after recording the starting lane change time, it is necessary to determine whether the target vehicle changes from the original lane to the changed lane, because in actual scenarios, the target vehicle may cancel the lane change in the middle of the lane change, that is, the actual result is that no lane change is made. Therefore, it is necessary to determine whether the lane where the target vehicle is located after recording the starting lane change time is inconsistent with the lane where the target vehicle was located at the starting lane change time, so as to determine whether the target vehicle has changed lanes.

[0100] S204: When it is detected that the target distance is less than the preset distance, and the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the start lane change time, the current time is recorded as the end lane change time of the target vehicle, and the lane where the target vehicle is located at the end lane change time is recorded.

[0101] Among them, when the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the time of the initial lane change, the target vehicle may cancel the lane change behavior midway. Although the lane is changed, the actual result may be that the lane is not changed. Therefore, it is also necessary to determine whether the target distance is less than the preset distance. After the lane change, the target distance is the distance between the positioning point of the target vehicle and the changed lane. If a lane change occurs, the target distance at this time should change from greater than the preset distance to less than the preset distance. When the target distance is less than the preset distance and the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the time of the initial lane change, it proves that the target vehicle has changed lanes and changed from the original lane to the new lane, so this moment is taken as the end of the lane change.

[0102] S205: Obtain lane change data of the target vehicle according to the lane change start time, the lane change end time, the lane where the target vehicle is located at the lane change start time, and the lane where the target vehicle is located at the lane change end time.

[0103] The lane change data may include: the time when the lane change occurs, the lane change process, and the time when the lane change ends. Through steps S202 to S204, the start time of the lane change, the time when the original lane changes to the new lane, and the time when the lane change ends are determined, the data standard is unified, the obtained data is neat, and the problem of uneven data due to manual labeling is solved, which improves the quality of the lane change data.

[0104] As a feasible implementation method, in order to improve the availability and accuracy of lane change data, see Figure 3 , Figure 3 The flowchart of the specific process of S205 provided in the embodiment of the present application is combined with Figure 3 As shown, step S205 may specifically include:

[0105] S2051: Determine an actual lane-changing start time and an actual lane-changing end time according to the lane-changing start time and the lane-changing end time, respectively.

[0106] S2052: Obtain lane change data of the target vehicle according to the actual lane change start time, the actual lane change end time, the lane where the target vehicle is located at the actual lane change start time, and the lane where the target vehicle is located at the actual lane change end time.

[0107] As a feasible implementation, step S2051 may specifically include:

[0108] Step 1: Check the shortest distance between the positioning point of the target vehicle before the start lane change moment and the center line of the lane where the target vehicle is located.

[0109] Step 2: When the shortest distance does not decrease, the current time is recorded as the actual starting lane change time of the target vehicle, and the lane where the target vehicle is located at the actual starting lane change time is recorded.

[0110] The starting lane change time is the time when the lane change has already started, not the time when the lane change behavior begins, that is, the actual lane change time is before the starting lane change time.

[0111] Therefore, before the initial lane change moment, the shortest distance can be checked one by one. From the initial lane change moment to the actual lane change moment, the shortest distance should gradually decrease. If the shortest distance does not change, then the moment when the shortest distance does not change should be the actual lane change moment.

[0112] Step 3: Check the shortest distance between the positioning point of the target vehicle after the lane change ends and the center line of the lane where the target vehicle is located.

[0113] Step 4: When the shortest distance does not decrease, the current time is recorded as the actual lane change end time of the target vehicle, and the lane where the target vehicle is located at the actual lane change end time is recorded.

[0114] Among them, the lane change end time is the time after the original road has been changed to the road, but the vehicle at this time may not be parallel to the road after the lane change, that is, at the end of the lane change moment, the vehicle body may not have turned straight, so the actual end of the lane change moment is after the end of the lane change moment.

[0115] Therefore, after determining the end time of lane change, continue to detect the change of the shortest distance. In the process from the end time of lane change to the actual end time of lane change, the shortest distance should gradually decrease. When the shortest distance no longer changes, the moment when the shortest distance does not change is the actual end time of lane change.

[0116] The specific implementation process of step S205 provided in the embodiment of the present application can make the extracted lane change data more accurate, and improve the availability and accuracy of the lane change data.

[0117] The lane change data extraction process in the optimization method for the autonomous driving model provided in the embodiment of the present application can realize the automatic extraction of lane change data, which is different from the manual labeling method in the prior art and improves the efficiency of lane change data extraction; and, through specific judgment conditions, the standards for judging the starting lane change time and the ending lane change time of the vehicle can be unified, so that the data is neat and the data quality is improved.

[0118] The lane change data obtained through the process of the above embodiment, when applied to optimize the autonomous driving model, will make the decisions generated by the optimized autonomous driving model more accurate and reliable.

[0119] Based on the contents of the above embodiments, the embodiments of the present application further provide a process for locating lane change data in an optimization method for an autonomous driving model. By using the steering information and vehicle orientation in the actual driving data, the lane change data in the actual driving data can be located, thereby facilitating the subsequent manual extraction of lane change data. Figure 4 , Figure 4 A flowchart of the positioning of lane change data in an optimization method for an autonomous driving model provided in an embodiment of the present application, combined with Figure 4 As shown, the positioning process of lane change data in the optimization method for the autonomous driving model provided in the embodiment of the present application may specifically include:

[0120] S401: Acquire actual driving data of a target vehicle, wherein the actual driving data includes steering information and vehicle direction.

[0121] The steering information refers to the steering value generated when the driver turns the steering wheel during vehicle driving. Taking a car turning as an example, when turning left, the steering wheel will turn from forward to left, that is, when the steering wheel is turned left, the difference generated from turning forward to left will be used as the steering value.

[0122] The vehicle orientation refers to the orientation of the front of the target vehicle during actual driving, including but not limited to using the front of the vehicle as the basis for orientation. Other positions of the target vehicle may also be used as a basis to determine the vehicle orientation, which is not specifically limited here.

[0123] S402: Divide the actual driving data into multiple time windows to obtain actual driving data corresponding to the multiple time windows respectively.

[0124] As a feasible implementation, step S402 may specifically include:

[0125] Step 1: Divide the actual driving data into a first time window according to a time sequence and a preset time length, and obtain the actual driving data corresponding to the first time window.

[0126] Step 2: When dividing the second time window, the starting position of the second time window is before the ending position of the first time window, and the starting position of the second time window is after the starting position of the first time window, and a window of the preset time length is divided from the starting position of the second time window as the second time window, and actual driving data corresponding to the second time window is obtained.

[0127] Step 3: Divide all remaining time windows in sequence according to the division process of the second time window.

[0128] The preset time length is determined according to the actual scenario and is not specifically limited here.

[0129] Among them, the starting position of the second time window is determined before the ending position of the first time window and after the starting position of the first time window, so that the data between the second time window and the first time window are overlapped. Because in the process of dividing the time windows of the actual driving data, a complete lane changing process may be divided into two time windows. Therefore, overlapping the data of the first window and the second window can avoid determining the lane changing process as non-lane changing data after the complete lane changing process is divided into two windows, which can improve the accuracy of the positioning of the lane changing data.

[0130] For steps 1 to 3, take a segment of actual driving data as an example and divide the actual driving data into multiple time windows. Figure 5 , Figure 5 A schematic diagram of dividing actual driving data into multiple time windows provided in an embodiment of the present application, combined with Figure 5 As shown, the process of dividing the actual driving data into multiple time windows can be:

[0131] The preset time length is set to 12 seconds, and the actual driving data is divided to obtain time window 1. When dividing time window 2, the overlapping time length of time window 2 and time window 1 is set to 6 seconds. The remaining time windows are divided in turn to obtain N time windows.

[0132] Among them, the above content and Figure 5 It is only used as an example and explanation of steps one to three, and is not a basis for limiting the scope of protection of this application.

[0133] S403: The actual driving data corresponding to each time window and in which the steering information conforms to the preset steering rule is used as the target data.

[0134] As a feasible implementation, S403 may specifically include:

[0135] It is determined whether the steering information in the actual driving data corresponding to each time window contains steering values ​​that exceed the preset upper limit value and the preset lower limit value respectively. If so, the actual driving data corresponding to the time window is marked as target data.

[0136] Among them, the preset upper limit value and the preset lower limit value can be determined according to the actual scenario and are not specifically limited here.

[0137] In actual driving, if the target vehicle changes lanes, the steering wheel should be turned to one side, then turned back to the center, then turned to the other side, and then turned back to the center. Therefore, if a lane change occurs, the corresponding steering value in the time window should exceed the preset upper limit and lower limit, respectively.

[0138] By utilizing the steering information in the actual driving data, a rough judgment is made on whether the target vehicle has changed lanes. The time windows that do not meet the rough judgment are excluded, and the actual driving data corresponding to the time windows that meet the rough judgment are used as the target data. This allows for a preliminary judgment of the actual driving data and a narrower positioning range.

[0139] S404: Determine whether the vehicle orientation in each target data complies with a preset orientation rule. If so, mark the target data as lane change data.

[0140] As a feasible implementation, S404 may specifically include:

[0141] Step 1: Determine a plurality of time nodes at which the wheel turning value in each target data is equal to the preset error value, wherein the plurality of time nodes are arranged in chronological order.

[0142] Among them, since in actual driving scenarios, the steering wheel is not always in the forward direction, and may fluctuate due to road bumps or other conditions, a preset error value is set, and the steering value in the target data that is less than the preset error value is defined as 0. By setting the preset error value, the target data that has been roughly judged can be made more accurate, and the errors caused by actual conditions can be eliminated, which can further narrow and accurately position the range.

[0143] Among them, multiple time nodes at which the steering value in each target data is equal to the preset error value are determined, and further judgment on the lane change behavior can be made according to the position and direction of the target vehicle corresponding to each time node.

[0144] Step 2: Determine whether the vehicle orientation in the target data corresponding to the multiple time nodes meets the preset orientation rule. If yes, mark the target data as lane change data.

[0145] Among them, in actual driving scenarios, target data that meets the rough judgment may be that the target vehicle is in a turning scene or on a non-straight road. Therefore, it is necessary to further use the vehicle orientation to make a detailed judgment on the target data to determine whether the target data is lane change data.

[0146] As a feasible implementation, step 2 may specifically include:

[0147] Step 2-1: Determine whether a difference in vehicle orientation in target data corresponding to a first time node and a last time node among the multiple time nodes is greater than a preset orientation difference.

[0148] Among them, the preset direction difference can be determined according to the actual driving scenario and is not specifically limited here.

[0149] Among them, when the target vehicle changes lanes in a straight lane, the vehicle orientation before and after the lane change will not exceed the preset orientation difference; if it exceeds the preset orientation difference, it proves that the road where the target vehicle is located is not a straight road, and it is necessary to further determine whether there is a lane change behavior.

[0150] Step 2-2: If the difference is greater than the preset direction difference, the lateral displacement between each two adjacent time nodes in the multiple time nodes is obtained respectively, and all lateral displacements are added together to obtain a total lateral displacement; determine whether the total lateral displacement is greater than a first preset lateral displacement, and if so, mark the target data as lane change data.

[0151] When the difference in vehicle orientation in the target data corresponding to the first time node and the last time node is greater than the preset orientation difference, the lane where the target vehicle is located is a non-straight road. It is necessary to determine whether a lane change occurs based on the lateral displacement in the case of a non-straight road. Specifically, the lateral displacement of every two adjacent time nodes can be calculated, and all lateral displacements are added together to obtain the total lateral displacement. If the total lateral displacement is greater than the first preset lateral displacement, a lane change occurs, and the target data is the lane change data.

[0152] As a possible implementation method, see Figure 6 , Figure 6 A schematic diagram of the lateral displacement of two adjacent time nodes provided in the embodiment of the present application, combined with Figure 6 As shown, the process of determining the lateral displacement of two adjacent time nodes can be:

[0153] The four time nodes are determined to be t1, t2, t3, and t4.

[0154] The process of determining the lateral displacement between t1 and t2 is as follows: determine the straight line K1 of the vehicle orientation of the target vehicle corresponding to t1, and determine the distance a between the positioning point of the target vehicle corresponding to t2 and the straight line K1, where a is the lateral displacement between t1 and t2.

[0155] The process of determining the lateral displacement between t2 and t3 is as follows: determine the center line K2 of the angle between the vehicle orientations of the target vehicles corresponding to t1 and t4, make a perpendicular line K3 to the center line K2 through the positioning point of the target vehicle corresponding to t2, make a perpendicular line K4 to the perpendicular line K3 through the positioning point of the target vehicle corresponding to t3, the foot of the perpendicular is point M, determine that the distance between point M and the positioning point of the target vehicle corresponding to t2 is b, and b is the lateral displacement between t2 and t3.

[0156] The process of determining the lateral displacement between t3 and t4 is as follows: draw a perpendicular line K4 through the positioning point of the target vehicle corresponding to t4 and the vehicle direction of the target vehicle corresponding to t4, draw a perpendicular line K5 through the positioning point of the target vehicle corresponding to t3 and the perpendicular line K4, the foot of the perpendicular is point N, and the distance between point N and the positioning point of the target vehicle corresponding to t4 is determined to be c, where c is the lateral displacement between t3 and t4.

[0157] The first preset lateral displacement is the lateral displacement when the target lane changes lanes in a non-straight road situation, which can be determined according to actual conditions and is not specifically limited here.

[0158] Step 2-3: If the difference is less than or equal to the preset direction difference, determine whether the lateral displacement between the first time node and the last time node is greater than a second preset lateral displacement. If greater, mark the time window as lane change data.

[0159] When the difference in the vehicle orientation in the target data corresponding to the first time node and the last time node is less than the preset orientation difference, the lane where the target vehicle is located is a straight road. At this time, it is only necessary to determine whether the lateral displacement between the first time node and the last time node is greater than the second preset lateral displacement to determine whether the target data is lane change data.

[0160] As a possible implementation method, see Figure 7 , Figure 7 A schematic diagram of the lateral displacement between the first time node and the last time node provided in the embodiment of the present application, combined with Figure 7 As shown, the process of determining the lateral displacement between the first time node and the last time node can be:

[0161] Four time nodes are determined, which are t1, t2, t3, and t4 in chronological order. After determining that the difference between the vehicle orientations at the two time points t1 and t4 is less than the preset orientation difference, the positioning point of the target vehicle corresponding to t4 is made parallel to the center line of the lane where the target vehicle is located, and the distance d between the positioning point of the target vehicle corresponding to t1 and the parallel line L is the lateral displacement.

[0162] Among them, the determination process of the lateral displacement between the first time node and the last time node and Figure 7 It is only used as an example and explanation and is not a basis for limiting the scope of protection of this application.

[0163] The second preset lateral displacement is the lateral displacement when the target lane changes lanes in a straight road situation, which can be determined according to actual conditions and is not specifically limited here.

[0164] In the process of the target vehicle traveling, lateral displacement will occur regardless of whether a lane change occurs. In the process of determining whether a lane change occurs, the lateral displacement of the target vehicle is greater than the minimum lateral displacement during the lane change to determine whether a lane change occurs.

[0165] In this embodiment, the direction along the lane is defined as the longitudinal direction, and the direction perpendicular to the longitudinal direction is defined as the lateral direction.

[0166] The lane change data positioning process in the optimization method for the automatic driving model provided in the embodiment of the present application divides the actual driving data into multiple time windows, and makes a rough judgment based on the steering information in the actual driving data corresponding to each time window, that is, the actual driving data corresponding to the time window that meets the preset steering rule is marked as target data, and then a fine judgment is made on the target data, that is, the target data that meets the preset orientation rule is marked as lane change data. Therefore, the staff only needs to judge the lane change data that has passed the rough judgment and the fine judgment to determine whether there is a lane change behavior in the lane change data. There is no need to manually check a large amount of actual driving data, and the lane change data can be automatically positioned, which saves the time of manual extraction of lane change data and improves the efficiency of lane change data extraction.

[0167] Through the positioning process of lane change data in the optimization method for the autonomous driving model provided by the above embodiment, the lane change is obtained, and the final judgment is made manually to obtain accurate lane change data. When the accurate lane change data is applied to the optimization of the autonomous driving model, the decision-making accuracy of the optimized autonomous driving model will be higher.

[0168] Based on the optimization method for the autonomous driving model provided in the above embodiment, the present application embodiment also provides an optimization device for the autonomous driving model, see Figure 8 , Figure 8 A schematic diagram of a structure of an optimization device for an automatic driving model provided in an embodiment of the present application, combined with Figure 8 As shown, the optimization device for the automatic driving model provided in the embodiment of the present application may include:

[0169] The data acquisition unit 801 is used to acquire vehicle data of a target vehicle and environmental data where the target vehicle is located;

[0170] An input unit 802, used for inputting the vehicle data and the environment data into the automatic driving model to be optimized;

[0171] an output result determination unit 803, configured to determine whether to change lanes according to the vehicle data and the environment data, and use the result of whether to change lanes as the output result of the automatic driving model to be optimized;

[0172] A result comparison unit 804 is used to compare the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle;

[0173] The autonomous driving model optimization unit 805 is used to optimize the autonomous driving model to be optimized according to the comparison result.

[0174] As a feasible implementation, the autonomous driving model optimization unit 805 may specifically include:

[0175] A comparison result accumulation unit, used for accumulating a preset number of comparison results;

[0176] A ratio determination unit, used to determine the ratio of the comparison results in which the output result is the same as the actual lane change result in the preset number of comparison results;

[0177] The autonomous driving model optimization subunit is used to determine whether the proportion is greater than or equal to a preset proportion value. If so, the optimization of the autonomous driving model is completed; if not, the parameters of the autonomous driving model to be optimized are adjusted, and the vehicle data and the environmental data are continued to be used to optimize the autonomous driving model to be optimized.

[0178] As an optional implementation, the device may further include:

[0179] An actual driving data acquisition unit, used to acquire actual driving data of a target vehicle, wherein the actual driving data includes steering information data, vehicle orientation data, target distance data, and preset distance data of the target vehicle, wherein the target distance data is obtained by the distance between the target vehicle and the lane where the target vehicle is located;

[0180] A lane change data determination unit is used to obtain the lane change data of the target vehicle according to at least two types of data in the actual driving data.

[0181] As an optional implementation manner, when the actual driving data is the target distance data and the preset distance data, the lane change data determination unit may specifically include:

[0182] A relative size detection unit, used to detect the relative size between the target distance and the preset distance in real time when the target vehicle moves;

[0183] a lane change start time determination unit, configured to record the current time as the lane change start time of the target vehicle and the lane in which the target vehicle is located at the lane change start time when detecting that the target distance is greater than the preset distance;

[0184] A lane determination unit, configured to continuously detect the relative size of the target distance and the preset distance after recording the starting lane change time of the target vehicle, and determine whether the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the starting lane change time;

[0185] a lane change end time determination unit, configured to record the current time as the lane change end time of the target vehicle and the lane where the target vehicle is located at the lane change end time when it is detected that the target distance is less than a preset distance and the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the start lane change time;

[0186] The first lane change data determination subunit is used to obtain the lane change data of the target vehicle according to the starting lane change time, the ending lane change time, the lane where the target vehicle is located at the starting lane change time, and the lane where the target vehicle is located at the ending lane change time.

[0187] As another optional implementation, when the actual driving data is the steering information and the vehicle direction, the lane change data determination unit may specifically include:

[0188] An actual driving data acquisition subunit, used to acquire actual driving data of a target vehicle, wherein the actual driving data includes steering information and vehicle orientation;

[0189] A time window division unit, used to divide the actual driving data into a plurality of time windows to obtain driving data corresponding to the plurality of time windows respectively;

[0190] A target data determination unit, configured to use the actual driving data corresponding to each time window, the actual driving data whose wheel turning information conforms to a preset wheel turning rule, as target data;

[0191] A pending lane change data determination unit is used to determine whether the vehicle orientation in each target data meets the preset orientation rule, and if so, mark the target data as pending lane change data;

[0192] The second lane change data determination subunit is used to manually determine whether the pending lane change data has a lane change behavior, and use the pending lane change data that has a lane change behavior as the lane change data.

[0193] The optimization device for the autonomous driving model provided in the embodiment of the present application has the same beneficial effects as the optimization method for the autonomous driving model provided in the above-mentioned embodiment, so they are not described in detail.

[0194] Based on the optimization method and device for the autonomous driving model provided in the aforementioned embodiments, accordingly, the present application also provides an optimization device for the autonomous driving model, including a processor and a memory; the memory is used to store a computer program; the processor is used to execute the optimization method for the autonomous driving model provided in the aforementioned method embodiment according to the computer program.

[0195] Based on the optimization methods, devices and equipment for the autonomous driving model provided in the aforementioned embodiments, the present application also provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, executes the optimization method for the autonomous driving model provided in the aforementioned method embodiments.

[0196] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and equipment embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and equipment embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without paying creative labor.

[0197] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0198] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An optimization method for an automatic driving model, characterized in that: The method comprises: Acquire vehicle data of a target vehicle and environmental data where the target vehicle is located; Inputting the vehicle data and the environment data into an autonomous driving model to be optimized; The autonomous driving model determines whether to change lanes according to the vehicle data and the environmental data, and uses the result of whether to change lanes as the output result of the autonomous driving model to be optimized; Comparing the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle; Optimizing the automatic driving model to be optimized according to the comparison result; The lane change data is obtained in the following manner: Acquiring actual driving data of a target vehicle, wherein the actual driving data includes steering information and vehicle orientation of the target vehicle; Dividing the actual driving data into multiple time windows to obtain driving data corresponding to the multiple time windows respectively; Determine whether the steering information in the actual driving data corresponding to each time window has a steering value that exceeds a preset upper limit value and a preset lower limit value, respectively; if so, mark the actual driving data corresponding to the time window as target data; Determine a plurality of time nodes at which the wheel turning value in each target data is equal to the preset error value, wherein the plurality of time nodes are arranged in chronological order; Determine whether a difference in vehicle orientation in target data corresponding to a first time node and a last time node among the multiple time nodes is greater than a preset orientation difference, and if the difference is greater than the preset orientation difference, respectively obtain a lateral displacement between each two adjacent time nodes among the multiple time nodes, and add all lateral displacements to obtain a total lateral displacement; determining whether the total lateral displacement is greater than a first preset lateral displacement, and if so, marking the target data as lane change data; If the difference is less than or equal to the preset direction difference, it is determined whether the lateral displacement between the first time node and the last time node is greater than a second preset lateral displacement. If so, the time window is marked as lane change data.

2. The optimization method for the automatic driving model according to claim 1, characterized in that: Optimizing the automatic driving model to be optimized according to the comparison result includes: Accumulate and obtain a preset number of comparison results; Determine the ratio of the comparison results that the output result is the same as the actual lane change result in the preset number of comparison results; Determine whether the proportion is greater than or equal to a preset proportion value. If so, complete the optimization of the autonomous driving model; if not, adjust the parameters of the autonomous driving model to be optimized, and continue to use the vehicle data and the environmental data to optimize the autonomous driving model to be optimized.

3. The optimization method for the automatic driving model according to claim 1, characterized in that: The actual driving data also includes the target distance and the lane where the target vehicle is located; The target distance is the shortest distance between the positioning point of the target vehicle and the center line of the lane where the target vehicle is located; the method further includes: The lane change data of the target vehicle is obtained according to the target distance and the lane where the target vehicle is located.

4. The optimization method for the automatic driving model according to claim 3, characterized in that: Obtaining lane change data of the target vehicle according to the target distance and the lane where the target vehicle is located, including: When the target vehicle moves, the relative size of the target distance and the preset distance is detected in real time; When it is detected that the target distance is greater than the preset distance, the current time is recorded as the starting lane change time of the target vehicle, and the lane in which the target vehicle is located at the starting lane change time is recorded; After recording the starting lane change time of the target vehicle, continuously detecting the relative size of the target distance and the preset distance, and determining whether the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the starting lane change time; When it is detected that the target distance is less than the preset distance, and the lane where the target vehicle is located is inconsistent with the lane where the target vehicle is located at the start lane change time, the current time is recorded as the end lane change time of the target vehicle, and the lane where the target vehicle is located at the end lane change time is recorded; The lane change data of the target vehicle is obtained according to the starting lane change time, the ending lane change time, the lane where the target vehicle is located at the starting lane change time, and the lane where the target vehicle is located at the ending lane change time.

5. An optimization device for an automatic driving model, characterized in that: The device comprises: a data acquisition unit, an input unit, an output result determination unit, a result comparison unit and an automatic driving model optimization unit; The data acquisition unit is used to acquire vehicle data of a target vehicle and environmental data where the target vehicle is located; The input unit is used to input the vehicle data and the environment data into the automatic driving model to be optimized; The output result determination unit is used to determine whether to change lanes according to the vehicle data and the environmental data, and use the result of whether to change lanes as the output result of the automatic driving model to be optimized; The result comparison unit is used to compare the output result with the actual lane change result to obtain a comparison result; wherein the actual lane change result is obtained through the lane change data of the target vehicle; The autonomous driving model optimization unit is used to optimize the autonomous driving model to be optimized according to the comparison result; The device further includes a lane change data determination unit, and the lane change data determination unit includes: An actual driving data acquisition subunit, used to acquire actual driving data of a target vehicle, wherein the actual driving data includes steering information and vehicle orientation of the target vehicle; A time window division unit, used to divide the actual driving data into a plurality of time windows to obtain driving data corresponding to the plurality of time windows respectively; a target data determination unit, for taking the actual driving data corresponding to each time window, whose wheel turning information conforms to the preset wheel turning rule, as the target data, including: determining whether the wheel turning information in the actual driving data corresponding to each time window has a wheel turning value exceeding a preset upper limit value and a preset lower limit value, respectively, and if so, taking the actual driving data corresponding to the time window as the target data; The pending lane change data determination unit is used to determine whether the vehicle orientation in each target data meets the preset orientation rule, and if so, mark the target data as pending lane change data, including: determining a plurality of time nodes at which the steering value in each target data is equal to the preset error value, wherein the plurality of time nodes are arranged in chronological order; determining whether the difference between the vehicle orientation in the target data corresponding to the first time node and the last time node in the plurality of time nodes is greater than the preset orientation difference, and if the difference is greater than the preset orientation difference, respectively obtaining the lateral displacement between each two adjacent time nodes in the plurality of time nodes, and adding all the lateral displacements to obtain the total lateral displacement; determining whether the total lateral displacement is greater than the first preset lateral displacement, and if so, marking the target data as pending lane change data; and if the difference is less than or equal to the preset orientation difference, determining whether the lateral displacement between the first time node and the last time node is greater than the second preset lateral displacement, and if so, marking the time window as pending lane change data; The lane change data determination subunit is used to manually determine whether the pending lane change data has a lane change behavior, and use the pending lane change data that has a lane change behavior as the lane change data.

6. The optimization device for the automatic driving model according to claim 5, characterized in that: The autonomous driving model optimization unit comprises: A comparison result accumulation unit, used for accumulating a preset number of comparison results; A ratio determination unit, configured to determine a ratio of the comparison results in which the output result is the same as the actual lane change result in the preset number of comparison results; The autonomous driving model optimization subunit is used to determine whether the proportion is greater than or equal to a preset proportion value. If so, the optimization of the autonomous driving model is completed; if not, the parameters of the autonomous driving model to be optimized are adjusted, and the vehicle data and the environmental data are continued to be used to optimize the autonomous driving model to be optimized.

7. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the optimization method for the autonomous driving model as described in any one of claims 1 to 4 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the optimization method for the autonomous driving model as described in any one of claims 1 to 4.

Images