Autonomous driving decision-making method, model training method, related devices and electronic equipment

Abstract

The present disclosure provides an autonomous driving decision-making method, a model training method, related devices and electronic equipment, which relate to the field of data processing technology, and specifically to the field of autonomous driving technology. The specific implementation scheme is as follows: when it is detected that a vehicle interacts with a first obstacle, a first scene feature and a first state of the vehicle interacting with the first obstacle are obtained; a first interaction key parameter corresponding to the first scene feature is determined, and the first interaction key parameter is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle; based on the first state, a first game tree is constructed to simulate the interaction between the vehicle and the first obstacle; when the interaction simulation between the vehicle and the first obstacle is completed, the target behavior decision of the vehicle when interacting with the first obstacle is determined based on the first game tree and the first interaction key parameter.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, in particular to the field of autonomous driving technology, and specifically to an autonomous driving decision-making method, a model training method, related devices, and electronic equipment. Background Art

[0002] Vehicles, such as autonomous vehicles, need to output safe and reasonable driving decisions, such as overtaking and giving way, while interacting with surrounding traffic participants. This is very important in the field of autonomous driving.

[0003] Currently, the way a vehicle makes autonomous driving decisions usually uses the predicted behavior of surrounding traffic participants to plan the vehicle's driving decisions. Summary of the Invention

[0004] The present disclosure provides an autonomous driving decision-making method, a model training method, related devices and electronic equipment.

[0005] According to a first aspect of the present disclosure, there is provided a method for autonomous driving decision-making, comprising:

[0006] When detecting that the vehicle interacts with the first obstacle, obtaining a first scene feature and a first state when the vehicle interacts with the first obstacle;

[0007] determining a first interaction key parameter corresponding to the first scene feature, where the first interaction key parameter is used to evaluate a driving behavior of the vehicle when interacting with the first obstacle;

[0008] Based on the first state, constructing a first game tree for simulating the interaction between the vehicle and the first obstacle;

[0009] When the interaction simulation between the vehicle and the first obstacle is completed, a target behavior decision of the vehicle when interacting with the first obstacle is determined based on the first game tree and the first interaction key parameter.

[0010] According to a second aspect of the present disclosure, a model training method is provided, comprising:

[0011] Acquire a simulation set, where the simulation set includes M scene samples of the vehicle interacting with the obstacle, where M is an integer greater than 1;

[0012] Determining M second interaction key parameters corresponding to the M scene samples;

[0013] Evaluating the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter, the target scenario sample being a scenario sample in which traffic efficiency and / or safety do not meet standards under the vehicle's third behavior decision, and the game tree is constructed based on the states of the vehicle and the obstacle during the interaction;

[0014] Determining a third interaction key parameter corresponding to the target scene sample, wherein the third interaction key parameter is an interaction key parameter indicating that an interaction result between the vehicle and the obstacle under the target scene sample meets standards in terms of traffic efficiency and safety;

[0015] Based on the scene features corresponding to the target scene sample and the third interaction key parameter, the model parameters of the target model are updated, and the target model is used to characterize the mapping relationship between the scene features and the interaction key parameters of the interaction between the vehicle and the obstacle.

[0016] According to a third aspect of the present disclosure, there is provided an autonomous driving decision-making device, comprising:

[0017] A first acquisition module is configured to, when detecting that an interaction exists between a vehicle and a first obstacle, acquire a first scene feature and a first state of the interaction between the vehicle and the first obstacle;

[0018] a first determining module, configured to determine a first interaction key parameter corresponding to the first scene feature, wherein the first interaction key parameter is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle;

[0019] A construction module, configured to construct a first game tree for simulating the interaction between the vehicle and the first obstacle based on the first state;

[0020] The second determination module is used to determine the target behavior decision of the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter when the interaction simulation between the vehicle and the first obstacle is completed.

[0021] According to a fourth aspect of the present disclosure, a model training device is provided, comprising:

[0022] A second acquisition module is used to acquire a simulation set, where the simulation set includes M scene samples of the interaction between the vehicle and the obstacle, where M is an integer greater than 1;

[0023] a fourth determining module, configured to determine M second interaction key parameters corresponding to the M scene samples;

[0024] an evaluation module, configured to evaluate the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter, wherein the target scenario sample is a scenario sample in which the traffic efficiency and / or safety do not meet the requirements under the vehicle's third behavior decision, and the game tree is constructed based on the states of the vehicle and the obstacle during the interaction;

[0025] a fifth determining module, configured to determine a third interaction key parameter corresponding to the target scene sample, wherein the third interaction key parameter is an interaction key parameter indicating that the interaction result between the vehicle and the obstacle in the target scene sample meets the requirements in terms of traffic efficiency and safety;

[0026] An updating module is used to update the model parameters of the target model based on the scene features corresponding to the target scene sample and the third interaction key parameters, wherein the target model is used to characterize the mapping relationship between the scene features and the interaction key parameters of the interaction between the vehicle and the obstacle.

[0027] According to a fifth aspect of the present disclosure, there is provided an electronic device, including:

[0028] at least one processor; and

[0029] a memory communicatively connected to at least one processor; wherein,

[0030] The memory stores instructions that can be executed by at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform any one of the methods in the first aspect or any one of the methods in the second aspect.

[0031] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute any one of the methods in the first aspect, or execute any one of the methods in the second aspect.

[0032] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements any one of the methods in the first aspect when executed by a processor, or implements any one of the methods in the second aspect when executed.

[0033] According to an eighth aspect of the present disclosure, an autonomous driving vehicle is provided, comprising the electronic device as described in the fifth aspect.

[0034] The technology disclosed herein solves the problem of low accuracy in behavioral decisions made by autonomous vehicles when interacting with obstacles, thereby improving the accuracy of behavioral decisions made by autonomous vehicles.

[0035] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are used to better understand the present invention and do not constitute a limitation of the present invention.

[0037] Figure 1 is a flowchart of an autonomous driving decision-making method according to the first embodiment of the present disclosure;

[0038] Figure 2 This is a schematic diagram of the structure of the first game tree;

[0039] Figure 3 is a flowchart of a model training method according to the second embodiment of the present disclosure;

[0040] Figure 4 It is a flowchart of iterative training of the target model;

[0041] Figure 5 is a structural diagram of an automatic driving decision-making device according to a third embodiment of the present disclosure;

[0042] Figure 6 is a structural diagram of a model training device according to a fourth embodiment of the present disclosure;

[0043] Figure 7 is a schematic block diagram of an example electronic device for implementing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0044] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0045] First embodiment

[0046] like Figure 1 As shown, the present disclosure provides an autonomous driving decision-making method, comprising the following steps:

[0047] Step S101: when it is detected that a vehicle interacts with a first obstacle, a first scene feature and a first state of the vehicle interacting with the first obstacle are acquired.

[0048] In this embodiment, the autonomous driving decision-making method relates to the field of data processing technology, and in particular to the field of autonomous driving technology, and can be widely applied in autonomous driving scenarios. The autonomous driving decision-making method of the disclosed embodiment can be executed by the autonomous driving decision-making device of the disclosed embodiment. The autonomous driving decision-making device of the disclosed embodiment can be configured in any electronic device to execute the autonomous driving decision-making method of the disclosed embodiment. The electronic device can be deployed in an autonomous vehicle to make behavioral decisions for the autonomous vehicle during driving.

[0049] The vehicle may be an autonomous driving vehicle, which may be referred to as a master vehicle, and the first obstacle may be an obstacle around the master vehicle, which may be a motor vehicle, which may be referred to as a slave vehicle.

[0050] The autonomous driving decision-making device can detect whether there are any slave vehicles interacting with the master vehicle. If so, the device can make autonomous driving decisions for the master vehicle to ensure safe driving. Specifically, the device can determine whether the master vehicle is interacting with any of the slave vehicles by detecting their positions, driving status, and other information. For example, if the two vehicles are detected to be relatively close, or if the acceleration of a slave vehicle is relatively high, the device can determine that the master vehicle is interacting with any of the slave vehicles.

[0051] When an interaction between a vehicle and a first obstacle is detected, a first scene feature and a first state of the interaction between the vehicle and the first obstacle can be obtained. The first state may include: the driving state and position of the vehicle during the interaction, and the motion state and position of the first obstacle (when the first obstacle is a follower vehicle, its motion state is its driving state). The driving state may include speed, acceleration, angle, angular velocity, etc. The first scene feature may include the speed of the main vehicle during the interaction, the map lane topology, etc.

[0052] The autonomous driving decision-making device can obtain the first scene characteristics and the first state when the vehicle interacts with the first obstacle by deploying detection modules (such as radar, inertial navigation sensors, etc.) and map applications.

[0053] Step S102: determining a first interaction key parameter corresponding to the first scene feature, where the first interaction key parameter is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle.

[0054] In this step, the first interaction key parameter can be a parameter for evaluating the strategy of behavioral decision-making when the autonomous driving vehicle interacts, and is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle.

[0055] The first interaction key parameters may include: the time step T for constructing the first game tree simulating the interaction between the vehicle and the first obstacle, standard parameters for scoring the results simulated by the first game tree (i.e., leaf nodes), such as the minimum distance for interaction safety in the scoring criteria, or the weight of the interaction safety item, etc.

[0056] Different time steps T, different constructed first game trees, and different standard parameters result in different scores for the results simulated by the first game tree. In this step, a first interaction key parameter corresponding to the first scene feature can be determined, and the first scene feature can be matched to the interaction scene between the vehicle and the first obstacle. This allows the autonomous driving decision-making device to construct a game tree suitable for simulating the interaction scene between the host vehicle and the first obstacle, and accurately evaluate the results simulated by the game tree, thereby making accurate behavioral decisions based on the evaluation results.

[0057] For example, if the interaction scenario indicates that the host vehicle is moving very fast, its driving state changes significantly. In this interaction scenario, the time step T can be relatively short so that the constructed game tree can more accurately simulate the interaction scenario. Conversely, the time step T can be adaptively increased to simplify the constructed game tree. In addition, if the interaction scenario indicates that the host vehicle is moving very fast, the standard parameters such as the minimum distance for interactive safety and the weight of the interactive safety item in the scoring criteria can be set relatively large. Conversely, these standard parameters can be adaptively reduced to ensure the accuracy of the scoring results.

[0058] For example, if the interactive scenario indicates a complex lane topology on the map, lane positions change rapidly. In this interactive scenario, the time step T can be shorter. Otherwise, it can be adaptively increased. Furthermore, if the interactive scenario indicates a complex lane topology on the map, standard parameters such as the minimum distance for interactive safety and the weight of the interactive safety item in the scoring criteria can be set higher. Otherwise, these standard parameters can be adaptively reduced to ensure the accuracy of the scoring results.

[0059] In an optional embodiment, the autonomous driving decision-making device may store a mapping table of scene features and interaction key parameters, and may determine the first interaction key parameter corresponding to the first scene feature by looking up the table.

[0060] In another optional embodiment, the first scene feature can be input into a target model to obtain a first interaction key parameter output by the target model. The target model can be a machine learning model, such as a regression model, and is used to characterize the mapping relationship between scene features of the vehicle-obstacle interaction and the interaction key parameter. The target model can be pre-trained, and its training process is described in detail in the following embodiments.

[0061] Step S103: Based on the first state, construct a first game tree for simulating the interaction between the vehicle and the first obstacle.

[0062] In this step, the first state may be referred to as an initial state, which includes the driving states of the host vehicle and the first obstacle when they interact (which may be referred to as time t0), and may also include the positions of the host vehicle and the first obstacle at time t0.

[0063] When interaction between the master vehicle and the first obstacle is detected, this initial state can be used as the node state of the root node of the first game tree. Starting from this initial state, the driving of the master vehicle and the slave vehicle is simulated according to their driving behaviors to simulate the interaction process between the master vehicle and the slave vehicle. The hierarchical nodes of the first game tree are constructed based on the states of the master vehicle and the slave vehicle during the interaction process.

[0064] The number of layers in the first game tree is determined based on a combination of the interaction simulation time between the host vehicle and the first obstacle and the time step T. The longer the interaction simulation time, the more layers of the game tree constructed using the same time step T. The shorter the time step T, the more layers of the game tree constructed within the same interaction simulation time. The time step T can be pre-set or determined based on the first scenario characteristics before constructing the first game tree. For example, the time step T can be included in the first interaction key parameters.

[0065] It should be noted that the driving behaviors of the master car and the slave car can be pre-set. In the process of simulating the interaction between the master car and the slave car, the driving behavior of the master car can correspond to the driving behavior that the master car may decide in the driving scenario, and the driving behavior of the slave car can correspond to the driving behavior that the slave car may decide. Since the driving behaviors that the master car and the slave car may decide in the interactive scenario include at least one, the at least one driving behavior that the master car may decide is combined with the at least one driving behavior that the slave car may decide to obtain multiple behavior combinations. Each behavior combination represents an interaction situation between the master car and the slave car. Different behavior combinations have different interaction situations. Accordingly, after simulating the master car and the slave car driving according to a behavior combination at the parent node (the root node can also be called a parent node), a child node can be constructed based on the state after driving. This construction is repeated until the interaction simulation between the master car and the slave car is completed.

[0066] Step S104: When the interaction simulation between the vehicle and the first obstacle is completed, a target behavior decision of the vehicle when interacting with the first obstacle is determined based on the first game tree and the first interaction key parameter.

[0067] In this step, the target behavior decision can be the actual driving behavior determined by the vehicle during its interaction with the first obstacle, such as overtaking, yielding, or observing. The vehicle can actually interact with the target behavior decision and the first obstacle to achieve safe and reasonable interaction and improve user experience.

[0068] The automatic driving decision-making device can determine whether the interactive simulation between the master vehicle and the slave vehicle has ended based on information such as the positions of the master vehicle and the slave vehicle during the interactive simulation, the driving status of the master vehicle and the slave vehicle, etc. For example, if it is detected that the distance between the two positions is relatively close during the simulation, or the acceleration of the slave vehicle is relatively large, it can be determined that the interactive simulation between the master vehicle and the slave vehicle has ended.

[0069] When the interaction simulation between the vehicle and the first obstacle is completed, the target behavior decision of the vehicle when interacting with the first obstacle can be determined based on the first game tree and the first interaction key parameter. Specifically, the autonomous driving decision-making device can score the interaction results (i.e., leaf nodes) simulated by the first game tree based on the first interaction key parameter to obtain a score value for each interaction result. From the leaf node corresponding to the interaction result with the highest score, the first game tree is traced back to the root node. The action from the root node to the first-level hierarchical node under the corresponding path is the optimal action, which is converted into an autonomous driving behavior decision through specific rules to obtain a target behavior decision.

[0070] For example, when the optimal action indicates that the acceleration of the host vehicle is greater than a threshold, the behavioral decision is converted to overtaking; when the optimal action indicates that the deceleration of the host vehicle is greater than a threshold, the behavioral decision is converted to yielding; when the optimal action indicates that the acceleration of the host vehicle is zero, the behavioral decision is converted to observing.

[0071] In one scenario, the master vehicle interacts with a slave vehicle. In this case, the target behavior decision of the vehicle when interacting with the first obstacle can be directly determined based on the first game tree and the first interaction key parameter.

[0072] In another scenario, the master vehicle interacts with multiple slave vehicles, such as when the master vehicle merges into traffic. The autonomous driving decision-making device can determine the key interaction parameters corresponding to the scene characteristics of the interaction with different slave vehicles, and construct game trees that simulate the interaction between the master vehicle and different slave vehicles. In this way, the master vehicle's behavioral decisions can be determined separately for the interaction with different slave vehicles, and the target behavioral decisions can be determined by integrating the various behavioral decisions of the master vehicle.

[0073] For example, in a scenario where the main vehicle merges into traffic, if the main vehicle's behavioral decisions are overtaking and yielding, the target behavioral decision instructs the main vehicle to merge between the two follower vehicles. If the main vehicle's behavioral decisions are all overtaking, the target behavioral decision instructs the main vehicle to merge in front of the two follower vehicles. If the main vehicle's behavioral decisions are all yielding, the target behavioral decision instructs the main vehicle to merge behind the two follower vehicles.

[0074] In this embodiment, by acquiring scene features during the vehicle's interaction with a first obstacle, determining key interaction parameters corresponding to these scene features, and modeling future interactions between the vehicle and the first obstacle, a game tree is generated to simulate the vehicle's interaction with the first obstacle. The game tree and key interaction parameters are then used to determine the vehicle's behavior during the interaction with the first obstacle. This approach, by simulating possible future interactions using the interaction game tree, addresses the behavioral decision-making problem of the master vehicle in scenarios with strong interactions between the master and slave vehicles, improving the accuracy of autonomous driving decisions.

[0075] Furthermore, by determining key interaction parameters corresponding to scenario characteristics, the behavioral decision-making strategies used by autonomous vehicles during interactions can be evaluated. This allows for accurate evaluation of the results of game tree simulations, thereby improving the accuracy of autonomous driving decisions. Furthermore, this method can improve the efficiency of autonomous driving decisions and can be applied to large-scale autonomous driving decision-making scenarios.

[0076] Optionally, step S102 specifically includes:

[0077] Inputting the first scene feature into a target model to obtain the first interaction key parameter output by the target model;

[0078] The target model is used to characterize the mapping relationship between the scene characteristics of the interaction between the vehicle and the obstacle and the key interaction parameters.

[0079] In this embodiment, a first scene feature can be input into a target model to obtain a first interaction key parameter output by the target model. The target model can be a machine learning model, such as a regression model, and is used to characterize the mapping relationship between scene features of the vehicle-obstacle interaction and the interaction key parameter. The target model can be pre-trained, and its training process is described in detail in the following embodiments.

[0080] Compared with manually adjusting the key interaction parameters to match the interaction scenario, it can greatly improve the efficiency of autonomous driving decision-making, can be used in large-scale autonomous driving decision-making scenarios, and can also improve the accuracy of determining the key interaction parameters and the application generalization in the interaction scenario.

[0081] Optionally, the first interaction key parameter includes a time step, and step S103 specifically includes:

[0082] Determining, based on the first state and a preset behavior combination, a second state at each moment of the time step starting from the interaction, the behavior combination including: driving behaviors of the vehicle and the first obstacle, respectively, within the time step; and the second state including: a state after simulating the vehicle and the first obstacle, respectively, driving according to the driving behaviors under the behavior combination;

[0083] The first game tree is constructed based on the first state and the second state, where the first state is the state of the root node of the first game tree, and the second state is the state of the hierarchical nodes of the first game tree.

[0084] In this embodiment, the first interaction key parameter may include a time step T. Different time steps T result in different first game trees being constructed.

[0085] Starting from the interaction, each moment with a duration of time step T can be expressed as t0+K*time step T, where K is a positive integer.

[0086] The following describes in detail the construction of a game tree by taking two adjacent moments (the first moment and the second moment) as an example.

[0087] The first moment is different, and the number of layers of the corresponding first game tree is different. For example, when the first moment is t0, it corresponds to the root node, and the node state of the root node is the initial state of the main vehicle and the first obstacle at t0. When the first moment is (t0 + time step T), it corresponds to the first level node in the first game tree, and the node state of the node is the state of the main vehicle and the first obstacle at (t0 + time step T), and so on.

[0088] The second moment may be a moment starting from the first moment and having a duration of the time step T, that is, the second moment is a moment that differs from the first moment by a time step T.

[0089] The preset behavior combination can be a combination of the driving behaviors of the host vehicle and the first obstacle, respectively, within a time period of time step T. At least one possible driving behavior of the host vehicle and at least one possible driving behavior of the first obstacle can be obtained for the driving scenario corresponding to the first moment. The at least one possible driving behavior of the host vehicle and the at least one possible driving behavior of the first obstacle are combined to obtain a behavior combination. The behavior combination can include a driving behavior determined by the host vehicle and a driving behavior determined by the first obstacle. The driving behaviors can include, for example, longitudinal acceleration and lateral angular velocity of the vehicle's head.

[0090] For example, at least one driving behavior that the main vehicle may decide includes driving behavior A1 and driving behavior A2, and at least one driving behavior that the first obstacle may decide includes driving behavior B1 and driving behavior B2. Then the behavior combinations in the driving scenario at the first moment may include (A1, B1), (A1, B2), (A2, B1) and (A2, B2).

[0091] Different first moments may correspond to different driving scenarios. For example, when the first moment is t0, the corresponding scenario is a straight-ahead scenario, while when the first moment is (t0 + time step T), the corresponding driving scenario is an intersection scenario. Accordingly, the behavior combinations may be different, and the number of behavior combinations may also be different.

[0092] For each behavior combination, based on the driving state of the main vehicle at the first moment, the driving of the main vehicle under the decided driving behavior can be simulated, and the driving state of the main vehicle at the second moment after the time step T can be calculated. Based on the driving state of the first obstacle at the first moment, the driving of the first obstacle under the decided driving behavior can be simulated, and the driving state of the first obstacle at the second moment after the time step T can be calculated to obtain the second state at the second moment.

[0093] The first state can be used as the node state of the root node of the first game tree, and the second state at each moment can be used as the node state of the hierarchical node of the first game tree to construct the first game tree. The hierarchical node constructed at the first moment and the hierarchical node constructed at the second moment are in a parent-child relationship, and the number of child nodes can be determined based on the number of behavior combinations in the driving scenario at the first moment.

[0094] It should be noted that the above description is a process of constructing a set of parent-child nodes in the first game tree based on the second state at the first moment and the second moment. All parent-child nodes in the first game tree can be constructed in the above manner until the simulation of the interaction between the main vehicle and the first obstacle is detected to be completed, and the first game tree for the interaction between the main vehicle and the first obstacle can be obtained accordingly.

[0095] Figure 2 This is a schematic diagram of the structure of the first game tree, such as Figure 2 As shown, the relationship between the root node of the first game tree and the hierarchical nodes of the first layer can be a set of parent-child nodes. The node state of the root node can be represented by S0, and the node state of the hierarchical nodes of the first layer can be represented by S1. j For each node in the first layer of the hierarchy, the nodes connected to it in the second layer can also be a group of parent and child nodes. The node status of the second layer of the hierarchy can be expressed as S2 jIt indicates that as the interaction between the main vehicle and the first obstacle proceeds, the above process of constructing the parent-child node is repeated until the interaction simulation between the main vehicle and the first obstacle is completed.

[0096] In this way, by determining the time step T that matches the interaction scenario, the autonomous driving decision-making device can construct a game tree suitable for simulating the interaction scenario based on the interaction scenario between the main vehicle and the first obstacle, thereby improving the accuracy of the autonomous driving decision.

[0097] Optionally, the first interaction key parameter includes a target standard parameter for scoring leaf nodes in the first game tree, and the step S104 specifically includes:

[0098] Scoring each leaf node in the first game tree based on the target standard parameter to obtain a scoring value of the leaf node;

[0099] Select the leaf node with the highest score from the first game tree to obtain the target leaf node;

[0100] The target behavior decision is determined based on the target driving behavior of the vehicle, where the target driving behavior is the driving behavior from the root node to the hierarchical node of the first layer under the path corresponding to the target leaf node.

[0101] In this embodiment, upon detecting the completion of the interaction simulation between the host vehicle and the first obstacle, a target leaf node can be determined from the first game tree based solely on the node status of the leaf node in the first game tree. The leaf node can be a node at the last level of the first game tree. Alternatively, the target leaf node can be determined from the first game tree based on at least one of the driving status, position, and driving behavior of the vehicle and the first obstacle along the first path. The first path can be the path from the root node to the leaf node.

[0102] All leaf results simulated in the first game tree can be scored, and the scoring criteria can take into account interaction safety, interaction sensation, traffic rules, etc., so that the higher the scoring value of the leaf node, the better the safety of the main vehicle and the first obstacle at the end of the interaction or during the interaction, the better the interaction sensation, and both comply with traffic rules; the lower the scoring value of the leaf node, the worse the safety of the main vehicle and the first obstacle at the end of the interaction or during the interaction, the worse the interaction sensation, and both may violate traffic rules.

[0103] The first interaction key parameter may include target standard parameters for scoring leaf nodes in the first game tree, such as standard parameters such as the minimum distance of interaction safety and the weight of interaction safety items in the scoring criteria.

[0104] Based on the first interactive key parameter, all leaf results simulated in the first game tree can be scored using an existing or new scoring method (such as a scoring equation) to obtain the scoring value of the leaf node, and the leaf node with the highest scoring value can be determined as the target leaf node.

[0105] Afterwards, the driving behavior from the root node to the first-layer hierarchical node under the path corresponding to the target leaf node can be determined as the target driving behavior (i.e., the optimal action), and converted into the behavior decision of autonomous driving through specific rules to obtain the target behavior decision.

[0106] In this way, the interaction results simulated by the game tree can be accurately evaluated based on the automatically determined target standard parameters corresponding to the interaction scenario, so that accurate behavioral decisions can be made based on the evaluation results, and the efficiency of autonomous driving decisions can be improved.

[0107] Optionally, the method further includes:

[0108] In a case where an interaction between the vehicle and a first obstacle is detected, if an interaction between the vehicle and a second obstacle is detected, determining a first behavior decision of the vehicle when interacting with the second obstacle;

[0109] The step S104 specifically includes:

[0110] Determining, based on the first game tree and the first interaction key parameter, a second behavior decision of the vehicle when interacting with the first obstacle;

[0111] The target behavior decision is determined based on the first behavior decision and the second behavior decision.

[0112] In this embodiment, the method for detecting the interaction between the vehicle and the second obstacle is similar to the method for detecting the interaction between the vehicle and the first obstacle, and will not be described in detail here.

[0113] The autonomous driving decision-making device can respectively determine the key interaction parameters corresponding to the scene characteristics of the interaction with different slave vehicles, and respectively construct game trees that simulate the interaction between the master vehicle and different slave vehicles. The method of determining the key interaction parameters and the method of constructing the game tree can be the same. In this way, the behavioral decision of the master vehicle (the first behavioral decision and the second behavioral decision respectively) can be determined for the interaction between the master vehicle and different slave vehicles. After that, the target behavioral decision can be determined by integrating the various behavioral decisions of the master vehicle.

[0114] For example, in a scenario where the main vehicle merges into traffic, if the main vehicle's behavioral decisions are overtaking and yielding, the target behavioral decision instructs the main vehicle to merge between the two follower vehicles. If the main vehicle's behavioral decisions are all overtaking, the target behavioral decision instructs the main vehicle to merge in front of the two follower vehicles. If the main vehicle's behavioral decisions are all yielding, the target behavioral decision instructs the main vehicle to merge behind the two follower vehicles.

[0115] In this way, the target behavior decision of the main vehicle can be accurately determined in complex interactive scenarios, improving the safety and rationality of autonomous driving.

[0116] Second embodiment

[0117] like Figure 3 As shown, the present disclosure provides a model training method, comprising the following steps:

[0118] Step S301: Acquire a simulation set, where the simulation set includes M scene samples of the interaction between the vehicle and the obstacle, where M is an integer greater than 1;

[0119] Step S302: Determine M second interaction key parameters corresponding to the M scene samples;

[0120] Step S303: Evaluate the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter. The target scenario sample is a scenario sample in which the traffic efficiency and / or safety do not meet the requirements under the vehicle's third behavior decision. The game tree is constructed based on the states of the vehicle and the obstacle during the interaction.

[0121] Step S304: determining a third interaction key parameter corresponding to the target scene sample, wherein the third interaction key parameter is an interaction key parameter indicating that the interaction result between the vehicle and the obstacle in the target scene sample meets the requirements in terms of traffic efficiency and safety;

[0122] Step S305: Based on the scene features corresponding to the target scene sample and the third interaction key parameter, the model parameters of the target model are updated, where the target model is used to characterize the mapping relationship between the scene features of the interaction between the vehicle and the obstacle and the interaction key parameters.

[0123] In this embodiment, the key interaction parameters when the vehicle interacts with the obstacle can be automatically determined based on the target model. Before the target model determines the key interaction scene based on the scene features, it needs to be pre-trained. The training process of the target model is described in detail below.

[0124] Specifically, in step S301, the simulation set refers to a set of simulation environments, which may include M scene samples of vehicle-obstacle interactions. Each scene sample corresponds to a simulation environment, and the scene sample is an interaction scene between the vehicle and the obstacle. For example, the host vehicle changes lanes to merge into the traffic flow, or the host vehicle turns left to merge into the traffic flow.

[0125] A pre-stored simulation set may be obtained, or a simulation set may be constructed through simulation software, which is not specifically limited here.

[0126] In step S302, in an optional implementation manner, M second interaction key parameters corresponding to M scene samples may be manually determined by manual parameter adjustment, with one scene sample corresponding to one second interaction key parameter.

[0127] In another optional implementation, scene features corresponding to the scene samples may be recorded and input into a target model, and the target model may output M second interaction key parameters corresponding to the M scene samples.

[0128] Among them, the content included in the second interaction key parameter can be the same as that included in the first interaction key parameter, which will not be described in detail here.

[0129] For each scenario sample, a game tree can be constructed to simulate the interaction between the vehicle and the obstacle in the scenario sample. The construction method of this game tree is similar to that of the first embodiment and is not further described here. Subsequently, the third behavior decision corresponding to the scenario sample can be determined based on the game tree and the second interaction key parameter. M third behavior decisions corresponding to M scenario samples are obtained. The third behavior decision indicates the driving behavior determined by the vehicle during the actual interaction process under the scenario sample.

[0130] Accordingly, an existing or new evaluation method can be used to evaluate the third behavior decision corresponding to each scenario sample to assess whether the vehicle's traffic efficiency and safety under the third behavior decision meet the standards.

[0131] If, under the third behavior decision, the traffic efficiency in the scene sample decreases, such as the third behavior decision instructs the vehicle to observe (i.e., the vehicle stops and does not move), this will affect the traffic efficiency in the interactive scene, and its traffic efficiency will not meet the standard. Or, if, under the third behavior decision, a collision risk will occur in the scene sample, such as the third behavior decision instructs the vehicle to overtake (i.e., the vehicle accelerates), this may affect the safety in the interactive scene, and its safety will not meet the standard.

[0132] If the traffic efficiency and / or safety of the evaluated vehicle under the third behavior decision does not meet the standards, the evaluated scenario sample is determined to be a target scenario sample (ie, a problem scenario sample).

[0133] In step S304, a third key interaction parameter can be found for the target scenario sample using a numerical optimization method, such as Bayesian optimization, so that the vehicle-obstacle interaction result in the target scenario sample meets the requirements for traffic efficiency and safety. The third key interaction parameter can be the optimal key interaction parameter for the target scenario sample, i.e., the interaction result with the highest score when the vehicle-obstacle interaction result in the target scenario sample meets the requirements for traffic efficiency and safety.

[0134] In step S305, the scene features of each target scene sample are recorded, such as the map lane topology, the speed of the main vehicle, etc., and the target model is trained based on the mapping relationship between the scene features and the third interaction key parameters to learn the mapping between the scene features and the interaction key parameters and update the model parameters of the target model.

[0135] In an optional embodiment, target model training may be completed based on updating model parameters of the target model.

[0136] In another optional embodiment, based on updating the model parameters of the target model, steps S302 to S305 may be repeated until the number of problem scenario samples found in the simulation set is minimal, at which point the target model training may be completed.

[0137] In this embodiment, by training the target model and automatically determining the interaction key parameters matching the interaction scene according to the scene characteristics based on the target model, the efficiency of autonomous driving decision-making can be greatly improved compared to manually adjusting the interaction key parameters matching the interaction scene. It can be used in large-scale autonomous driving decision-making scenarios, and can also improve the accuracy of determining the interaction key parameters and the application generalization in the interaction scenario.

[0138] Optionally, step S302 specifically includes:

[0139] Obtaining M scene features corresponding to the M scene samples;

[0140] The M scene features are respectively input into the target model to obtain the M second interaction key parameters output by the target model.

[0141] In this embodiment, the scene features of each scene sample can be recorded, and M scene features can be input into the target model to obtain M second interaction key parameters output by the target model. In this way, based on the updated model parameters of the target model, the target model can be iteratively trained based on the simulation set, repeating steps S302 to S305 until the number of problem scene samples found in the simulation set is minimized. At this point, the target model training can be completed. In this way, iterative training of the target model through the reinforcement learning algorithm can improve the training effect of the target model.

[0142] Optionally, step S304 specifically includes:

[0143] Based on N preset interaction key parameters, score the interaction results between the vehicle and the obstacle in the target scene sample respectively to obtain N score values ​​corresponding to the N interaction key parameters;

[0144] The interaction key parameter corresponding to the highest score value among the N score values ​​is determined as the third interaction key parameter.

[0145] In this embodiment, through the numerical optimization method, under the preset N interaction key parameters, the interaction results between the vehicle and the obstacle in the target scene sample are scored based on the preset interaction key parameters respectively to find the optimal interaction key parameters under the target scene sample, that is, the interaction key parameters with the highest score value. In this way, the accuracy of determining the mapping relationship between scene features and interaction key parameters can be improved.

[0146] like Figure 4 As shown in the figure, it is a flowchart of iterative training of the target model. Figure 4 As shown, the training is continuously iterated under the simulation set, the simulation results are obtained by constructing a game tree under the scenario samples, the problem scenario samples are found by evaluating the simulation results, and the third interaction key parameters corresponding to the problem scenario samples are found through the parameter optimization algorithm. Based on the scenario features corresponding to the problem scenario samples and the third interaction key parameters, the model parameters of the target model are updated.

[0147] Third embodiment

[0148] like Figure 5 As shown, the present disclosure provides an automatic driving decision-making device 500, comprising:

[0149] A first acquisition module 501 is configured to, when detecting that a vehicle interacts with a first obstacle, acquire a first scene feature and a first state of the vehicle interacting with the first obstacle;

[0150] a first determining module 502, configured to determine a first interaction key parameter corresponding to the first scene feature, wherein the first interaction key parameter is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle;

[0151] A construction module 503 is configured to construct a first game tree for simulating the interaction between the vehicle and the first obstacle based on the first state;

[0152] The second determination module 504 is configured to determine, when the interaction simulation between the vehicle and the first obstacle is completed, a target behavior decision for the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter.

[0153] Optionally, the first determining module 502 is specifically configured to:

[0154] Inputting the first scene feature into a target model to obtain the first interaction key parameter output by the target model;

[0155] The target model is used to characterize the mapping relationship between the scene characteristics of the interaction between the vehicle and the obstacle and the key interaction parameters.

[0156] Optionally, the first interaction key parameter includes a time step, and the construction module 503 is specifically configured to:

[0157] Determining, based on the first state and a preset behavior combination, a second state at each moment of the time step starting from the interaction, the behavior combination including: driving behaviors of the vehicle and the first obstacle, respectively, within the time step; and the second state including: a state after simulating the vehicle and the first obstacle, respectively, driving according to the driving behaviors under the behavior combination;

[0158] The first game tree is constructed based on the first state and the second state, where the first state is the state of the root node of the first game tree, and the second state is the state of the hierarchical nodes of the first game tree.

[0159] Optionally, the first interaction key parameter includes a target standard parameter for scoring leaf nodes in the first game tree, and the second determination module 504 is specifically configured to:

[0160] Scoring each leaf node in the first game tree based on the target standard parameter to obtain a scoring value of the leaf node;

[0161] Select the leaf node with the highest score from the first game tree to obtain the target leaf node;

[0162] The target behavior decision is determined based on the target driving behavior of the vehicle, where the target driving behavior is the driving behavior from the root node to the hierarchical node of the first layer under the path corresponding to the target leaf node.

[0163] Optionally, the device further includes:

[0164] a third determining module, configured to, when detecting that the vehicle has interacted with the first obstacle, determine a first behavior decision of the vehicle when interacting with the second obstacle if detecting that the vehicle has interacted with the second obstacle;

[0165] The second determining module 504 is specifically configured to:

[0166] Determining, based on the first game tree and the first interaction key parameter, a second behavior decision of the vehicle when interacting with the first obstacle;

[0167] The target behavior decision is determined based on the first behavior decision and the second behavior decision.

[0168] The autonomous driving decision-making device 500 provided in the present disclosure can implement each process implemented in the embodiment of the autonomous driving decision-making method and can achieve the same beneficial effects. To avoid repetition, it will not be described here.

[0169] Fourth embodiment

[0170] like Figure 6 As shown, the present disclosure provides a model training device 600, comprising:

[0171] A second acquisition module 601 is configured to acquire a simulation set, where the simulation set includes M scene samples of the interaction between the vehicle and the obstacle, where M is an integer greater than 1;

[0172] A fourth determining module 602 is configured to determine M second interaction key parameters corresponding to the M scene samples;

[0173] An evaluation module 603 is configured to evaluate the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter, wherein the target scenario sample is a scenario sample in which the traffic efficiency and / or safety do not meet the requirements under the vehicle's third behavior decision, and the game tree is constructed based on the states of the vehicle and the obstacle during the interaction;

[0174] A fifth determining module 604 is configured to determine a third interaction key parameter corresponding to the target scene sample, where the third interaction key parameter is a key interaction parameter indicating that the interaction result between the vehicle and the obstacle in the target scene sample meets the requirements in terms of traffic efficiency and safety.

[0175] The updating module 605 is used to update the model parameters of the target model based on the scene features corresponding to the target scene sample and the third interaction key parameters. The target model is used to characterize the mapping relationship between the scene features and the interaction key parameters of the interaction between the vehicle and the obstacle.

[0176] Optionally, the fourth determining module 602 is specifically configured to:

[0177] Obtaining M scene features corresponding to the M scene samples;

[0178] The M scene features are respectively input into the target model to obtain the M second interaction key parameters output by the target model.

[0179] Optionally, the fifth determining module 604 is specifically configured to:

[0180] Based on N preset interaction key parameters, score the interaction results between the vehicle and the obstacle in the target scene sample respectively to obtain N score values ​​corresponding to the N interaction key parameters;

[0181] The interaction key parameter corresponding to the highest score value among the N score values ​​is determined as the third interaction key parameter.

[0182] The model training device 600 provided in the present disclosure can implement each process implemented in the embodiment of the model training method and can achieve the same beneficial effects. To avoid repetition, it will not be described here.

[0183] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0184] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0185] Figure 7 A schematic block diagram of an example electronic device that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0186] like Figure 7As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. Various programs and data required for the operation of the device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0187] Various components in device 700 are connected to I / O interface 705, including an input unit 706, such as a keyboard, mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, optical disk, etc.; and a communication unit 709, such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0188] The computing unit 701 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as the autonomous driving decision-making method or the model training method. For example, in some embodiments, the autonomous driving decision-making method or the model training method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the autonomous driving decision-making method described above or one or more steps of the model training method described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to execute the autonomous driving decision-making method or the model training method in any other appropriate manner (e.g., by means of firmware).

[0189] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0190] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0191] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0192] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0193] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0194] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0195] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0196] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. An autonomous driving decision-making method, comprising: When detecting that the vehicle interacts with the first obstacle, obtaining a first scene feature and a first state when the vehicle interacts with the first obstacle; Inputting the first scene feature into a target model to obtain a first interaction key parameter output by the target model and corresponding to the first scene feature; the first interaction key parameter is used to evaluate the driving behavior of the vehicle when interacting with the first obstacle; Based on the first state, constructing a first game tree for simulating the interaction between the vehicle and the first obstacle; When the interaction simulation between the vehicle and the first obstacle is completed, determining a target behavior decision for the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter; The target model is trained in the following way: Acquire a simulation set, where the simulation set includes M scene samples of the vehicle interacting with the obstacle, where M is an integer greater than 1; Determining M second interaction key parameters corresponding to the M scene samples; Evaluating the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter, the target scenario sample being a scenario sample in which traffic efficiency and / or safety do not meet standards under the vehicle's third behavior decision, and the game tree is constructed based on the states of the vehicle and the obstacle during the interaction; Determining a third interaction key parameter corresponding to the target scene sample, wherein the third interaction key parameter is an interaction key parameter indicating that an interaction result between the vehicle and the obstacle under the target scene sample meets standards in terms of traffic efficiency and safety; Based on the scene features corresponding to the target scene sample and the third interaction key parameter, the model parameters of the target model are updated, and the target model is used to characterize the mapping relationship between the scene features and the interaction key parameters of the interaction between the vehicle and the obstacle.

2. The method according to claim 1, wherein The first interaction key parameter includes a time step. Constructing a first game tree for simulating the interaction between the vehicle and the first obstacle based on the first state includes: Determining, based on the first state and a preset behavior combination, a second state at each moment of the time step starting from the interaction, the behavior combination including: driving behaviors of the vehicle and the first obstacle, respectively, within the time step; and the second state including: a state after simulating the vehicle and the first obstacle, respectively, driving according to the driving behaviors under the behavior combination; The first game tree is constructed based on the first state and the second state, where the first state is the state of the root node of the first game tree, and the second state is the state of the hierarchical nodes of the first game tree.

3. The method according to claim 1, wherein The first interaction key parameter includes a target standard parameter for scoring leaf nodes in the first game tree. Determining a target behavior decision of the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter includes: Scoring each leaf node in the first game tree based on the target standard parameter to obtain a scoring value of the leaf node; Select the leaf node with the highest score from the first game tree to obtain the target leaf node; The target behavior decision is determined based on the target driving behavior of the vehicle, where the target driving behavior is the driving behavior from the root node to the hierarchical node of the first layer under the path corresponding to the target leaf node.

4. The method according to claim 1, further comprising: In a case where an interaction between the vehicle and a first obstacle is detected, if an interaction between the vehicle and a second obstacle is detected, determining a first behavior decision of the vehicle when interacting with the second obstacle; Determining a target behavior decision of the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter includes: Determining, based on the first game tree and the first interaction key parameter, a second behavior decision of the vehicle when interacting with the first obstacle; The target behavior decision is determined based on the first behavior decision and the second behavior decision.

5. The method according to claim 1, wherein The determining of the M second interaction key parameters corresponding to the M scene samples includes: Obtaining M scene features corresponding to the M scene samples; The M scene features are respectively input into the target model to obtain the M second interaction key parameters output by the target model.

6. The method according to claim 1, wherein The determining of the third interaction key parameter corresponding to the target scene sample includes: Based on N preset interaction key parameters, score the interaction results between the vehicle and the obstacle in the target scene sample respectively to obtain N score values ​​corresponding to the N interaction key parameters; The interaction key parameter corresponding to the highest score value among the N score values ​​is determined as the third interaction key parameter.

7. An autonomous driving decision-making device comprising: A first acquisition module is configured to, when detecting that an interaction exists between a vehicle and a first obstacle, acquire a first scene feature and a first state of the interaction between the vehicle and the first obstacle; a first determination module, configured to determine a first interaction key parameter corresponding to the first scene feature, the first interaction key parameter being used to evaluate the driving behavior of the vehicle when interacting with the first obstacle; the first determination module being specifically configured to input the first scene feature into a target model to obtain the first interaction key parameter output by the target model; A construction module, configured to construct a first game tree for simulating the interaction between the vehicle and the first obstacle based on the first state; a second determining module, configured to determine, when the interaction simulation between the vehicle and the first obstacle is completed, a target behavior decision for the vehicle when interacting with the first obstacle based on the first game tree and the first interaction key parameter; A second acquisition module is used to acquire a simulation set, where the simulation set includes M scene samples of the interaction between the vehicle and the obstacle, where M is an integer greater than 1; a fourth determining module, configured to determine M second interaction key parameters corresponding to the M scene samples; an evaluation module, configured to evaluate the M third behavior decisions corresponding to the M scenario samples to obtain a target scenario sample, wherein the third behavior decision is determined based on a game tree simulating the interaction between the vehicle and the obstacle in the scenario sample and the second interaction key parameter, wherein the target scenario sample is a scenario sample in which the traffic efficiency and / or safety do not meet the requirements under the vehicle's third behavior decision, and the game tree is constructed based on the states of the vehicle and the obstacle during the interaction; a fifth determining module, configured to determine a third interaction key parameter corresponding to the target scene sample, wherein the third interaction key parameter is an interaction key parameter indicating that the interaction result between the vehicle and the obstacle in the target scene sample meets the requirements in terms of traffic efficiency and safety; An updating module is used to update the model parameters of the target model based on the scene features corresponding to the target scene sample and the third interaction key parameters, wherein the target model is used to characterize the mapping relationship between the scene features and the interaction key parameters of the interaction between the vehicle and the obstacle.

8. The device according to claim 7, wherein The first interaction key parameter includes a time step, and the building module is specifically used to: Determining, based on the first state and a preset behavior combination, a second state at each moment of the time step starting from the interaction, the behavior combination including: driving behaviors of the vehicle and the first obstacle, respectively, within the time step; and the second state including: a state after simulating the vehicle and the first obstacle, respectively, driving according to the driving behaviors under the behavior combination; The first game tree is constructed based on the first state and the second state, where the first state is the state of the root node of the first game tree, and the second state is the state of the hierarchical nodes of the first game tree.

9. The device according to claim 7, wherein The first interaction key parameter includes a target standard parameter for scoring the leaf nodes in the first game tree, and the second determination module is specifically configured to: Scoring each leaf node in the first game tree based on the target standard parameter to obtain a scoring value of the leaf node; Select the leaf node with the highest score from the first game tree to obtain the target leaf node; The target behavior decision is determined based on the target driving behavior of the vehicle, where the target driving behavior is the driving behavior from the root node to the hierarchical node of the first layer under the path corresponding to the target leaf node.

10. The apparatus according to claim 7, further comprising: a third determining module, configured to, when detecting that the vehicle has interacted with the first obstacle, determine a first behavior decision of the vehicle when interacting with the second obstacle if detecting that the vehicle has interacted with the second obstacle; The second determining module is specifically configured to: Determining, based on the first game tree and the first interaction key parameter, a second behavior decision of the vehicle when interacting with the first obstacle; The target behavior decision is determined based on the first behavior decision and the second behavior decision.

11. The device according to claim 7, wherein The fourth determining module is specifically configured to: Obtaining M scene features corresponding to the M scene samples; The M scene features are respectively input into the target model to obtain the M second interaction key parameters output by the target model.

12. The device according to claim 7, wherein The fifth determining module is specifically configured to: Based on N preset interaction key parameters, score the interaction results between the vehicle and the obstacle in the target scene sample respectively to obtain N score values ​​corresponding to the N interaction key parameters; The interaction key parameter corresponding to the highest score value among the N score values ​​is determined as the third interaction key parameter.

13. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 6.

15. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

16. An autonomous driving vehicle comprising the electronic device according to claim 13.

Images