A method, system, and storage medium for selecting a mode of driving a vehicle

Abstract

The application relates to the technical field of automatic driving, and discloses a method, a system and a storage medium for selecting a car driving mode, the method comprising the following steps: a vehicle-mounted system receives user expected income and expenditure, a starting node and a terminal node, and transmits the starting node and the terminal node to a cloud server; the cloud server obtains an optimal path comprising M nodes and M road sections according to the starting node and the terminal node, and transmits the optimal path to the vehicle-mounted system; and a model contained in the vehicle-mounted system determines a driving mode of the M nodes and an interaction mode of the M road sections between a car and a roadside unit. The application gives the decision-making right of the car driving mode to the vehicle-mounted system, reduces the blindness of human selection, and improves user experience; the expected income and expenditure are introduced, the difference between actual expenditure and the expected income and expenditure is reduced during training, and user experience is further improved; a hierarchical reinforcement learning method is introduced, a problem is refined, the convergence speed of an algorithm is improved, and user experience is further improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method, system, and storage medium for selecting a car driving mode based on satisfaction. Background Technology

[0002] With the continuous improvement of computer processing power and the development of artificial intelligence technology, autonomous driving technology has made great strides. Various sensors and computer systems are installed in vehicles, enabling them to autonomously perceive, understand, and make decisions about road conditions. Currently, some automakers have launched autonomous electric vehicle products, such as Tesla's Autopilot and Baidu's Apollo autonomous driving project. As the level of automated driving continues to improve, in the future, whether for private car travel or ride-hailing, drivers will need to add the option of autonomous driving in addition to manual driving. While ensuring safety, adding autonomous driving options will inevitably lead to an increasing number of vehicles choosing autonomous driving on the road. The convergence of numerous autonomous vehicles on the same road segment at the same time creates a huge demand for resources. If the computing resources or computing power of intelligent roadside equipment are insufficient, it will be unable to process the tasks assigned by vehicles in a timely manner, resulting in a poor user experience. However, current technologies only consider the task offloading of intelligent roadside equipment for autonomous driving, or only consider vehicles sharing their surplus computing power with RSUs lacking computing power, without combining the two. This leads to a regional imbalance in computing resource demand and low user satisfaction.

[0003] Existing technology discloses a shared autonomous vehicle system, which includes a vehicle-side component, a cloud platform, sensors, a data acquisition system, and an autonomous driving system. The vehicle-side component and the cloud platform are connected via data upload and model download. The vehicle-side component includes a driver identification module connected to a driving mode selection module, which in turn connects to a manual driving mode and an autonomous driving mode. The manual driving mode and the autonomous driving mode are respectively connected to the data acquisition system and the autonomous driving system. The cloud platform includes scene reconstruction, machine learning, and an autonomous driving model. Scene reconstruction is connected to machine learning, and machine learning is connected to the autonomous driving model. This existing shared autonomous vehicle system requires driver identification before each use, and there is no prior reference for selecting a driving mode. Often, some road sections are unsuitable for autonomous driving, but the driver mistakenly selects autonomous driving due to a lack of feedback from the driving system regarding the target road conditions, resulting in a poor driver experience. Summary of the Invention The purpose of this invention is to provide a method, system, and storage medium for selecting car driving modes that can improve user satisfaction.

[0004] To achieve the above objectives, the present invention provides a method for selecting a car driving mode, comprising the following steps: Step S1: The vehicle system receives the expected income, expected expenditure, starting node, and ending node, and transmits the starting node and the ending node to the cloud server; Step S2: The cloud server includes a path planning model. The path planning model obtains the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Step S3: The vehicle system inputs the optimal path, expected input, and expected expenditure of the M nodes and M road segments into the node driving mode selection model to obtain the driving modes of the M nodes; Step S4: Input the driving modes of the M nodes into the vehicle-roadside unit computing power sharing model to obtain the interaction mode between the vehicle and the roadside unit for the next road segment corresponding to the M nodes, and complete the driving mode selection.

[0005] Furthermore, the interaction method between the vehicle and the roadside unit in step S4 includes: If the vehicle system selects manual driving mode at the node, the interaction decision factor β between the car and the roadside unit is β∈{0,1}. The interaction decision factor β is determined by the driver. If β=0, the interaction between the car and the roadside unit is: the car refuses to receive the transmission task from the roadside unit. If β=1, the interaction between the car and the roadside unit is: the car receives the transmission task from the roadside unit and receives income. If the vehicle system selects the autonomous driving mode at the node, the interaction decision factor β between the vehicle and the roadside unit is [0,1]. The interaction method between the vehicle and the roadside unit is as follows: the vehicle pays an auxiliary computing fee to offload (β*100)% of the task to the roadside unit for auxiliary computing.

[0006] Furthermore, the path planning model described in step S2 is obtained by training the Astar algorithm, which obtains the optimal path between the starting point and the ending point based on the starting point and the ending point.

[0007] Furthermore, the process by which the Astar algorithm obtains the optimal path between the starting point and the ending point based on the starting point and the ending point includes: Step S2.1: The Astar algorithm is initialized and the starting node is determined, and the starting node is stored in the open list; Step S2.2: Determine the node with the smallest distance from the starting node and the smallest congestion coefficient, and store the smallest node in the open list; Step S2.3: Expand the neighboring nodes of the node with the smallest distance from the starting node, calculate the congestion coefficient and distance of the neighboring nodes, select the node with the smallest congestion coefficient and distance, and store the smallest node in the open list; Step S2.4: Repeat steps S2.1 to S2.3 until the smallest node is found as the termination node. In the open list, an optimal path with the minimum cost from the starting node to the termination node can be obtained.

[0008] Furthermore, the node driving mode selection model mentioned in step S3 and the vehicle-roadside unit computing power sharing model mentioned in step S4 are determined by a hierarchical reinforcement learning algorithm, which includes upper and lower layers. The upper layer of the hierarchical reinforcement learning trains the node driving mode selection model, and the node driving mode selection model determines the driving mode of the node. The lower layer of the hierarchical reinforcement learning trains the vehicle-roadside unit computing power sharing model, and the vehicle-roadside unit computing power sharing model determines the interaction mode between the vehicle and the roadside unit in the next road segment corresponding to the node. The vehicle system has a reward function for each road segment. The interaction method between the vehicle and the roadside unit determines the reward function, and the reward function optimizes the node driving mode selection model.

[0009] Furthermore, the reward function includes: If the vehicle system selected manual driving mode in the previous node, then the vehicle system will select manual driving mode in the corresponding next node. A reward function set for a road segment for: If the vehicle system selected autonomous driving mode in the previous node, then the vehicle system will proceed in the corresponding next node. A reward function set for a road segment for: The optimal path comprises M road segments, where m represents any one of the M road segments, A represents the automatic driving mode, and H represents the manual driving mode. This represents the satisfaction rating for autonomous driving. Indicates the current road segment length, and E represents the driver's expected income and expenditure. This represents the difference coefficient between actual income and expenditure and the driver's expected income and expenditure, where L represents the total road segment length. Indicates the task is in a time slot The driver's total income and expenditure are completed within the time limit. This indicates the total number of time slots in the current road segment. This indicates the nth time slot of the current road segment.

[0010] Furthermore, the node driving mode selection model is determined in the following way: Construct a node driving mode selection model; Define the state, where the node driving mode selection model is a high-level state, and the main state of the high-level state is... include: Where E represents the user's expected income and expenditure. This indicates the current intersection number for vehicle x. Current traffic expenditure, Indicates the vehicle's current battery level. ; Indicates the current time slot The amount of work, where h indicates that the state category is a high-level state; Define actions, whereby the higher-level states correspond to higher-level actions. The high-level actions The high-level action is used to determine the driving mode of the next road segment corresponding to the current node. Determined by the following formula: [0, 1] If it is 1, then it is in the road section For autonomous driving, If it is 0, then it is in the road section For manual driving, the high-level actions Determine the lower-level actions; Define a reward, whereby the reward for a high-level action is the sum of the rewards for low-level actions. The reward for the high-level action is used to optimize the driving mode selection model for the aforementioned node. The reward for the high-level action is determined by the following formula: in This represents the total number of time slots for the current road segment. For low-level actions in time slots The reward; Repeat the above process to obtain a trained node driving mode selection model.

[0011] Furthermore, the computing power sharing model between the vehicle and the roadside unit is determined in the following way: Construct a computing power sharing model between vehicles and roadside units; Define the state; the vehicle-roadside unit computing power sharing model is a low-level state, and the main state of the low-level state is... include: in, Indicates the vehicle's current location. Indicates the vehicle's current battery level. Indicates the current time slot of the vehicle The workload, The current flow status of the roadside unit in the current time slot, where ι indicates that the status category is a low-level state; Define actions, where the lower-level states correspond to lower-level actions. The low-level actions The low-level actions are used to determine the interaction between the vehicle and the roadside unit. Depends on the actions of higher-ups The low-level actions Determined by the following formula: If high-level actions =1, Indicates in The time-slot vehicle will currently be performing the task %Unload to the roadside unit for auxiliary calculations, The value is a floating-point number between 0 and 1; if the higher-level action... =0, A value of 1 indicates that the vehicle has decided to share its computing power and accept computing tasks transmitted from the roadside unit. A value of 0 indicates that the vehicle does not share computing power. A binary value that is either 0 or 1; Define rewards, each road segment is divided into Each time slot For one of the time slots, the reward for the lower-level action in each time slot is: If high-level actions =1, the time slot reward is Where a and b are non-zero coefficients. For the task in the time slot The driver's total income and expenditure are completed within the time limit. Total time required to complete the task; If high-level actions =0, the slot reward is ,in The compensation received for car-assisted roadside units A coefficient greater than 1 The energy consumed by the car to receive the task. The onboard system calculates the energy consumed for the tasks received by the vehicle. The local electricity price for the driver; Repeat the above process to obtain a trained model for sharing computing power between cars and roadside units.

[0012] Furthermore, a system for selecting a car driving mode includes: Receiving module: used by the vehicle system to receive expected income, expected expenditure, starting node and ending node, and transmit the starting node and ending node to the cloud server; Optimal Path Acquisition Module: Used to obtain the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Node driving mode selection module: used to input the optimal paths, expected inputs and expected expenditures of M nodes and M road segments into the node driving mode selection model to obtain the driving modes of the M nodes; The module for determining the interaction mode between the car and the roadside unit is used to input the driving modes of the M nodes into the car-roadside unit computing power sharing model, obtain the interaction mode between the car and the roadside unit for the next road segment corresponding to the M nodes, and complete the current driving mode selection.

[0013] Furthermore, the present invention also provides a computer-readable storage medium storing thereon a computer program including a method for selecting a vehicle driving mode, wherein when the computer program for selecting a vehicle driving mode is executed by a processor, it implements the steps of the method for selecting a vehicle driving mode.

[0014] Compared with the prior art, the advantages of this invention are as follows: This invention relates to the field of autonomous driving technology, and discloses a method, system, and storage medium for selecting a vehicle driving mode. By delegating the decision-making power of the vehicle driving mode to the onboard system, allowing the onboard system to select the appropriate driving mode based on actual road conditions, the problem of blind human selection is solved, improving the user experience. By introducing the concept of expected travel expenditure, the difference between actual traffic expenditure and expected expenditure is reduced during training, further improving the user experience. By introducing a hierarchical reinforcement learning method, the problem is decomposed into multiple levels, with each level having a relatively small state space, greatly reducing the state space of the algorithm in practical applications, thereby improving the convergence speed of the algorithm, learning and optimizing driving mode selection more efficiently, and further improving the user experience. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for selecting a car driving mode according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a system for selecting a car driving mode according to an embodiment of the present invention; Figure 3 This is a system model diagram of selecting a car driving mode according to an embodiment of the present invention; Figure 4 This is a flowchart illustrating the selection of a car driving mode according to an embodiment of the present invention. Detailed Implementation

[0016] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0017] Example 1: like Figure 1 As shown, a preferred embodiment of the present invention provides a method for selecting a car driving mode, comprising the following steps: Step S1: The vehicle system receives the expected income, expected expenditure, start node, and end node, and transmits the start node and end node to the cloud server; Step S2: The cloud server contains a path planning model. The path planning model obtains the optimal path between the start node and the end node based on the start node and the end node. The optimal path contains M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Step S3: The vehicle system inputs the optimal paths, expected inputs, and expected expenditures of M nodes and M road segments into the node driving mode selection model to obtain the driving modes of the M nodes; Step S4: Input the driving modes of the M nodes into the vehicle-roadside unit computing power sharing model to obtain the interaction mode between the vehicle and the roadside unit for the next road segment corresponding to the M nodes, and complete the driving mode selection.

[0018] This invention addresses the issue of blind human choice by delegating the decision-making power for driving modes to the in-vehicle system, allowing the system to select the appropriate mode based on actual road conditions, thus improving user experience. Furthermore, by introducing the concept of expected travel expenditure, this invention reduces the difference between actual travel expenditure and expected expenditure during training, further enhancing user experience. Additionally, by employing a hierarchical reinforcement learning method, the problem is decomposed into multiple levels, each with a relatively small state space. This significantly reduces the state space of the algorithm in practical applications, thereby improving convergence speed, enabling more efficient learning and optimization of driving mode selection, and further enhancing user experience.

[0019] Example 2: like Figure 1 As shown, a preferred embodiment of the present invention provides a method for selecting a car driving mode, comprising the following steps: Step S1: The vehicle system receives the expected income, expected expenditure, start node, and end node, and transmits the start node and end node to the cloud server; In this embodiment, the in-vehicle system receives expected income, expected expenditure, start node, and end node in the following ways: before driving the car, the user inputs the information in the in-vehicle system or through a third party, and then the third party transmits it to the in-vehicle system. The third party can be a mobile phone, computer, or other device.

[0020] Step S2: The cloud server includes a path planning model. The path planning model obtains the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. In this embodiment, the path planning model is obtained by training with the A* algorithm, which is a direct search method for finding the shortest path in a static road network. In a map containing various obstacles, the goal is to find the shortest path to the target location for the game character's movement. The A* algorithm combines the node information from Dijkstra's algorithm (favoring nodes closer to the starting point) with the best-first search information from a greedy algorithm (favoring nodes closer to the target). Like Dijkstra's algorithm, it guarantees finding the shortest path, while also using heuristics to guide the algorithm, similar to the greedy best-first search algorithm. In this embodiment, the process of obtaining the optimal path between the starting and ending points using the A* algorithm includes: The Astar algorithm initializes and determines the starting node, and stores the starting node in the open list. Determine the node with the minimum distance from the starting node and the minimum congestion coefficient, and store the minimum node in the open list; Expand the neighboring nodes of the node with the shortest distance from the starting node, calculate the congestion coefficient and distance of the neighboring nodes, select the node with the smallest congestion coefficient and distance, and store the node with the smallest distance in the open list. Repeat steps S2.1 to S2.3 until the smallest node is found as the terminal node. From the open list, an optimal path with the minimum cost from the starting node to the terminal node can be obtained.

[0021] If we assume For road nodes, Starting from, The starting point is the destination, and there are a total of M road segments between the starting point and the destination. For the m-th road segment, the shortest path model obtained by the A* algorithm can be represented as: ,m∈M On the road section Several roadside units are distributed on it. , Indicates in The k-th roadside unit on the road segment can be represented by the following model: K Vehicle X in The total driving time on the road segment is It is further discretized into n parts, that is, the time slot length is For the length of the road segment, For the average speed, then in The time slot model for the road segment is as follows: Step S3: The vehicle system inputs the optimal paths, expected inputs, and expected expenditures of M nodes and M road segments into the node driving mode selection model to obtain the driving modes of the M nodes; In this embodiment, the node driving mode selection model is determined in the following way: Construct a node driving mode selection model; Define the state, where the node driving mode selection model is a high-level state, and the main state of the high-level state is... include: Where E represents the user's expected income and expenditure. This indicates the current intersection number for vehicle x. Current traffic expenditure, Indicates the vehicle's current battery level. ; Indicates the current time slot The amount of work, where h indicates that the state category is a high-level state; Define actions, whereby the higher-level states correspond to higher-level actions. The high-level actions The high-level action is used to determine the driving mode of the next road segment corresponding to the current node. Determined by the following formula: [0, 1] If it is 1, then it is in the road section For autonomous driving, If it is 0, then it is in the road section For manual driving, the high-level actions Determine the lower-level actions; Define a reward, whereby the reward for a high-level action is the sum of the rewards for low-level actions. The reward for the high-level action is used to optimize the driving mode selection model for the aforementioned node. The reward for the high-level action is determined by the following formula: in This represents the total number of time slots for the current road segment. For low-level actions in time slots The reward; Repeat the above process to obtain a trained node driving mode selection model.

[0022] It should be noted that if the car selects α(m) = 1 at the node, i.e., in autonomous driving mode, then... Within the road segment, the interactive decision factors between vehicle x and the RSU are expressed as follows: This refers to electric vehicles using the current time slot. workload Unload to RSU for auxiliary calculations. The value is a floating-point number between 0 and 1; if the car selects α(m) = 0 at the node, i.e., manual driving mode, then... Within the road segment, the interactive decision factors between vehicle x and the RSU are expressed as follows: , Electric vehicles Does the time slot accept the workload passed from RSU? A binary value that is either 0 or 1.

[0023] Step S4: Input the driving modes of M nodes into the vehicle-roadside unit computing power sharing model to obtain the interaction mode between the vehicle and the roadside unit for the next road segment corresponding to the M nodes, thus completing this driving mode selection.

[0024] In this embodiment, the computing power sharing model between vehicles and roadside units is determined in the following way: Construct a computing power sharing model between vehicles and roadside units; Define the state; the vehicle-roadside unit computing power sharing model is a low-level state, and the main state of the low-level state is... include: in, Indicates the vehicle's current location. Indicates the vehicle's current battery level. Indicates the current time slot of the vehicle The workload, The current flow status of the roadside unit in the current time slot, where ι indicates that the status category is a low-level state; Define actions, where the lower-level states correspond to lower-level actions. The low-level actions The low-level actions are used to determine the interaction between the vehicle and the roadside unit. Depends on the actions of higher-ups The low-level actions Determined by the following formula: If high-level actions =1, Indicates in The time-slot vehicle will currently be performing the task %Unload to the roadside unit for auxiliary calculations, The value is a floating-point number between 0 and 1; if the higher-level action... =0, A value of 1 indicates that the vehicle has decided to share its computing power and accept computing tasks transmitted from the roadside unit. A value of 0 indicates that the vehicle does not share computing power. A binary value that is either 0 or 1; Define rewards, each road segment is divided into Each time slot For one of the time slots, the reward for the lower-level action in each time slot is: If high-level actions =1, the time slot reward is Where a and b are non-zero coefficients. For the task in the time slot The driver's total income and expenditure are completed within the time limit. Total time required to complete the task; If high-level actions =0, the slot reward is ,in The compensation received for car-assisted roadside units A coefficient greater than 1 The energy consumed by the car to receive the task. The onboard system calculates the energy consumed for the tasks received by the vehicle. The local electricity price for the driver; Repeat the above process to obtain a trained model for sharing computing power between cars and roadside units.

[0025] If the vehicle selects autonomous driving at a node, it can choose to offload tasks to the RSU for auxiliary computation in autonomous driving mode. (During time slot intervals) To ensure the computational speed for autonomous driving tasks, the vehicle can choose to... The workload is offloaded to RSU for auxiliary computation. The workload is computed locally, ultimately minimizing the computation latency of autonomous driving tasks. Offloading tasks to the RSU requires transmission latency. and transmission costs Because the wireless channel loss between vehicles and roadside units is more dominant than that of other channels, therefore, in time slots... From vehicle x to roadside unit Channel gain h( This can be represented by the free space path loss model as follows: where This represents the received power at a reference distance of 1 meter and a transmit power of 1 watt. Indicates in time slot roadside unit The Euclidean distance between vehicle x and vehicle x.

[0026] To improve the signal-to-noise ratio, assume the vehicle operates at its maximum power. Transmission is performed. Since vehicle x can receive at most one roadside unit in each time slot... The computational offloading is achieved, thus communication interference between channels can be ignored. Therefore, in the time slot Vehicles and roadside units The instantaneous transmission rate (bps / Hz) between them can be given as: , E, in This represents the noise power at the vehicle location. Then, in the time slot... Transfer task from vehicle x To the roadside unit The resulting time delay Displayed as: B refers to channel bandwidth, task The energy required to transfer energy from the vehicle to the RSU is: Considering that the output of various computationally intensive tasks (such as speech processing and image classification) is much smaller than the input, the output is taken from the roadside unit. The time and effort required to transmit the data to vehicle x is negligible. Unload the vehicle x task to the roadside unit Traffic pricing, then computing tasks Unloading from vehicle x to roadside unit The calculated cost is as follows: The number of CPU clocks required to process each bit / byte of task. It is a roadside unit Local computing power, task The computation delay for RSU is: Remaining tasks Calculated locally on the vehicle. If the local computing power of vehicle x is the computation latency, then the computation latency is: The coefficients related to the CPU architecture of vehicle x in the time slot The power of vehicle x is: In the time slot Local computing tasks on vehicle x energy consumption Expressed as the product of power and computation time: If the local electricity price is used, then the task is... In the time slot The total cost to the user upon completion is the sum of the data usage and electricity costs required for autonomous driving. In the time slot Internal mission The calculation is mainly divided into two parts: one part is calculated locally, and the other part is performed as an auxiliary calculation in a roadside unit. The total time required to complete the task is... The larger of the two calculation times: a and b are non-zero coefficients, and in autonomous driving mode, they are in time slots. Inside, the second layer of reinforcement learning actions The reward function r( It is inversely proportional to the computational latency of the task and the total cost of autonomous driving: If the vehicle chooses manual driving at the node, it can share its computing power with the RSU, meaning it can choose whether or not to receive computing tasks from the RSU. The computational model for task transmission is roughly the same as that in autonomous driving mode, and the transmission delay for the vehicle to receive RSU tasks can be obtained as follows: for The power of the vehicle receiving the task, then the vehicle x receives The energy required is: Task The computational latency at the vehicle's local location is: The vehicle's own computing tasks The local computation latency is: Then in the time slot Vehicle x performs local computation tasks The energy required is: ) According to RSU The pricing, Traffic pricing, then vehicle-assisted RSU calculation task The reward received is: For local electricity prices, task In the time slot The total revenue and expenditure for users who complete the task are as follows: 'd' is a coefficient greater than 1 because the expenses caused by power consumption not only involve monetary costs but also require the owner to spend time charging. Excessive power consumption can cause range anxiety, hence a coefficient greater than 1 is added to the power consumption expenses. Therefore, in manual driving mode, during time slots... Inside, the second layer of reinforcement learning actions The reward function r( Equivalent to total revenue and expenditure: Example 3: The difference between this embodiment and embodiment two is that, based on embodiment two, the node driving mode selection model and the roadside unit computing power sharing model are determined by a hierarchical reinforcement learning algorithm. The hierarchical reinforcement learning algorithm consists of two layers, upper and lower. By decomposing the task into multiple sub-tasks, the sample utilization rate is higher.

[0027] The upper layer of the hierarchical reinforcement learning training node driving mode selection model determines the driving mode of the node. The lower layer of the hierarchical reinforcement learning training is a car-roadside unit computing power sharing model. The car-roadside unit computing power sharing model determines the interaction mode between the car and the roadside unit in the next road segment corresponding to the node. The vehicle system has a reward function for each road segment. The interaction between the car and the roadside unit determines the reward function, which optimizes the node driving mode selection model.

[0028] Example 4: The difference between this embodiment and Embodiment 3 is that the reward function is determined in the following way: If the vehicle system selected manual driving mode in the previous node, then the vehicle system will select manual driving mode in the corresponding next node. A reward function set for a road segment for: If the vehicle system selected autonomous driving mode in the previous node, then the vehicle system will proceed in the corresponding next node. A reward function set for a road segment for: The optimal path comprises M road segments, where m represents any one of the M road segments, A represents the automatic driving mode, and H represents the manual driving mode. This represents the satisfaction rating for autonomous driving. Indicates the current road segment length, and E represents the driver's expected income and expenditure. This represents the difference coefficient between actual income and expenditure and the driver's expected income and expenditure, where L represents the total road segment length. Indicates the task is in a time slot The driver's total income and expenditure are completed within the time limit. This indicates the total number of time slots in the current road segment. This indicates the nth time slot of the current road segment.

[0029] Example 5: like Figure 2 As shown, a preferred embodiment of the present invention provides a system for selecting a car driving mode, comprising: Receiving module: used by the vehicle system to receive expected income, expected expenditure, starting node and ending node, and transmit the starting node and ending node to the cloud server; Optimal Path Acquisition Module: Used to obtain the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Node driving mode selection module: used to input the optimal paths, expected inputs and expected expenditures of M nodes and M road segments into the node driving mode selection model to obtain the driving modes of the M nodes; The module for determining the interaction mode between the car and the roadside unit is used to input the driving modes of the M nodes into the car-roadside unit computing power sharing model, obtain the interaction mode between the car and the roadside unit for the next road segment corresponding to the M nodes, and complete the current driving mode selection.

[0030] This invention addresses the issue of blind human choice by delegating the decision-making power for driving modes to the in-vehicle system, allowing the system to select the appropriate mode based on actual road conditions, thus improving user experience. Furthermore, by introducing the concept of expected travel expenditure, this invention reduces the difference between actual travel expenditure and expected expenditure during training, further enhancing user experience. Additionally, by employing a hierarchical reinforcement learning method, the problem is decomposed into multiple levels, each with a relatively small state space. This significantly reduces the state space of the algorithm in practical applications, thereby improving convergence speed, enabling more efficient learning and optimization of driving mode selection, and further enhancing user experience.

[0031] Example 6: A preferred embodiment of the present invention provides a computer-readable storage medium having a computer program thereon storing a method for selecting a vehicle driving mode, wherein when the computer program for selecting a vehicle driving mode is executed by a processor, the steps of the method for implementing the vehicle driving mode are described.

[0032] In summary, the embodiments of the present invention provide a method, system, and storage medium for selecting a car driving mode. By delegating the decision-making power of the car driving mode to the on-board system, allowing the on-board system to select the appropriate driving mode based on actual road conditions, the invention solves the problem of blind human selection and improves user experience. Furthermore, by introducing the concept of expected travel expenditure, the invention reduces the difference between actual traffic expenditure and expected expenditure during training, further improving user experience. Additionally, by introducing a hierarchical reinforcement learning method, the invention decomposes the problem into multiple levels, each with a relatively small state space, significantly reducing the state space of the algorithm in practical applications. This improves the convergence speed of the algorithm, enabling more efficient learning and optimization of driving mode selection, and further enhancing user experience.

[0033] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for selecting a car driving mode, characterized in that, Includes the following steps: Step S1: The vehicle system receives the expected income, expected expenditure, starting node, and ending node, and transmits the starting node and the ending node to the cloud server; Step S2: The cloud server includes a path planning model. The path planning model obtains the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Step S3: The vehicle system inputs the optimal route, expected income, and expected expenditure, which includes M nodes and M road segments, into the node driving mode selection model to obtain the driving modes of the M nodes; Step S4: Input the driving modes of the M nodes into the vehicle-roadside unit computing power sharing model to obtain the interaction mode between the vehicle and the roadside unit for the next road segment corresponding to the M nodes, and complete the driving mode selection.

2. The method for selecting a car driving mode according to claim 1, characterized in that, The interaction methods between the vehicle and the roadside unit in step S4 include: If the vehicle system selects manual driving mode at the node, the interaction decision factor β between the car and the roadside unit is ∈{0,1}. The interaction decision factor β is determined by the driver. If β=0, the interaction between the car and the roadside unit is: the car refuses to receive the transmission task from the roadside unit. If β=1, the interaction between the car and the roadside unit is: the car receives the transmission task from the roadside unit and receives income. If the vehicle system selects the autonomous driving mode at the node, the interaction decision factor β between the vehicle and the roadside unit is [0,1]. The interaction method between the vehicle and the roadside unit is as follows: the vehicle pays an auxiliary computing fee to offload (β*100)% of the task to the roadside unit for auxiliary computing.

3. The method for selecting a car driving mode according to claim 1, characterized in that, The path planning model described in step S2 is obtained by training the Astar algorithm. The Astar algorithm obtains the optimal path between the starting node and the ending node based on the starting node and the ending node.

4. The method for selecting a car driving mode according to claim 3, characterized in that, The Astar algorithm obtains the optimal path between the starting node and the ending node based on the starting node and the ending node, including the following process: Step S2.1: The Astar algorithm is initialized and the starting node is determined, and the starting node is stored in the open list; Step S2.2: Determine the node with the smallest distance from the starting node and the smallest congestion coefficient, and store the smallest node in the open list; Step S2.3: Expand the neighboring nodes of the node with the smallest distance from the starting node, calculate the congestion coefficient and distance of the neighboring nodes, select the node with the smallest congestion coefficient and distance, and store the smallest node in the open list; Step S2.4: Repeat steps S2.1 to S2.3 until the smallest node is found as the termination node. In the open list, an optimal path with the minimum cost from the starting node to the termination node can be obtained.

5. The method for selecting a car driving mode according to claim 1, characterized in that, The node driving mode selection model in step S3 and the vehicle-roadside unit computing power sharing model in step S4 are determined by a hierarchical reinforcement learning algorithm, which consists of upper and lower layers. The upper layer of the hierarchical reinforcement learning trains the node driving mode selection model, and the node driving mode selection model determines the driving mode of the node. The lower layer of the hierarchical reinforcement learning trains the vehicle-roadside unit computing power sharing model, and the vehicle-roadside unit computing power sharing model determines the interaction mode between the vehicle and the roadside unit in the next road segment corresponding to the node. The vehicle system has a reward function for each road segment. The interaction method between the vehicle and the roadside unit determines the reward function, and the reward function optimizes the node driving mode selection model.

6. The method for selecting a car driving mode according to claim 5, characterized in that, The reward function includes: If the vehicle system selected manual driving mode in the previous node, then the vehicle system will... The reward function set for the next road segment for: If the vehicle system selected autonomous driving mode in the previous node, then the vehicle system will proceed in the corresponding... The reward function set for the next road segment for: The optimal path comprises M road segments, where m represents any one of the M road segments, A represents the automatic driving mode, and H represents the manual driving mode. This represents the satisfaction rating for autonomous driving. Indicates the current road segment length, and E represents the driver's expected income and expenditure. This represents the difference coefficient between actual income and expenditure and the driver's expected income and expenditure, where L represents the total road segment length. Indicates the task is in a time slot The driver's total income and expenditure are completed within the time limit. This indicates the total number of time slots in the current road segment. This indicates the nth time slot of the current road segment.

7. The method for selecting a car driving mode according to claim 6, characterized in that, Place The driving mode selection model for the aforementioned nodes is determined in the following way: Construct a node driving mode selection model; Define the state, where the node driving mode selection model is a high-level state, and the main state of the high-level state is... include: Where E represents the user's expected income and expenditure. This indicates the current intersection number for vehicle x. Current traffic expenditure, Indicates the vehicle's current battery level. ; Indicates the current time slot The amount of work, where h indicates that the state category is a high-level state; Define actions, whereby the higher-level states correspond to higher-level actions. The high-level actions The high-level action is used to determine the driving mode of the next road segment corresponding to the current node. Determined by the following formula: [0, 1] If it is 1, then it is in the road section For autonomous driving, If it is 0, then it is in the road section For manual driving, the high-level actions Determine the lower-level actions; Define a reward: the reward for a high-level action is the sum of the rewards for low-level actions. The reward for the high-level action is used to optimize the node driving mode selection model. The reward for the high-level action is determined by the following formula: in This represents the total number of time slots for the current road segment. For low-level actions in time slots The reward; Repeat the above process to obtain a trained node driving mode selection model.

8. The method for selecting a car driving mode according to claim 5, characterized in that, The computing power sharing model between vehicles and roadside units is determined in the following way: Construct a computing power sharing model between vehicles and roadside units; Define the state; the vehicle-roadside unit computing power sharing model is a low-level state, and the main state of the low-level state is... include: in, Indicates the vehicle's current location. Indicates the vehicle's current battery level. Indicates the current time slot of the vehicle The workload, The current flow status of the roadside unit in the current time slot, where ι indicates that the status category is a low-level state; Define actions, where the lower-level states correspond to lower-level actions. The low-level actions The low-level actions are used to determine the interaction between the vehicle and the roadside unit. Depends on the actions of higher-ups The low-level actions Determined by the following formula: If high-level actions =1, Indicates in The time-slot vehicle will currently be performing the task %Unload to the roadside unit for auxiliary calculations, The value is a floating-point number between 0 and 1; if the higher-level action... =0, A value of 1 indicates that the vehicle has decided to share its computing power and accept computing tasks transmitted from the roadside unit. A value of 0 indicates that the vehicle does not share computing power. A binary value that is either 0 or 1; Define rewards, each road segment is divided into Each time slot For one of the time slots, the reward for the lower-level action in each time slot is: If high-level actions =1, the time slot reward is Where a and b are non-zero coefficients. For the task in the time slot The driver's total income and expenditure are completed within the time limit. Total time required to complete the task; If high-level actions =0, the slot reward is ,in The compensation received for car-assisted roadside units A coefficient greater than 1 The energy consumed by the car to receive the task. The onboard system calculates the energy consumed for the tasks received by the vehicle. The local electricity price for the driver; Repeat the above process to obtain a trained model for sharing computing power between cars and roadside units.

9. A system for selecting a car driving mode, characterized in that, include: Receiving module: used by the vehicle system to receive expected income, expected expenditure, starting node and ending node, and transmit the starting node and ending node to the cloud server; Optimal Path Acquisition Module: Used to obtain the optimal path between the starting node and the ending node based on the starting node and the ending node. The optimal path includes M nodes and M road segments. The cloud server transmits the optimal path to the vehicle system. Node driving mode selection module: used to input the optimal route, expected income and expected expenditure of M nodes and M road segments into the node driving mode selection model to obtain the driving mode of the M nodes; The module for determining the interaction mode between the car and the roadside unit is used to input the driving modes of the M nodes into the car-roadside unit computing power sharing model, obtain the interaction mode between the car and the roadside unit for the next road segment corresponding to the M nodes, and complete the current driving mode selection.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for selecting a car driving mode as described in any one of claims 1 to 8.

Images