Multi-target collaborative path planning and scheduling system and method for autonomous vehicle

Abstract

The embodiment of the invention discloses an automatic driving vehicle multi-target cooperative path planning and scheduling system and method, and the method comprises the steps: obtaining the environment state information of a current vehicle in a communication range and the traffic information of a current planning path, inputting the environment state information into the trained agent model to obtain execution action information, controlling the current vehicle to run according to the execution action information, and when the traffic information meets a preset path updating condition, obtaining a driving task of the current vehicle and each preset vehicle navigation target, adjusting the priority of each vehicle navigation target according to the driving task and the traffic information, adjusting the current planned path based on the adjusted priority of each vehicle navigation target and a pre-loaded traffic map to obtain an updated path, and performing navigation control on the current vehicle based on the updated path. According to the scheme, by fusing deep reinforcement learning and the graph neural network and dynamically adjusting the priority, the accuracy and efficiency of vehicle navigation control can be improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of navigation technology, and in particular to a multi-objective collaborative path planning and scheduling system and method for an autonomous driving vehicle. Background Art

[0002] The navigation control technology of a vehicle is a feedback control that allows the vehicle to autonomously travel along a planned path. It mainly includes technologies such as environmental perception, precise positioning, path planning, and wire control execution. These technologies use various sensors (such as video cameras, radar sensors, and laser rangefinders) to obtain information about the surrounding environment, and use algorithms to process it, so that the vehicle can understand and respond to complex traffic conditions. The navigation control technology of a vehicle is also widely used in many fields in business, such as logistics distribution, public transportation, and special operations. For example, self-driving trucks can achieve fully automatic navigation driving for long-distance transportation through precise map data, sensors, and computer algorithms. Self-driving buses can achieve precise navigation control to stop at various stops by sensing the surrounding environment in real time.

[0003] Among the existing navigation technologies, most are based on traditional methods and rely on sensors (such as lidar and cameras) to obtain information about the vehicle's surroundings, and use traditional algorithms to plan navigation paths and set fixed weights for each navigation target (driving time, energy, etc.). This type of navigation method does not have enough depth of perception of the environment and cannot fully and accurately model the environment, which affects the accuracy of navigation path planning. In addition, the navigation target weighting method cannot be adjusted in real time, which may delay the navigation task. Summary of the invention The embodiments of the present application provide a multi-target collaborative path planning and scheduling system and method for an autonomous driving vehicle, which solves the problems in existing navigation methods, such as the vehicle's insufficient perception depth of the environment, the inability to comprehensively and accurately model the environment, which affects the accuracy of navigation path planning, and the inability to adjust the navigation target weights in real time, which may cause delays in navigation tasks. The system and method can enhance the vehicle's perception and understanding of complex dynamic environments, realize adaptive adjustment of navigation target weights, and improve the accuracy and efficiency of vehicle navigation control.

[0004] In a first aspect, an embodiment of the present application provides a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle, comprising: Obtaining environmental status information within the communication range of the current vehicle and traffic information of the current planned path, inputting the environmental status information into the trained agent model to obtain execution action information of the current vehicle, and controlling the operation of the current vehicle according to the execution action information, wherein the environmental status information includes driving information sent by other vehicles and detected road surface information; When it is detected that the traffic information satisfies a preset path update condition, obtaining the driving task of the current vehicle and each preset vehicle navigation target, and adjusting the priority of each vehicle navigation target according to the driving task and the traffic information; The currently planned path is adjusted based on the adjusted priorities of the navigation targets of each vehicle and the pre-loaded traffic map to obtain an updated path, and navigation control of the current vehicle is performed based on the updated path.

[0005] Optionally, the training process of the agent model includes: A deep neural network is constructed, sample environmental state information is determined as input data of the deep neural network, and action probability distribution information is determined as output data of the deep neural network, the sample environmental state information includes driving information of other vehicles and road surface information, the driving information includes vehicle position, speed and acceleration, the road surface information includes road curvature, road slope and traffic signal parameters, and reinforcement learning training is performed on the deep neural network based on a preset reward mechanism.

[0006] Optionally, the performing reinforcement learning training on the deep neural network based on a preset reward mechanism includes: When the sample environment state information is input, randomly select an action in the action probability distribution information output by the deep neural network, and input the environment state information and the action into a reward function to obtain a reward result; The parameters of the deep neural network are updated and adjusted according to the preset deep Q network algorithm, and the training is repeated to obtain the intelligent agent model. The update formula of the deep Q network algorithm is: , is the learning rate, is the discount factor, is the state-action value function, which means that in the state Take action expected long-term rewards.

[0007] Optionally, the vehicle navigation target includes driving time and energy reserve, and adjusting the priority of each vehicle navigation target according to the driving task and the traffic information includes: The corresponding weights are calculated according to the preset weight calculation formulas of each of the vehicle navigation targets. The weight calculation formula of the driving time is: , the energy reserve weight calculation formula is: , represents the traffic congestion index, Indicates the urgency of the task. Represents the vehicle energy value, Indicates the energy-saving area marking value; The traffic congestion index and the task urgency are calculated according to the driving task and the traffic information, and the vehicle energy value of the current vehicle and the energy-saving area marking value in the traffic information are obtained, and the traffic congestion index, the task urgency, the vehicle energy value and the energy-saving area marking value are respectively input into the corresponding preset weight calculation formula to obtain the weight of each of the vehicle navigation targets, and the priority is adjusted according to the weight of each of the vehicle navigation targets.

[0008] Optionally, the traffic information includes a vehicle density and a vehicle speed at a congested location, the driving task includes a task type, and the calculating the traffic congestion index and the task urgency according to the driving task and the traffic information includes: Substituting the vehicle density and vehicle speed of the congested location in the traffic information into a preset congestion calculation formula to obtain a traffic congestion index value, the preset congestion calculation formula includes , represents the vehicle density at the congested location, Indicates the vehicle speed at the congested location; According to the task type in the driving task, a preset task urgency comparison table is queried to obtain a task urgency value.

[0009] Optionally, before adjusting the current planned path based on the adjusted priorities of the vehicle navigation targets and the preloaded traffic map to obtain an updated path, the method further includes: Constructing the traffic map according to the traffic information, the traffic map comprising a plurality of traffic nodes and edges of connection relationships between the traffic nodes, the traffic nodes comprising road network nodes, vehicle nodes and obstacle nodes, and the edges comprising weight information; A graph convolutional layer is used to continuously update and aggregate the features of each traffic node to obtain a new feature representation of each traffic node, which is used to capture the spatial relationship and semantic information between the elements.

[0010] Optionally, adjusting the current planned path based on the adjusted priorities of the vehicle navigation targets and the preloaded traffic map to obtain an updated path includes: Acquire the current position and the target position of the current vehicle, determine a plurality of optional paths in the traffic map according to the current position and the target position, and calculate reference values ​​corresponding to each vehicle navigation target of each optional path based on characteristic information in the traffic map; An updated path is determined according to reference values ​​corresponding to each vehicle navigation target of each optional path, priorities of each vehicle navigation target, and a preset path planning method.

[0011] Optionally, the step of determining the updated path according to the reference values ​​corresponding to the vehicle navigation targets of the optional paths, the priorities of the vehicle navigation targets and a preset path planning method includes: The reference values ​​corresponding to the vehicle navigation targets of the optional paths and the priorities of the vehicle navigation targets are input into the trained optimal path output model to obtain an updated path.

[0012] Optionally, the step of determining the updated path according to the reference values ​​corresponding to the vehicle navigation targets of the optional paths, the priorities of the vehicle navigation targets and a preset path planning method includes: The reference values ​​corresponding to the vehicle navigation targets of the optional paths are fuzzy processed to obtain corresponding fuzzy sets, and the updated paths are determined based on the fuzzy sets corresponding to the vehicle navigation targets of the optional paths, the priorities of the vehicle navigation targets and preset fuzzy rules.

[0013] Optionally, before obtaining the environmental status information within the communication range of the current vehicle and the traffic information of the current planned path, the following steps are included: Each vehicle on the road is identified as a vehicle node, and based on the Internet of Vehicles technology, each vehicle node within a preset communication range is enabled to communicate with each other so as to share their own information.

[0014] Optionally, the current vehicle in the navigation control process further includes: Parameters of the strategy network and the value network are continuously updated according to a preset strategy network update formula and a preset value network update formula, wherein the preset strategy network update formula includes:

[0015] Among them, the It is The policy network parameters of the vehicle, is the objective function, yes - function, It is Car strategy.

[0016] In a second aspect, the embodiment of the present application further provides a path planning and scheduling device for an autonomous driving vehicle, including: An acquisition module is used to obtain the environmental status information within the communication range of the current vehicle and the traffic information of the current planned path; An action determination module, used for inputting the environmental state information into the trained intelligent agent model to obtain the execution action information of the current vehicle; A control module, configured to control the operation of the current vehicle according to the execution action information, wherein the environmental state information includes driving information received from other vehicles and detected road surface information; A priority adjustment module, configured to obtain the driving task of the current vehicle and each preset vehicle navigation target when detecting that the traffic information satisfies a preset path update condition, and adjust the priority of each vehicle navigation target according to the driving task and the traffic information; A path updating module, configured to adjust the currently planned path based on the adjusted priorities of the respective vehicle navigation targets and a pre-loaded traffic map to obtain an updated path; A navigation module is used to perform navigation control of the current vehicle based on the updated path.

[0017] In a third aspect, an embodiment of the present application further provides a path planning and scheduling device for an autonomous driving vehicle, the device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-objective collaborative path planning and scheduling method for the autonomous driving vehicle described in the embodiment of the present application.

[0018] In a fourth aspect, an embodiment of the present application further provides a storage medium storing computer executable instructions, which, when executed by a computer processor, are used to execute the multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle described in an embodiment of the present application.

[0019] In the embodiment of the present application, the environmental state information within the communication range of the current vehicle and the traffic information of the current planned path are obtained, and the environmental state information is input into the trained intelligent model to obtain the execution action information of the current vehicle, and the operation of the current vehicle is controlled according to the execution action information. The environmental state information includes the driving information sent by other vehicles and the detected road surface information. When the traffic information is detected to meet the preset path update conditions, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information. The current planned path is adjusted based on the adjusted priority of each vehicle navigation target and the pre-loaded traffic map to obtain an updated path, and the navigation control of the current vehicle is performed based on the updated path. This solution solves the problem that the vehicle's perception depth of the environment is not enough in the existing navigation method, the environment cannot be fully and accurately modeled, and the accuracy of navigation path planning is affected. The navigation target weight cannot be adjusted in real time, which may delay the navigation task. It can enhance the vehicle's perception and understanding of complex dynamic environments, realize the adaptive adjustment of navigation target weights, and improve the accuracy and efficiency of vehicle navigation control. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle provided in an embodiment of the present application; Figure 2 A flowchart of another method for multi-objective collaborative path planning and scheduling of an autonomous driving vehicle provided in an embodiment of the present application; Figure 3 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including a method for constructing a traffic map provided in an embodiment of the present application; Figure 4 A schematic diagram of a traffic map provided in an embodiment of the present application; Figure 5 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including an updated path determination method provided in an embodiment of the present application; Figure 6 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including an inter-vehicle communication method provided in an embodiment of the present application; Figure 7 A module structure diagram of a path planning and scheduling device for an autonomous driving vehicle provided in an embodiment of the present application; Figure 8 A schematic diagram of the structure of a path planning and scheduling device for an autonomous driving vehicle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the embodiments of the present application, rather than to limit the embodiments of the present application. It should also be noted that, for ease of description, only parts related to the embodiments of the present application are shown in the accompanying drawings, rather than all structures.

[0022] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and or or" in the specification and claims represents at least one of the connected objects, and the character "or" generally indicates that the objects associated before and after are in an "or" relationship.

[0023] The embodiment of the present application provides a method for multi-objective collaborative path planning and scheduling of an autonomous driving vehicle, which can be applied to the path planning and scheduling scenarios of the vehicle during autonomous driving. For example, when an autonomous driving vehicle encounters an accident on the road, the path can be replanned using the method for multi-objective collaborative path planning and scheduling of an autonomous driving vehicle provided in the embodiment of the present application. The embodiment of the present application provides a method for multi-objective collaborative path planning and scheduling of an autonomous driving vehicle, and the execution subject of each step can be an autonomous driving vehicle.

[0024] Figure 1 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, specifically including: Step S101: Obtain environmental status information within the communication range of the current vehicle and traffic information of the current planned path.

[0025] Among them, the communication range can be the geographical range in which the current vehicle can carry out communication activities. The communication range can be set according to the actual situation or according to the specifications of the communication equipment, and the current vehicle can communicate with other vehicles within the communication range. Environmental status information can be status information of various factors that affect vehicle driving in the environment of the current vehicle. The planned path is used to characterize the driving path referenced by the vehicle for navigation control. The current vehicle can automatically navigate to the destination using the planned path. Traffic information can be traffic-related information on the current planned path, such as road congestion information, traffic accident information and other traffic information. In one embodiment, before obtaining the environmental status information within the communication range of the current vehicle and the traffic information of the current planned path, the driving information of the vehicle is collected by a centralized control center so that the vehicles can share their own driving information. Optionally, a vehicle driving information sharing method can be to determine each vehicle on the road as a vehicle node, and enable each vehicle node within the preset communication range to communicate based on the vehicle networking technology, so as to share their own information between each vehicle. By treating each vehicle as an independent node and using vehicle networking technology to communicate with surrounding vehicles, multiple autonomous vehicles can exchange information and make collaborative decisions in a decentralized environment, improving the robustness and scalability of the system.

[0026] Step S102: input the environmental state information into the trained intelligent body model to obtain the execution action information of the current vehicle, and control the operation of the current vehicle according to the execution action information. The environmental state information includes the driving information sent by other vehicles and the detected road surface information.

[0027] Among them, the agent model is used to characterize a system that can perceive the environment, make decisions and perform actions. The agent corresponding to the agent model can be an autonomous driving vehicle. The execution action information of the current vehicle can be obtained by using the agent model. The execution action information can be the action information performed by the current vehicle in the current state, such as acceleration, deceleration, steering and other action information. The execution action information can be used to control the current vehicle to perform corresponding actions. Driving information can be the state information of other vehicles when driving, such as speed, acceleration and other state information. Road surface information can be relevant parameter information of the road surface around the current vehicle, such as road curvature, road slope and other parameter information. The road surface information can be collected by sensors installed on the current vehicle. In one embodiment, a training process of an agent model can be to construct a deep neural network, determine the sample environment state information as the input data of the deep neural network, determine the action probability distribution information as the output data of the deep neural network, the sample environment state information includes the driving information of other vehicles and the road surface information, the driving information includes the vehicle position, speed and acceleration, the road surface information includes the road curvature, road surface slope and traffic signal parameters, and perform reinforcement learning training on the deep neural network based on a preset reward mechanism. Building an intelligent agent model through deep neural networks can enable autonomous vehicles to continuously learn and make adaptive decisions in complex dynamic environments.

[0028] Step S103: when it is detected that the traffic information satisfies a preset path update condition, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information.

[0029] Among them, the preset path update condition can be a pre-set condition for determining whether the current vehicle is updating the path. The preset path update condition can be used to determine whether to adjust the priority of each preset vehicle navigation target to update the vehicle's driving path. The driving task can be the task information that the current vehicle needs to complete. For example, the driving task of the current vehicle is to go to a central hospital. The vehicle navigation target can be the target that the vehicle expects to achieve during navigation control, such as vehicle navigation targets such as short driving time, low energy consumption, and high comfort. Priority is used to characterize the order in which the autonomous driving vehicle considers each vehicle navigation target during navigation. Optionally, a way to determine whether the traffic information meets the preset path update condition can be to detect whether there is a preset path update event in the traffic information. If there is a preset path update event, it is determined that the traffic information meets the preset path update condition, otherwise, it is determined that the traffic information does not meet the preset path update condition. An exemplary example can be that the preset path update event includes traffic congestion, traffic accidents, and road repairs. It is detected whether there is a traffic congestion event or a traffic accident event or a road repair event in the traffic information. By detecting that there is a traffic accident event in the traffic information, it is determined that the preset path update condition is met. In another embodiment, a method for determining whether traffic information meets the preset path update conditions may be to detect whether the vehicle flow value in the traffic information is greater than the corresponding preset value. If the vehicle flow value is greater than the corresponding preset value, it is determined that the traffic information meets the preset path update conditions; otherwise, it is determined that the traffic information does not meet the preset path update conditions.

[0030] In one embodiment, the vehicle navigation target includes driving time and energy reserve. A method for adjusting the priority of the vehicle navigation target may be to calculate the corresponding weight according to the preset weight calculation formula of each vehicle navigation target. The weight calculation formula of the driving time is: , the energy reserve weight calculation formula is: , represents the traffic congestion index, Indicates the urgency of the task. Represents the vehicle energy value, Indicates the energy-saving area labeling value, calculates the traffic congestion index and the task urgency according to the driving task and traffic information, obtains the vehicle energy value of the current vehicle and the energy-saving area labeling value in the traffic information, inputs the traffic congestion index, the task urgency, the vehicle energy value and the energy-saving area labeling value into the corresponding preset weight calculation formula to obtain the weight of each vehicle navigation target, and adjusts the priority according to the weight of each vehicle navigation target. Among them, the preset weight calculation formula can be a pre-set formula for calculating the importance of each vehicle navigation target. The driving time can be the time for the current vehicle to complete the driving task. The weight of the driving time can be used to determine the current vehicle's requirements for the length of the driving time. The energy reserve can be the current vehicle's reserve of energy consumed when driving. The weight of the energy reserve can be used to determine the current vehicle's requirements for the amount of energy reserve. The traffic congestion index is used to characterize the congestion of the current planned path. The higher the traffic congestion index, the more congested the path. The task urgency is used to characterize the urgency of the completion of the current vehicle's driving task. The higher the task urgency, the shorter the time required to complete the driving task. The vehicle energy value is used to represent the current energy value of the vehicle. The lower the vehicle energy value, the more energy-saving the vehicle needs and the higher the demand for energy reserves. The energy-saving area marking value can be the control value of the energy-saving area for vehicle energy consumption. The lower the energy-saving area marking value, the higher the demand for energy reserves.

[0031] Optionally, the traffic information includes the vehicle density at the congestion location and the vehicle speed at the congestion location, and the driving task includes the task type. A method for calculating the traffic congestion index and the task urgency may be to substitute the vehicle density at the congestion location and the vehicle speed at the congestion location in the traffic information into a preset congestion calculation formula to obtain a traffic congestion index value. The preset congestion calculation formula includes , represents the vehicle density at the congested location, Indicates the vehicle speed at the congested location. According to the task type in the driving task, the preset task urgency comparison table is queried to obtain the task urgency value. Among them, the vehicle density at the congested location can be the density of vehicles at the location where the traffic congestion occurs, the vehicle speed at the congested location can be the driving speed of the vehicle at the location where the traffic congestion occurs, and the task type can be the classification of the driving task, such as medical type, fire fighting type, etc. An exemplary example can be that the traffic congestion index ranges from 0 to 10, the vehicle density at the congested location is 170-250 vehicles / km, and the vehicle speed at the congested location is 20 km / h. Substitute 170-250 vehicles / km and 20 km / h into the formula The traffic congestion index is 6, and the preset task urgency comparison table is shown in the following table.

[0032] The above comparison table specifically describes the comparison relationship between the task type and task urgency of the driving task, and the specific description is as follows: when the task type is medical type, the corresponding task urgency is level three; when the task type is firefighting type, the corresponding task urgency is level three; when the task type is logistics type, the corresponding task urgency is level two; when the task type is daily travel type, the corresponding task urgency is level one.

[0033] The task type of the current driving task is a medical type, and the corresponding task urgency is obtained by querying the above comparison table as level three.

[0034] Alternatively, a method for calculating the traffic congestion index and the task urgency may be to query the corresponding preset congestion index comparison table according to the vehicle density at the congested location and the vehicle speed at the congested location in the traffic information to obtain the corresponding congestion index to be calculated, multiply the congestion index to be calculated of the vehicle density at the congested location and the congested location vehicle speed by the corresponding preset weights and add them together to obtain the traffic congestion index, and query the preset task urgency comparison table according to the destination in the driving task to obtain the task urgency value.

[0035] In one embodiment, a method for adjusting the priority of a vehicle navigation target may be to obtain a weight calculation formula for each vehicle navigation target and a corresponding parameter to be input, match each parameter to be input with various parameter data in a driving task and traffic information to obtain a corresponding input value, input the input value of each parameter to be input into the corresponding weight calculation formula to obtain the weight of each vehicle navigation target, and adjust the priority according to the weight of each vehicle navigation target.

[0036] Step S104: adjusting the current planned path based on the adjusted priorities of the navigation targets of the vehicle and the pre-loaded traffic map to obtain an updated path, and performing navigation control of the current vehicle based on the updated path.

[0037] Among them, the traffic map can be a map constructed by various elements in the traffic environment and the relationship between the elements. The update path is used to represent the driving path obtained after the current planned path is updated. Optionally, a method of constructing a traffic map can be to construct a traffic map based on traffic information, the traffic map includes multiple traffic nodes and edges of connection relationships between each traffic node, the traffic nodes include road network nodes, vehicle nodes and obstacle nodes, the edges include weight information, and the graph convolution layer is used to continuously update and aggregate the features of each traffic node to obtain a new feature representation of each traffic node, which is used to capture the spatial relationship and semantic information between the elements. By abstracting the traffic elements into a graph structure and accurately capturing the spatial relationship and semantic information between the elements, comprehensive and accurate environmental perception can be provided for subsequent path planning.

[0038] In one embodiment, a method for determining an updated path may be to obtain the current position and target position of the current vehicle, determine multiple optional paths in the traffic map based on the current position and target position, determine the vehicle navigation target with the highest priority, calculate the reference value of the vehicle navigation target with the highest priority for each optional path based on the characteristic information in the traffic map, and compare the reference values ​​corresponding to each optional path to obtain an updated path. Optionally, a method for determining an updated path may be to obtain the current position and target position of the current vehicle, determine multiple optional paths in the traffic map based on the current position and target position, calculate the reference value corresponding to each vehicle navigation target of each optional path based on the characteristic information in the traffic map, and determine the updated path based on the reference value corresponding to each vehicle navigation target of each optional path, the priority of each vehicle navigation target, and a preset path planning method. By updating the path based on the adjusted priority of each vehicle navigation target and the traffic map, the actual needs in different scenarios can be met, and the comprehensive performance and user experience of the vehicle are improved.

[0039] In one embodiment, during the navigation control process, the current vehicle continuously updates the parameters of the strategy network and the value network according to a preset strategy network update formula and a preset value network update formula, wherein the preset strategy network update formula includes:

[0040] Among them, the It is The policy network parameters of the vehicle, is the objective function, yes - function, It is The strategy of the vehicle. Among them, the policy network is a neural network used to model the strategy of the intelligent agent, that is, the probability distribution of selecting an action in a given state. Its input is the current state, and the output is the probability of selecting each action in this state. The purpose of the policy network is to guide the intelligent agent to make the best decision in each state, so as to maximize the cumulative return. The value network is a neural network used to estimate the expected cumulative return that can be obtained in a given state or after taking a certain action. During the navigation control process, the vehicle repeatedly interacts and collaborates with the surrounding vehicles, shares information and experience between vehicles, substitutes the shared information into the preset policy network update formula and the preset value network update formula, and continuously updates the vehicle's policy network parameters and value network parameters. As the learning progresses, the vehicle cluster gradually forms an efficient collaborative scheduling strategy, such as maintaining appropriate spacing and speed in the convoy formation driving, and reasonably diverting vehicles in congested sections. By continuously updating the parameters of the policy network and the value network, the vehicle cluster can adaptively respond to various complex traffic scenarios and improve the operating efficiency and safety of the entire vehicle cluster.

[0041] As can be seen from the above, the environmental state information within the communication range of the current vehicle and the traffic information of the current planned path are obtained, and the environmental state information is input into the trained intelligent model to obtain the execution action information of the current vehicle, and the current vehicle operation is controlled according to the execution action information. The environmental state information includes the driving information sent by other vehicles and the detected road surface information. When the traffic information is detected to meet the preset path update conditions, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and traffic information. The current planned path is adjusted based on the adjusted priority of each vehicle navigation target and the pre-loaded traffic map to obtain an updated path, and the navigation control of the current vehicle is performed based on the updated path. This solution solves the problem that the vehicle's perception depth of the environment is not enough in the existing navigation method, and the environment cannot be fully and accurately modeled, which affects the accuracy of navigation path planning, and the navigation target weight cannot be adjusted in real time, which may delay the navigation task. It can enhance the vehicle's perception and understanding of complex dynamic environments, realize the adaptive adjustment of navigation target weights, and improve the accuracy and efficiency of vehicle navigation control.

[0042] Figure 2 A flowchart of another method for multi-objective collaborative path planning and scheduling of an autonomous driving vehicle provided in an embodiment of the present application is provided, which provides a specific method for optional intelligent agent model training, such as Figure 2 As shown, specifically including: Step S201, construct a deep neural network, determine sample environmental state information as input data of the deep neural network, and determine action probability distribution information as output data of the deep neural network, the sample environmental state information includes driving information of other vehicles and road surface information, the driving information includes vehicle position, speed and acceleration, and the road surface information includes road curvature, road slope and traffic signal parameters.

[0043] Among them, the deep neural network is composed of multiple layers of neurons, including an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving the original data, the hidden layer processes and transforms the data layer by layer, and finally the output layer outputs the processing results. The neurons in each layer receive the output of the previous layer, perform weighted summation and nonlinear transformation, and then pass it to the next layer. Through this layer-by-layer information transmission and transformation, the network can gradually learn the high-level feature representation of the data. The action probability distribution information can be the distribution information of the probability of an autonomous driving vehicle performing each action. The driving information and road information of other vehicles are used as the input of the deep neural network, and the action probability distribution information of the current vehicle is used as the output of the deep neural network, which can learn the relationship between the driving information and road information of other vehicles and the execution action of the current vehicle.

[0044] Step S202: Perform reinforcement learning training on the deep neural network based on a preset reward mechanism.

[0045] Among them, the preset reward mechanism is used to represent the pre-input reward mechanism. The reward mechanism refers to the process of guiding the behavior of the agent by giving it rewards or punishments in the reinforcement learning process so that it can achieve the expected goal. Optionally, a reinforcement learning training method can be to randomly select an action from the action probability distribution information output by the deep neural network when inputting sample environmental state information, input the environmental state information and the action into the reward function to obtain the reward result, update and adjust the parameters of the deep neural network according to the preset deep Q network algorithm, and repeat the execution to train the agent model. The update formula of the deep Q network algorithm is , is the learning rate, is the discount factor, is the state-action value function, which means that in the state Take action An illustrative example could be to define a reward function , when the vehicle performs action a and is in state s, the vehicle can maintain a safe distance and move in the target direction, a positive reward is given, and if a collision occurs or traffic rules are violated, a negative reward is given. In another embodiment, when the sample environment state information is input, an action in the action probability distribution information output by the deep neural network is randomly selected, the environment state information and the action are input into the reward function to obtain the reward result, and the parameters of the deep neural network are updated and adjusted according to the preset priority experience playback DQN algorithm, and the execution is repeated to train the intelligent agent model.

[0046] Step S203: Obtain environmental status information within the communication range of the current vehicle and traffic information of the current planned path.

[0047] Step S204: input the environmental status information into the trained intelligent body model to obtain the execution action information of the current vehicle, and control the operation of the current vehicle according to the execution action information. The environmental status information includes the driving information sent by other vehicles and the detected road surface information.

[0048] Step S205: When it is detected that the traffic information satisfies the preset path update condition, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information.

[0049] Step S206: adjusting the current planned path based on the adjusted priorities of the navigation targets of the vehicle and the pre-loaded traffic map to obtain an updated path, and performing navigation control of the current vehicle based on the updated path.

[0050] From the above, we can know that a deep neural network is constructed, sample environment state information is determined as the input data of the deep neural network, and action probability distribution information is determined as the output data of the deep neural network. The sample environment state information includes driving information of other vehicles and road information. Driving information includes vehicle position, speed and acceleration. Road information includes road curvature, road slope and traffic signal parameters. The deep neural network is reinforced based on a preset reward mechanism. This solution constructs an intelligent agent model through a deep neural network, which enables autonomous driving vehicles to continuously learn and make adaptive decisions in complex dynamic environments.

[0051] Figure 3 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including a method for constructing a traffic map is provided in an embodiment of the present application, such as Figure 3 As shown, specifically including: Step S301: Obtain environmental status information within the communication range of the current vehicle and traffic information of the current planned path.

[0052] Step S302: input the environmental status information into the trained intelligent model to obtain the execution action information of the current vehicle, and control the operation of the current vehicle according to the execution action information. The environmental status information includes the driving information sent by other vehicles and the detected road surface information.

[0053] Step S303: When it is detected that the traffic information satisfies the preset path update condition, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information.

[0054] Step S304: construct the traffic map according to the traffic information, wherein the traffic map includes a plurality of traffic nodes and edges that connect the traffic nodes, the traffic nodes include road network nodes, vehicle nodes, and obstacle nodes, and the edges include weight information.

[0055] The weight information on the edge can be used to represent the distance between connected traffic nodes or traffic flow information. Figure 4 As shown, Figure 4 A schematic diagram of a traffic map provided for an embodiment of the present application, wherein 01 represents a vehicle node, 02 represents a road network node, 03 represents an obstacle node, the lines connecting the nodes represent the edges of the connection relationship between the nodes, and the decimals on each edge represent the weight information on the edge.

[0056] Step S305: Use a graph convolution layer to continuously update and aggregate the features of each of the traffic nodes to obtain a new feature representation of each of the traffic nodes, which is used to capture the spatial relationship and semantic information between the elements.

[0057] Among them, the graph convolution layer is a neural network layer specially designed to process graph structure data. Different from the traditional convolution layer that processes Euclidean data, the graph convolution layer processes non-Euclidean data, that is, graph data. The graph convolution layer can capture the complex relationship between nodes and the structural characteristics of the graph by performing convolution operations on the graph data. Traffic node features can be attributes of traffic nodes, such as the type and speed of vehicle nodes, the type of road nodes, and other attributes. In one embodiment, a continuous update and aggregation method of traffic node features based on the graph convolution layer can be to continuously update and aggregate the features of each traffic node through multi-layer graph convolution operations, and finally obtain a new feature representation of each traffic node.

[0058] Step S306: adjusting the current planned path based on the adjusted priorities of the navigation targets of the vehicle and the pre-loaded traffic map to obtain an updated path, and performing navigation control of the current vehicle based on the updated path.

[0059] From the above, we can see that a traffic map is constructed based on traffic information. The traffic map includes multiple traffic nodes and edges of connection relationships between traffic nodes. Traffic nodes include road network nodes, vehicle nodes, and obstacle nodes. Edges include weight information. The graph convolution layer is used to continuously update and aggregate the features of each traffic node to obtain a new feature representation of each traffic node, which is used to capture the spatial relationship and semantic information between elements. This solution accurately captures the spatial relationship and semantic information between elements by abstracting traffic elements into a graph structure, which can provide comprehensive and accurate environmental perception for subsequent path planning.

[0060] Figure 5 A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including an updated path determination method provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, specifically including: Step S401: Obtain environmental status information within the communication range of the current vehicle and traffic information of the current planned path.

[0061] Step S402: input the environmental status information into the trained intelligent model to obtain the execution action information of the current vehicle, and control the operation of the current vehicle according to the execution action information. The environmental status information includes the driving information sent by other vehicles and the detected road surface information.

[0062] Step S403: When it is detected that the traffic information satisfies the preset path update condition, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information.

[0063] Step S404, obtaining the current position and the target position of the current vehicle, determining multiple optional paths in the traffic map according to the current position and the target position, and calculating reference values ​​corresponding to each vehicle navigation target of each optional path based on feature information in the traffic map.

[0064] Among them, the current position is used to characterize the current position of the vehicle, and the target position is used to characterize the geographical location that the current vehicle is expected to reach. The optional path can be a path that can be driven from the current position to the target position. The reference value can be a parameter value of each vehicle navigation target used to evaluate the optional path, such as a reference value such as a driving time value and an energy consumption value. Optionally, a reference value calculation method can be to obtain a preset calculation formula and a demand parameter of the reference value corresponding to each vehicle navigation target, and input the characteristic data of the optional path in the characteristic information that matches the demand parameter of the preset calculation formula into the preset calculation formula to obtain the reference value of the corresponding vehicle navigation target. An illustrative example may be that the reference value corresponding to the driving time navigation target of optional path a is the driving time of optional path a, and the driving time calculation formula of the path is the sum of the driving times of the sections obtained by dividing the section lengths of all sections by the product of the real-time section traffic flow and the average speed of the vehicles on the sections, and the demand parameters are the section lengths of each section of optional path a, the real-time section traffic flow and the average speed of the vehicles on the sections, and the section lengths of each section of optional path a, the real-time section traffic flow and the average speed of the vehicles on the sections in the characteristic information of optional path a in the traffic map are substituted into the driving time calculation formula of the path to obtain the driving time of optional path a.

[0065] Step S405: determining an updated path according to reference values ​​corresponding to each vehicle navigation target of each optional path, priorities of each vehicle navigation target and a preset path planning method, and performing navigation control of the current vehicle based on the updated path.

[0066] Among them, the preset path planning method can be a pre-set method for planning the driving path of the autonomous driving vehicle. In one embodiment, a method for determining an updated path can be to input the reference value corresponding to each vehicle navigation target of each optional path and the priority of each vehicle navigation target into the trained optimal path output model to obtain an updated path. Among them, the optimal path output model can be obtained by using a multi-layer neural network to learn the mapping relationship between input and output through a large amount of training data, and the training data includes sample reference values ​​corresponding to each vehicle navigation target of the sample path and sample priorities of each vehicle navigation target.

[0067] In another embodiment, a method for determining an updated path may be to fuzzy the reference values ​​corresponding to each vehicle navigation target of each optional path to obtain a corresponding fuzzy set, and determine the updated path based on the fuzzy sets corresponding to each vehicle navigation target of each optional path, the priority of each vehicle navigation target, and preset fuzzy rules. Among them, the fuzzy set may be a set that expresses fuzzy concepts. For example, for driving time, fuzzy sets such as "short", "medium", and "long" may be defined. The preset fuzzy rules may be pre-set fuzzy rules formulated based on expert knowledge or experimental data. For example, if the driving time is "short", the energy consumption is "low", and the comfort level is "high", then the decision of the path selection is "excellent"; if the driving time is "long", the energy consumption is "high", and the comfort level is "low", then the decision is "poor", etc. An exemplary example may be that the optional paths are path a, path b, path c and path d, and the reference values ​​of the driving time, energy consumption and comfort of each optional path are compared with the numerical ranges of the corresponding fuzzy sets to obtain that the fuzzy set of the driving time of path a is "short", the fuzzy set of energy consumption is "high", and the fuzzy set of comfort is "high", the fuzzy set of the driving time of path b is "short", the fuzzy set of energy consumption is "low", and the fuzzy set of comfort is "low", the fuzzy set of the driving time of path c is "short", the fuzzy set of energy consumption is "low", and the fuzzy set of comfort is "medium", the fuzzy set of the driving time of path d is "long", the fuzzy set of energy consumption is "low", and the fuzzy set of comfort is "low", the priority of each vehicle navigation target is driving time>energy consumption>comfort, and the fuzzy rule is that the driving time is "short" and the energy consumption is "low". If the comfort level is "high", the decision of path selection is "excellent", and the paths with the best travel time among the optional paths are preferentially determined as paths a, b and c. Then, the paths with the best energy consumption are determined from paths a, b and c. Finally, the path with the best comfort level is determined from paths b and c as path c, and path c is determined as the updated path.

[0068] As can be seen from the above, the current position and target position of the current vehicle are obtained, multiple optional paths in the traffic map are determined based on the current position and target position, and the reference values ​​corresponding to the navigation targets of each vehicle on each optional path are calculated based on the characteristic information in the traffic map. The updated path is determined based on the reference values ​​corresponding to the navigation targets of each vehicle on each optional path, the priority of each vehicle navigation target, and the preset path planning method. This solution updates the path by adjusting the priority of each vehicle navigation target and the traffic map, which can meet the actual needs in different scenarios and improve the comprehensive performance and user experience of the vehicle.

[0069] Figure 6A flowchart of a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle including an inter-vehicle communication method provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, specifically including: Step S501: Determine each vehicle on the road as a vehicle node, and enable each vehicle node within a preset communication range to communicate based on the Internet of Vehicles technology, so as to share information between each vehicle.

[0070] Among them, the preset communication range can be a preset range within which vehicles can communicate with each other, and the preset communication range can be set according to the actual traffic conditions or the specifications of the communication equipment. ‌Internet of Vehicles technology‌ refers to a technology that closely connects vehicles, vehicle owners, other vehicles, road infrastructure and service platforms through a new generation of information and communication technology to achieve all-round network interconnection and information sharing, such as dedicated short-range communication technology (DSRC) or cellular vehicle-to-everything (C-V2X). An exemplary example may be that the preset communication range is ten meters, and the vehicles currently within the ten-meter range are vehicle 1, vehicle 2, vehicle 3, vehicle 4 and vehicle 5. The five vehicles currently within the ten-meter range are respectively determined as an independent vehicle node, and the cellular vehicle-to-everything (C-V2X) technology is used to enable vehicles 1, vehicle 2, vehicle 3, vehicle 4 and vehicle 5 to communicate with each other to share their own location, speed, mission status and other information.

[0071] Step S502: Obtain environmental status information within the communication range of the current vehicle and traffic information of the current planned path.

[0072] Step S503: input the environmental status information into the trained intelligent model to obtain the execution action information of the current vehicle, and control the operation of the current vehicle according to the execution action information. The environmental status information includes the driving information sent by other vehicles and the detected road surface information.

[0073] Step S504: when it is detected that the traffic information satisfies the preset path update condition, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and the traffic information.

[0074] Step S505: adjusting the current planned path based on the adjusted priorities of the navigation targets of the vehicle and the pre-loaded traffic map to obtain an updated path, and performing navigation control of the current vehicle based on the updated path.

[0075] From the above, we can see that each vehicle on the road is identified as a vehicle node, and each vehicle node within the preset communication range is enabled to communicate based on the Internet of Vehicles technology, so as to share information between each vehicle. By treating each vehicle as an independent node and using the Internet of Vehicles technology to communicate with surrounding vehicles, multiple autonomous driving vehicles can exchange information and make collaborative decisions in a decentralized environment, thus improving the robustness and scalability of the system.

[0076] Figure 7 This is a module structure diagram of a path planning and scheduling device for an autonomous driving vehicle provided in an embodiment of the present application. The system is used to execute a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle provided in the above embodiment, and has the corresponding functional modules and beneficial effects of the execution method. Figure 7 As shown, the system specifically includes: An acquisition module 101 is used to acquire environmental status information within the communication range of the current vehicle and traffic information of the current planned path; An action determination module 102 is used to input the environmental state information into the trained agent model to obtain the execution action information of the current vehicle; A control module 103, configured to control the operation of the current vehicle according to the execution action information, wherein the environmental state information includes driving information received from other vehicles and detected road surface information; The priority adjustment module 104 is used to obtain the driving task of the current vehicle and each preset vehicle navigation target when it is detected that the traffic information meets the preset path update condition, and adjust the priority of each vehicle navigation target according to the driving task and the traffic information; A path updating module 105, configured to adjust the currently planned path based on the adjusted priorities of the respective vehicle navigation targets and the pre-loaded traffic map to obtain an updated path; The navigation module 106 is used to perform navigation control of the current vehicle based on the updated path.

[0077] It can be seen from the above scheme that the environmental state information within the communication range of the current vehicle and the traffic information of the current planned path are obtained, and the environmental state information is input into the trained intelligent model to obtain the execution action information of the current vehicle, and the current vehicle operation is controlled according to the execution action information. The environmental state information includes the driving information sent by other vehicles and the detected road surface information. When the traffic information is detected to meet the preset path update conditions, the driving task of the current vehicle and each preset vehicle navigation target are obtained, and the priority of each vehicle navigation target is adjusted according to the driving task and traffic information. The current planned path is adjusted based on the adjusted priority of each vehicle navigation target and the pre-loaded traffic map to obtain an updated path, and the navigation control of the current vehicle is performed based on the updated path. This scheme solves the problem that the vehicle's perception depth of the environment is not enough in the existing navigation method, and the environment cannot be fully and accurately modeled, which affects the accuracy of navigation path planning, and the navigation target weight cannot be adjusted in real time, which may delay the navigation task. It can enhance the vehicle's perception and understanding of complex dynamic environments, realize the adaptive adjustment of navigation target weights, and improve the accuracy and efficiency of vehicle navigation control.

[0078] In a possible embodiment, a model training module is further included, which is specifically used to: A deep neural network is constructed, sample environmental state information is determined as input data of the deep neural network, and action probability distribution information is determined as output data of the deep neural network, the sample environmental state information includes driving information of other vehicles and road surface information, the driving information includes vehicle position, speed and acceleration, the road surface information includes road curvature, road slope and traffic signal parameters, and reinforcement learning training is performed on the deep neural network based on a preset reward mechanism.

[0079] In a possible embodiment, the model training module is further used to: When the sample environment state information is input, randomly select an action in the action probability distribution information output by the deep neural network, and input the environment state information and the action into a reward function to obtain a reward result; The parameters of the deep neural network are updated and adjusted according to the preset deep Q network algorithm, and the training is repeated to obtain the intelligent agent model. The update formula of the deep Q network algorithm is: , is the learning rate, is the discount factor, is the state-action value function, which means that in the state Take action expected long-term rewards.

[0080] In a possible embodiment, the priority adjustment module 104 is specifically configured to: The corresponding weights are calculated according to the preset weight calculation formulas of each of the vehicle navigation targets. The weight calculation formula of the driving time is: , the energy reserve weight calculation formula is: , represents the traffic congestion index, Indicates the urgency of the task. Represents the vehicle energy value, Indicates the energy-saving area marking value; The traffic congestion index and the task urgency are calculated according to the driving task and the traffic information, and the vehicle energy value of the current vehicle and the energy-saving area marking value in the traffic information are obtained, and the traffic congestion index, the task urgency, the vehicle energy value and the energy-saving area marking value are respectively input into the corresponding preset weight calculation formula to obtain the weight of each of the vehicle navigation targets, and the priority is adjusted according to the weight of each of the vehicle navigation targets.

[0081] In a possible embodiment, the priority adjustment module 104 is further configured to: Substituting the vehicle density and vehicle speed of the congested location in the traffic information into a preset congestion calculation formula to obtain a traffic congestion index value, the preset congestion calculation formula includes , represents the vehicle density at the congested location, Indicates the vehicle speed at the congested location; According to the task type in the driving task, a preset task urgency comparison table is queried to obtain a task urgency value.

[0082] In a possible embodiment, a graph neural network modeling module is also included, which is specifically used to: Constructing the traffic map according to the traffic information, the traffic map comprising a plurality of traffic nodes and edges of connection relationships between the traffic nodes, the traffic nodes comprising road network nodes, vehicle nodes and obstacle nodes, and the edges comprising weight information; A graph convolutional layer is used to continuously update and aggregate the features of each traffic node to obtain a new feature representation of each traffic node, which is used to capture the spatial relationship and semantic information between the elements.

[0083] In a possible embodiment, the path updating module 105 is specifically configured to: Acquire the current position and the target position of the current vehicle, determine a plurality of optional paths in the traffic map according to the current position and the target position, and calculate reference values ​​corresponding to each vehicle navigation target of each optional path based on characteristic information in the traffic map; An updated path is determined according to reference values ​​corresponding to each vehicle navigation target of each optional path, priorities of each vehicle navigation target, and a preset path planning method.

[0084] In a possible embodiment, the path updating module 105 is further configured to: The reference values ​​corresponding to the vehicle navigation targets of the optional paths and the priorities of the vehicle navigation targets are input into the trained optimal path output model to obtain an updated path.

[0085] In a possible embodiment, the path updating module 105 is further configured to: The reference values ​​corresponding to the vehicle navigation targets of the optional paths are fuzzy processed to obtain corresponding fuzzy sets, and the updated paths are determined based on the fuzzy sets corresponding to the vehicle navigation targets of the optional paths, the priorities of the vehicle navigation targets and preset fuzzy rules.

[0086] In a possible embodiment, a communication module is further included, specifically configured to: Each vehicle on the road is identified as a vehicle node, and based on the Internet of Vehicles technology, each vehicle node within a preset communication range is enabled to communicate with each other so as to share their own information.

[0087] In a possible embodiment, a parameter updating module is further included, which is specifically used to: Parameters of the strategy network and the value network are continuously updated according to a preset strategy network update formula and a preset value network update formula, wherein the preset strategy network update formula includes:

[0088] Among them, the It is The policy network parameters of the vehicle, is the objective function, yes - function, It is Car strategy.

[0089] Figure 8 A schematic diagram of a path planning and scheduling device for an autonomous driving vehicle provided in an embodiment of the present application is shown in FIG. Figure 8As shown, the device includes a processor 201, a memory 202, an input device 203 and an output device 204; the number of processors 201 in the device can be one or more. Figure 8 A processor 201 is taken as an example; the processor 201, memory 202, input device 203 and output device 204 in the device can be connected by a bus or other means. Figure 8 The example of connecting through a bus is taken. The memory 202, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions or modules corresponding to a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle in an embodiment of the present application. The processor 201 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 202, that is, realizes the above-mentioned multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle. The input device 203 can be used to receive input digital or character information, and to generate key signal input related to user settings and function controls of the device. The output device 204 may include a display device such as a display screen.

[0090] The present application also provides a storage medium containing computer executable instructions, which are used to execute a multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle when executed by a computer processor. The method includes: Obtaining environmental status information within the communication range of the current vehicle and traffic information of the current planned path, inputting the environmental status information into the trained agent model to obtain execution action information of the current vehicle, and controlling the operation of the current vehicle according to the execution action information, wherein the environmental status information includes driving information sent by other vehicles and detected road surface information; When it is detected that the traffic information satisfies a preset path update condition, obtaining the driving task of the current vehicle and each preset vehicle navigation target, and adjusting the priority of each vehicle navigation target according to the driving task and the traffic information; The currently planned path is adjusted based on the adjusted priorities of the navigation targets of each vehicle and the pre-loaded traffic map to obtain an updated path, and navigation control of the current vehicle is performed based on the updated path.

[0091] It is worth noting that in the above-mentioned embodiment of the multi-objective collaborative path planning and scheduling method system for an autonomous driving vehicle, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of the present application.

[0092] Note that the above are only preferred embodiments of the present application and the technical principles used. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made to the present application without departing from the protection scope of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle, characterized in that: The method comprises: Obtaining environmental status information within the communication range of the current vehicle and traffic information of the current planned path, inputting the environmental status information into the trained agent model to obtain execution action information of the current vehicle, and controlling the operation of the current vehicle according to the execution action information, wherein the environmental status information includes driving information sent by other vehicles and detected road surface information; When it is detected that the traffic information satisfies a preset path update condition, obtaining the driving task of the current vehicle and each preset vehicle navigation target, and adjusting the priority of each vehicle navigation target according to the driving task and the traffic information; The currently planned path is adjusted based on the adjusted priorities of the navigation targets of each vehicle and the pre-loaded traffic map to obtain an updated path, and navigation control of the current vehicle is performed based on the updated path.

2. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to claim 1, characterized in that: The training process of the agent model includes: A deep neural network is constructed, sample environmental state information is determined as input data of the deep neural network, and action probability distribution information is determined as output data of the deep neural network, the sample environmental state information includes driving information of other vehicles and road surface information, the driving information includes vehicle position, speed and acceleration, the road surface information includes road curvature, road slope and traffic signal parameters, and reinforcement learning training is performed on the deep neural network based on a preset reward mechanism.

3. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to claim 2, characterized in that: The reinforcement learning training of the deep neural network based on a preset reward mechanism includes: When the sample environment state information is input, randomly select an action in the action probability distribution information output by the deep neural network, and input the environment state information and the action into a reward function to obtain a reward result; The parameters of the deep neural network are updated and adjusted according to the preset deep Q network algorithm, and the training is repeated to obtain the intelligent agent model. The update formula of the deep Q network algorithm is: , is the learning rate, is the discount factor, is the state-action value function, which means that in the state Take action expected long-term rewards.

4. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to any one of claims 1 to 3, characterized in that: The vehicle navigation objectives include driving time and energy reserves, and the adjusting the priorities of the vehicle navigation objectives according to the driving task and the traffic information includes: The corresponding weights are calculated according to the preset weight calculation formulas of each of the vehicle navigation targets. The weight calculation formula of the driving time is: , the energy reserve weight calculation formula is: , represents the traffic congestion index, Indicates the urgency of the task. Represents the vehicle energy value, Indicates the energy-saving area marking value; The traffic congestion index and the task urgency are calculated according to the driving task and the traffic information, and the vehicle energy value of the current vehicle and the energy-saving area marking value in the traffic information are obtained, and the traffic congestion index, the task urgency, the vehicle energy value and the energy-saving area marking value are respectively input into the corresponding preset weight calculation formula to obtain the weight of each of the vehicle navigation targets, and the priority is adjusted according to the weight of each of the vehicle navigation targets.

5. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to claim 4, characterized in that: The traffic information includes vehicle density and vehicle speed at the congested location, the driving task includes a task type, and the calculating of the traffic congestion index and the task urgency according to the driving task and the traffic information includes: Substituting the vehicle density and vehicle speed of the congested location in the traffic information into a preset congestion calculation formula to obtain a traffic congestion index value, the preset congestion calculation formula includes , represents the vehicle density at the congested location, Indicates the vehicle speed at the congested location; According to the task type in the driving task, a preset task urgency comparison table is queried to obtain a task urgency value.

6. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to any one of claims 1 to 3, characterized in that: Before adjusting the current planned path to obtain an updated path based on the adjusted priorities of the vehicle navigation targets and the preloaded traffic map, the method further includes: Constructing the traffic map according to the traffic information, the traffic map comprising a plurality of traffic nodes and edges of connection relationships between the traffic nodes, the traffic nodes comprising road network nodes, vehicle nodes and obstacle nodes, and the edges comprising weight information; A graph convolutional layer is used to continuously update and aggregate the features of each traffic node to obtain a new feature representation of each traffic node, which is used to capture the spatial relationship and semantic information between the elements.

7. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to any one of claims 1 to 3, characterized in that: The adjusting the current planned path based on the adjusted priorities of each of the vehicle navigation targets and the pre-loaded traffic map to obtain an updated path includes: Acquire the current position and the target position of the current vehicle, determine a plurality of optional paths in the traffic map according to the current position and the target position, and calculate reference values ​​corresponding to each vehicle navigation target of each optional path based on characteristic information in the traffic map; An updated path is determined according to reference values ​​corresponding to each vehicle navigation target of each optional path, priorities of each vehicle navigation target, and a preset path planning method.

8. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to claim 7, characterized in that: The updating path is determined according to the reference value corresponding to each vehicle navigation target of each optional path, the priority of each vehicle navigation target and the preset path planning method, including: The reference values ​​corresponding to the vehicle navigation targets of the optional paths and the priorities of the vehicle navigation targets are input into the trained optimal path output model to obtain an updated path.

9. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to claim 7, characterized in that: The updating path is determined according to the reference value corresponding to each vehicle navigation target of each optional path, the priority of each vehicle navigation target and the preset path planning method, including: The reference values ​​corresponding to the vehicle navigation targets of the optional paths are fuzzy processed to obtain corresponding fuzzy sets, and the updated paths are determined based on the fuzzy sets corresponding to the vehicle navigation targets of the optional paths, the priorities of the vehicle navigation targets and preset fuzzy rules.

10. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to any one of claims 1 to 3, characterized in that: Before obtaining the environmental status information within the communication range of the current vehicle and the traffic information of the current planned path, it includes: Each vehicle on the road is identified as a vehicle node, and based on the Internet of Vehicles technology, each vehicle node within a preset communication range is enabled to communicate with each other so as to share their own information.

11. The multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle according to any one of claims 1 to 3, characterized in that: The current vehicle in the navigation control process also includes: Parameters of the strategy network and the value network are continuously updated according to a preset strategy network update formula and a preset value network update formula, wherein the preset strategy network update formula includes: Among them, the It is The policy network parameters of the vehicle, is the objective function, yes - function, It is Car strategy.

12. A multi-objective collaborative path planning and scheduling system for an autonomous driving vehicle, characterized in that: include: An acquisition module is used to obtain the environmental status information within the communication range of the current vehicle and the traffic information of the current planned path; An action determination module, used for inputting the environmental state information into the trained intelligent agent model to obtain the execution action information of the current vehicle; A control module, configured to control the operation of the current vehicle according to the execution action information, wherein the environmental state information includes driving information received from other vehicles and detected road surface information; A priority adjustment module, configured to obtain the driving task of the current vehicle and each preset vehicle navigation target when detecting that the traffic information satisfies a preset path update condition, and adjust the priority of each vehicle navigation target according to the driving task and the traffic information; A path updating module, configured to adjust the currently planned path based on the adjusted priorities of the respective vehicle navigation targets and a pre-loaded traffic map to obtain an updated path; A navigation module is used to perform navigation control of the current vehicle based on the updated path.

13. A multi-objective collaborative path planning and scheduling device for an autonomous driving vehicle, the device comprising: one or more processors; A storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle as described in any one of claims 1-11.

14. A storage medium storing computer executable instructions, which, when executed by a computer processor, are used to execute the multi-objective collaborative path planning and scheduling method for an autonomous driving vehicle as described in any one of claims 1-11.

Images