Knowledge-enhanced reinforcement learning vehicle decision control method and system
By introducing the integration of expert driving strategy distribution constraints and Gaussian mixed probability model in the reinforcement learning strategy network, the reinforcement learning vehicle decision control method is optimized, and the problem of vehicle decision safety in complex dynamic interaction scenarios is solved, and a safer and more reliable Driving strategy updated.
Patent Information
- Application Number
- CN202411156088.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional vehicle decision-making methods are difficult to adapt to all situations in complex dynamic interaction scenarios, rules-based methods are difficult to ensure driving safety, and deep reinforcement learning methods require exploring a large number of actions to maximize cumulative rewards, resulting in driving safety being threatened.
Using knowledge-enhanced reinforcement learning vehicle decision-making control method, by obtaining a bird's-eye view of three consecutive frames of scenes, inputting a preset-built reinforcement learning model, introducing a loss function based on the distribution constraints of expert driving strategies, regularly integrating new expert driving strategies, and generating optimized Control instructions and optimize the policy network through reward functions to ensure the security and reliability of policy updates.
Improve the safety and reliability of decision-making of vehicles in complex dynamic traffic scenarios. By introducing expert driving strategy constraints, the accused vehicles are encouraged to make decisions that are close to expert behavior, ensuring the stability and effectiveness of each step of strategy update, and maximizing the accumulation. award.
Smart Images

Figure CN118928468A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of vehicle driving decision technology, and in particular to a knowledge-enhanced reinforcement learning vehicle decision control method and system. Background Art
[0002] With the development of sensor technology and deep learning, research related to autonomous driving has made significant progress. Behavioral decision-making is one of the basic functions that autonomous vehicles need to achieve. Its basic purpose is to select the optimal driving operation to achieve an efficient, safe and comfortable driving experience while ensuring the safety of passengers and other road users. However, in the face of complex dynamic interactive scenarios, it is necessary to comprehensively consider multiple factors such as the driving behavior of traffic participants in complex dynamic interactive scenarios, road boundary constraints, etc., which puts forward high requirements on the vehicle decision-making control level.
[0003] Traditional decision-making methods are mainly rule-based methods, which divide the vehicle's motion behavior, establish a behavior rule library based on the mission route, traffic environment, traffic regulations, driving rules knowledge base, etc., perform behavior decision logic reasoning for different environmental states, output driving behavior, and receive feedback from the motion planning layer on the control execution status for real-time dynamic adjustment. However, in the face of complex and uncertain environments, rule-based methods are difficult to adapt to all situations.
[0004] The method based on deep reinforcement learning can update the model's own parameters by generating data through the controlled vehicle's continuous interaction with the environment during the learning process, and can better cope with complex and changing driving environments. However, the controlled vehicle needs to explore a large number of actions to maximize the cumulative reward in order to obtain an effective driving strategy, which seriously endangers the driving safety of the controlled vehicle. Summary of the invention
[0005] The main purpose of the present invention is to provide a knowledge-enhanced reinforcement learning vehicle decision control method and system that can improve the ability of vehicles to make safe and reliable decision control instructions in complex dynamic traffic scenarios.
[0006] The technical solution adopted by the present invention is:
[0007] A knowledge-enhanced reinforcement learning vehicle decision control method is provided, comprising the following steps:
[0008] Obtain three consecutive frames of a scene bird's-eye view, which contains vehicle status information, obstacle information, and road topology information;
[0009] The acquired bird's-eye view of the scene is input into a preset reinforcement learning model, and the vehicle acceleration and yaw rate are output through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; wherein a loss function based on the distribution constraint of the expert driving strategy is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and the new expert driving strategy is regularly integrated with the current expert driving strategy through a Gaussian mixture probability model to generate a new expert driving strategy;
[0010] A control command is generated according to the vehicle acceleration and the yaw rate so that the vehicle moves according to the control command.
[0011] Following the above technical solution, the process of generating an expert driving strategy is as follows: collect a small batch of human driving data in advance, and generate an expert driving strategy based on behavior cloning and uncertainty estimation methods.
[0012] Following the above technical solution, the generation process of the new expert driving strategy is as follows: during the operation of the vehicle, excellent cases that meet the preset conditions are collected. When a certain number of excellent cases are reached, a new expert driving strategy is generated based on the behavior cloning and uncertainty estimation method.
[0013] Following the above technical solution, the expert driving strategy distribution and the learning strategy distribution in the reinforcement learning strategy network are constrained by KL divergence, and incorporated into the loss function of the reinforcement learning strategy network to generate a loss function based on the expert driving strategy distribution constraint.
[0014] Following the above technical solution, the method further comprises the steps of:
[0015] Generate a reward function that includes negative rewards for vehicle collisions, road boundary violations, efficiency, and positive rewards for reaching the end point without collision;
[0016] Optimize the learning policy of a reinforcement learning policy network based on a reward function.
[0017] Following the above technical solution, the specific generation process of the expert driving strategy is as follows:
[0018] Manually drive the vehicle online based on the SMARTS simulation platform and collect expert driving data;
[0019] The expert driving data is constructed as a dataset containing N pairs of state-action sequences;
[0020] The mapping relationship from state to action of expert knowledge is fitted based on the behavior cloning method, the Gaussian probability density model is used to evaluate the uncertainty of human behavior, and the maximum likelihood loss function is used to generate expert driving strategies.
[0021] Following the above technical solution, the specific generation process of the new expert driving strategy is as follows:
[0022] During the online operation of the controlled vehicle, the state-action pairs of the expert driving strategy and the controlled vehicle driving strategy are recorded in the temporary storage space, and the action value network is used to score them. If the controlled vehicle successfully completes the task and its cumulative score exceeds that of the expert, the temporarily stored expert state-action pairs are stored in the expert data expansion storage space;
[0023] When the amount of data stored in the expert data expansion storage space exceeds a set threshold, a new expert driving strategy is generated based on behavior cloning and uncertainty estimation methods.
[0024] The present invention also provides a knowledge-enhanced reinforcement learning vehicle decision control system, comprising the following steps:
[0025] A data acquisition module is used to acquire three consecutive frames of a scene bird's-eye view, which contains vehicle status information, obstacle information, and road topology information;
[0026] A reinforcement learning model module is used to input the acquired scene bird's-eye view into a preset reinforcement learning model, and output the vehicle acceleration and yaw rate through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; wherein a loss function based on the expert driving strategy distribution constraint is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and the new expert driving strategy is regularly integrated with the current expert driving strategy through a Gaussian mixture probability model to generate a new expert driving strategy;
[0027] The control instruction generation module is used to generate a control instruction according to the vehicle acceleration and yaw angular velocity so that the vehicle travels according to the control instruction.
[0028] Following the above technical solution, the expert driving strategy distribution and the learning strategy distribution in the reinforcement learning strategy network are constrained by KL divergence, and incorporated into the loss function of the reinforcement learning strategy network to generate a loss function based on the expert driving strategy distribution constraint.
[0029] The present invention also provides a computer storage medium, which stores a computer program that can be executed by a processor, and the computer program executes the knowledge-enhanced reinforcement learning vehicle decision-making control method described in the above technical solution.
[0030] The beneficial effects of the present invention are as follows: the present invention introduces a loss function based on the distribution constraint of the expert driving strategy into the reinforcement learning strategy network, thereby encouraging the controlled vehicle to make decision-making control instructions close to the expert behavior, thereby improving the driving safety of the vehicle; and generates an optimized expert driving strategy by regularly integrating the new expert driving strategy with the current expert driving strategy through a Gaussian mixture probability model, thereby realizing the self-evolution of the expert strategy.
[0031] Furthermore, by constraining the action distribution between the learning strategy of the model and the expert driving strategy, it is ensured that each step of the strategy update will not deviate too far from the expert driving strategy, thereby ensuring the safety and reliability of the training process.
[0032] Furthermore, the learning strategy of the reinforcement learning policy network is optimized based on the reward function, which encourages the policy network to learn the decision-making control strategy that maximizes the cumulative reward function.
[0033] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0035] Figure 1 is a flow chart of a knowledge-enhanced reinforcement learning vehicle decision control method according to an embodiment of the present invention;
[0036] Figure 2 is a schematic diagram of the structure of a reinforcement learning model based on knowledge enhancement in an embodiment of the present invention;
[0037] Figure 3 is a specific network structure diagram in the reinforcement learning model of an embodiment of the present invention;
[0038] Figure 4 It is a training simulation environment created based on the SMARTS environment in an embodiment of the present invention, wherein the white vehicle is a surrounding vehicle and the red vehicle is a controlled vehicle;
[0039] Figure 5 This is a bird's-eye view of the scene of an embodiment of the present invention, where the yellow solid line is the road boundary line and the white dotted line is the road center line. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0041] It should be noted that the illustrations provided in the embodiments of the present invention are only used to illustrate the basic concept of the present invention in a schematic manner. Therefore, the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout type may also be more complicated.
[0042] In the present invention, it is also necessary to explain that, if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, if the terms "first" and "second" appear, they are only used for description and distinction purposes, and cannot be understood as indicating or implying relative importance.
[0043] like Figure 1 As shown, the knowledge-enhanced reinforcement learning vehicle decision control method includes the following steps:
[0044] S1, obtaining three consecutive frames of a scene bird's-eye view, which includes vehicle state information, obstacle information, and road topology information;
[0045] S2. Input the acquired bird's-eye view of the scene into a preset reinforcement learning model, and output the vehicle acceleration and yaw rate through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; wherein a loss function based on the distribution constraint of the expert driving strategy is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and regularly integrate the new expert driving strategy with the current expert driving strategy through a Gaussian mixture probability model to generate an optimized expert driving strategy;
[0046] S3. Generate a control instruction according to the vehicle acceleration and yaw angular velocity, so that the vehicle travels according to the control instruction.
[0047] Among them, Figure 3As shown in the figure, we can first construct a state space based on the bird's-eye view of the historical three-frame scene (as the input of the reinforcement learning model), use a 4-layer convolutional neural network as the SAC (Soft Actor-Critic) state encoding network to extract the spatial features of the scene in multiple time dimensions; build an action space including vehicle acceleration and yaw velocity (as the output of the reinforcement learning model); based on the multi-layer perceptron network, establish a reinforcement learning policy network, a state value network, and an action value network, so that the controlled vehicle can control the acceleration and yaw velocity based on the control instructions output by the policy network, and according to the feedback of the reward function, use the state value network and the action value network to further guide the policy network to optimize and update its control decision strategy.
[0048] Furthermore, by collecting small batches of human driving data, expert driving strategies can be generated based on behavior cloning and uncertainty estimation methods. Based on the expert driving strategy, a loss function based on the distribution constraint of the expert driving strategy is introduced into the reinforcement learning strategy network to encourage the controlled vehicle to make decision-making control instructions close to the expert behavior.
[0049] Furthermore, since small batches of expert data are difficult to cover complex interactive scenarios, "excellent cases" of the controlled vehicles during operation can be collected (the vehicles arrive at the target destination without collision and efficiently). When the number of "excellent cases" reaches a certain threshold, the behavioral cloning and uncertainty estimation methods are used again to generate new expert behavior decision-making strategies; based on the new and current expert driving strategies, the Gaussian mixture probability model is used to integrate them to generate an optimized expert driving strategy, which can better guide the controlled vehicles to make safe and efficient decision-making control instructions.
[0050] The steps of building a reinforcement learning model mainly include:
[0051] (1) Construct a state space containing a bird's-eye view of the historical three-frame scene, an action space containing vehicle acceleration and yaw angular velocity, and a reward function including negative rewards for vehicle collision, road boundary violation, and efficiency, and positive rewards for reaching the end point without collision. The reward function is as follows:
[0052]
[0053] In the formula, r v is the speeding penalty, v is the speed of the vehicle being charged, and v limit is the road speed limit, r collision = -1 for collision or exceeding the road boundary penalty, r success =1 The reward is provided only when the controlled vehicle reaches the end point without any collision with other traffic participants or exceeding the road boundary.
[0054] (2) A state encoding network consisting of a four-layer convolutional neural network, an action encoding network based on the ReLU activation function, and a reinforcement learning strategy network, a state value network, and an action value network based on a multi-layer perceptron network are established. This allows the model to output vehicle action information based on state space and action space inputs, and to update and optimize its control decision strategy based on the reward function to maximize the cumulative reward function.
[0055] The specific method for generating an expert driving strategy includes the following steps:
[0056] Manually drive the vehicle online based on the SMARTS simulation platform and collect driving data;
[0057] The expert data is transformed into a sequence D containing M pairs of state-action E ={s1,a1,...,s m ,a m};
[0058] Based on the behavior cloning method, the mapping relationship from state to action of expert knowledge is fitted, the Gaussian probability model is used to evaluate the uncertainty of human behavior, and the maximum likelihood loss function is used to generate expert driving strategies. The loss function is as follows:
[0059]
[0060] In the formula, θ represents the parameters of the policy network, and They represent the mean and variance of Gaussian distribution respectively, and c is a constant.
[0061] like Figure 2 As shown in FIG, the generation of the loss function of the expert driving strategy distribution constraint specifically includes the following steps:
[0062] Based on the expert driving strategy, a loss function based on the expert driving strategy distribution constraint is introduced into the reinforcement learning strategy network to encourage the controlled vehicle to make decision-making control instructions close to the expert behavior:
[0063]
[0064] In the formula, Denote the network parameter as ω j The action value network, is the KL divergence constraint (Kullback-Leibler Divergence), To constrain the threshold, by limiting the KL divergence between the learning strategy and the expert driving strategy, we ensure that each step of the strategy update will not deviate too far from the expert driving strategy, thereby ensuring the stability and effectiveness of the training.
[0065] In order to more effectively handle the optimization problem with constraints, the present invention uses Lagrangian duality to transform the constraints of the original problem into an optimization problem:
[0066]
[0067] In the formula, the Lagrange multiplier α is updated only when the constraints are not met and is always non-negative. Its loss function is as follows:
[0068]
[0069] Furthermore, the generation of a new expert driving strategy specifically includes the following steps:
[0070] During the online operation of the controlled vehicle, the state-action pairs of the expert driving strategy and the controlled vehicle driving strategy are recorded in the temporary storage space, and the action value network is used to score them. If the controlled vehicle successfully completes the task and its cumulative score exceeds that of the expert, the temporarily stored expert state-action pairs are stored in the expert data expansion storage space:
[0071]
[0072] When the data stored in the expert data expansion storage space exceeds the set threshold, a new expert driving strategy is generated through training;
[0073] In order to enable the existing expert driving strategy to surpass its original capability boundary, the present invention uses a Gaussian mixture probability model to integrate the existing and new expert driving strategy networks to achieve the self-evolution of the expert driving strategy:
[0074]
[0075] Updating the loss function according to the new expert driving strategy can enable the controlled vehicle to quickly move closer to the integrated expert driving strategy and improve the safety and reliability of decision-making control instructions.
[0076] Based on the SMARTS simulation platform, the initial position of the vehicle is randomly generated at the beginning of each training set. Based on the initial speed, the surrounding vehicles are allowed to travel at a speed disturbance in the range of [-1,2] m / s. In addition, the random driving style coefficient is used to determine whether to change lanes, give way or not. Figure 4 As shown, an unprotected left turn scenario is shown, where the controlled vehicle must turn left into a main road without the control of traffic lights, cross a two-way four-lane road, and reach the rightmost lane.
[0077] The knowledge-enhanced reinforcement learning vehicle decision control system of the embodiment of the present invention is mainly used to implement the above method embodiment, and the system includes:
[0078] A data acquisition module is used to acquire three consecutive frames of a scene bird's-eye view, which contains vehicle status information, obstacle information, and road topology information;
[0079] A reinforcement learning model module is used to input the acquired scene bird's-eye view into a preset reinforcement learning model, and output the vehicle acceleration and yaw rate through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; wherein a loss function based on the distribution constraint of the expert driving strategy is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and the new expert driving strategy is regularly integrated with the current expert driving strategy through a Gaussian mixture probability model to generate an optimized expert driving strategy;
[0080] The control instruction generation module is used to generate a control instruction according to the vehicle acceleration and yaw angular velocity so that the vehicle travels according to the control instruction.
[0081] Each module is mainly used to implement each step of the above method embodiment, which will not be described in detail here.
[0082] The method of the above embodiment can be applied to a server, a terminal, or a system including a terminal, a server, and a system, and implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here.
[0083] The present application also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App application store, etc., on which a computer program is stored, and when the program is executed by a processor, a corresponding function is realized. When the computer-readable storage medium of this embodiment is executed by a processor, the knowledge-enhanced reinforcement learning vehicle decision control method of the method embodiment is realized.
[0084] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.
[0085] The order of execution of each step in the above embodiment does not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.
[0086] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.
Claims
1. A knowledge-enhanced reinforcement learning vehicle decision control method, characterized in that: The following steps are involved: Obtain three consecutive frames of a scene bird's-eye view, which contains vehicle status information, obstacle information, and road topology information; Input the acquired bird's-eye view of the scene into a preset reinforcement learning model, and output the vehicle acceleration and yaw rate through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; A loss function based on the expert driving strategy distribution constraint is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and the new expert driving strategy is regularly integrated with the current expert driving strategy through a Gaussian mixture probability model to generate an optimized expert driving strategy; A control command is generated according to the vehicle acceleration and the yaw rate so that the vehicle moves according to the control command.
2. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 1 is characterized in that: The process of generating expert driving strategies is as follows: collect small batches of human driving data in advance, and generate expert driving strategies based on behavior cloning and uncertainty estimation methods.
3. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 1, characterized in that: The generation process of the new expert driving strategy is as follows: during the operation of the vehicle, excellent cases that meet the preset conditions are collected. When a certain number of excellent cases are reached, a new expert driving strategy is generated based on the behavior cloning and uncertainty estimation methods.
4. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 1, characterized in that: Specifically, the expert driving strategy distribution and the learning strategy distribution in the reinforcement learning strategy network are constrained by KL divergence, and incorporated into the loss function of the reinforcement learning strategy network to generate a loss function based on the expert driving strategy distribution constraint.
5. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 1, characterized in that: The method further comprises the steps of: Generate a reward function that includes negative rewards for vehicle collisions, road boundary violations, efficiency, and positive rewards for reaching the end point without collision; Optimize the learning policy of a reinforcement learning policy network based on a reward function.
6. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 2, characterized in that: The specific generation process of the expert driving strategy is as follows: Manually drive the vehicle online based on the SMARTS simulation platform and collect expert driving data; The expert driving data is constructed as a dataset containing N pairs of state-action sequences; The mapping relationship from state to action of expert knowledge is fitted based on the behavior cloning method, Gaussian probability density is used to evaluate the uncertainty of human behavior, and the maximum likelihood loss function is used to generate expert driving strategies.
7. The knowledge-enhanced reinforcement learning vehicle decision control method according to claim 3 is characterized in that: The specific generation process of the new expert driving strategy is as follows: During the online operation of the controlled vehicle, the state-action pairs of the expert driving strategy and the controlled vehicle driving strategy are recorded in the temporary storage space, and the action value network is used to score them. If the controlled vehicle successfully completes the task and its cumulative score exceeds that of the expert, the temporarily stored expert state-action pairs are stored in the expert data expansion storage space; When the amount of data stored in the expert data expansion storage space exceeds a set threshold, a new expert driving strategy is generated based on behavior cloning and uncertainty estimation methods.
8. A knowledge-enhanced reinforcement learning vehicle decision control system, characterized in that: include: A data acquisition module is used to acquire three consecutive frames of a scene bird's-eye view, which contains vehicle status information, obstacle information, and road topology information; A reinforcement learning model module, which is used to input the acquired bird's-eye view of the scene into a preset reinforcement learning model, and output the vehicle acceleration and yaw rate through the reinforcement learning model; the reinforcement learning model includes a reinforcement learning strategy network, a state value network, and an action value network; A loss function based on the expert driving strategy distribution constraint is introduced into the reinforcement learning strategy network to make the output of the reinforcement learning model closer to the expert driving strategy; and the new expert driving strategy is regularly integrated with the current expert driving strategy through a Gaussian mixture probability model to generate an optimized expert driving strategy; The control instruction generation module is used to generate a control instruction according to the vehicle acceleration and yaw angular velocity so that the vehicle travels according to the control instruction.
9. The knowledge-enhanced reinforcement learning vehicle decision control system according to claim 8, characterized in that: Specifically, the expert driving strategy distribution and the learning strategy distribution in the reinforcement learning strategy network are constrained by KL divergence, and incorporated into the loss function of the reinforcement learning strategy network to generate a loss function based on the expert driving strategy distribution constraint.
10. A computer storage medium, characterized in that: A computer program executable by a processor is stored therein, and the computer program executes the knowledge-enhanced reinforcement learning vehicle decision control method described in any one of claims 1 to 7.
Citation Information
Cited By
Vehicle risk avoiding planning method and device, vehicle, electronic equipment and storage medium
CN119408570A
Vehicle risk avoidance planning method, device, vehicle, electronic device and storage medium
CN119408570B
Priori knowledge fused end-to-end automatic driving decision-making method
CN119975417A
An end-to-end autonomous driving decision-making method integrating prior knowledge
CN119975417B
