A driving assistance method and device for longitudinal following scenarios

Abstract

This invention discloses an assisted driving method and device for longitudinal following scenarios, relating to the field of vehicle assisted driving technology. The assisted driving method includes: dynamically fusing a rule model and a data-driven model based on the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario, to construct a dynamic fusion control law for the vehicle specifically for longitudinal following scenarios. The rule model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on a deep deterministic policy gradient model. The dynamic fusion control law is used to generate the following acceleration of the vehicle relative to the vehicle in front. The vehicle is adjusted according to the following acceleration to improve the driving experience and traffic efficiency.

Description

Technical Field

[0001] This invention relates to the field of vehicle driver assistance technology, and in particular to a driver assistance method and device for longitudinal following scenarios. Background Technology

[0002] Advanced Driver Assistance Systems (ADAS) are technologies that enhance driving safety and comfort by utilizing high-performance sensors (such as cameras, radar, and ultrasonic sensors), high-performance computing chips, high-precision maps, and advanced algorithms. Among these, longitudinal following control, a key technology of ADAS, improves the accuracy of controlling the vehicle's speed and the distance between the vehicle and the vehicle in front, ensuring safe following of the vehicle ahead, reducing driver workload, and enhancing driving safety and comfort.

[0003] Existing longitudinal following control technologies for ADAS mainly use a single model as the control law for longitudinal following. However, the single model has certain limitations, which results in poor control accuracy during longitudinal following control. In particular, existing longitudinal following control technologies are poorly adapted to low-speed congestion scenarios, leading to less than ideal driving experience and vehicle traffic efficiency. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an assisted driving method and device for longitudinal following scenarios. The method can dynamically construct a dynamic fusion control law for the longitudinal following scenario in which the vehicle is located, so that the dynamic fusion control law can be better matched with the longitudinal following scenario in which the vehicle is located, thereby generating following acceleration for the vehicle more accurately, so as to more accurately control the vehicle, and enabling the assisted driving method to adapt to different scenarios, so as to improve the driving experience and traffic efficiency of the vehicle.

[0005] To achieve the above objectives, in a first aspect, according to embodiments of the present invention, an assisted driving method for longitudinal following scenarios is provided, comprising: Based on the current distance between this vehicle and the vehicle in front, this vehicle's driving data, and this vehicle's driving scenario, a rule-based model and a data-driven model are dynamically fused to construct a dynamic fusion control law for this vehicle in longitudinal following scenarios. The rule-based model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on a deep deterministic policy gradient model; Using the aforementioned dynamic fusion control law, a following acceleration relative to the preceding vehicle is generated for this vehicle; Adjust the vehicle's acceleration according to the stated following vehicle acceleration.

[0006] Optionally, the assisted driving method further includes: A first mapping relationship is constructed for the rule model, wherein, in the first mapping relationship, a combination of a vehicle spacing and a vehicle speed corresponds to a weight coefficient of the rule model; Construct a second mapping relationship between at least one driving scenario and the weight coefficients corresponding to the rule model; The dynamic fusion of rule-based models and data-driven models includes: Based on the first mapping relationship and the second mapping relationship, a target weight coefficient is determined for the rule model, wherein the target weight coefficient satisfies the current vehicle spacing and the vehicle's driving data and / or the target weight coefficient satisfies the vehicle's driving scenario; The target weight coefficients are used to fuse the rule model and the data-driven model.

[0007] Optionally, determining the target weight coefficients for the rule model includes: From the first mapping relationship, find the first weighting coefficient corresponding to the current vehicle speed included in the current vehicle distance and the vehicle's driving data; From the second mapping relationship, find the second weight coefficient corresponding to the driving scenario of the vehicle; If the first weighting coefficient and the second weighting coefficient are inconsistent, and the driving scenario of the vehicle is a low-speed driving scenario, the larger of the first weighting coefficient and the second weighting coefficient is determined to be the target weighting coefficient; if the driving scenario of the vehicle is a medium-high speed driving scenario, the smaller of the first weighting coefficient and the second weighting coefficient is determined to be the target weighting coefficient.

[0008] Optionally, the assisted driving method further includes: Based on the training data, including vehicle status information, preceding vehicle information, and surrounding environment information, for H ∞ Robust control determines the state variable matrix and state space parameters; The state variable matrix and the state space parameters are compared with H. ∞ Robust control is combined to construct a rule model.

[0009] Optionally, for H ∞ Robust control determines the state variable matrix, which includes: Based on the vehicle status information and the preceding vehicle information, determine the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment cycle, and the expected acceleration of the current adjustment cycle. Using the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment period, and the expected acceleration of the current adjustment period, the following state variable matrix corresponding to the current adjustment period is constructed. :

[0010] in, ; ; ; Indicates the following distance error of the vehicle; Indicates the actual following distance of the vehicle; Indicates the expected following distance of the vehicle; This indicates the longitudinal speed error of the vehicle; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; T Indicates the time constant of the sampling step; This indicates the vehicle's actual acceleration; Indicates the vehicle's expected acceleration; This represents the time constant of the chassis system; Indicates the first One adjustment cycle.

[0011] Optionally, the desired following distance of the vehicle is determined, including: The following constant interval model is used to calculate the expected following distance of the vehicle; Constant workshop time-distance model:

[0012] in, This indicates the minimum static inter-vehicle distance set for the vehicle. This indicates the upper limit of the longitudinal following distance set for the vehicle; This indicates the lower limit of the vehicle's set longitudinal following distance. Indicates the vehicle's actual speed; This indicates the relative speed between the two workshops; The actual acceleration of the vehicle in front; Indicates the mass of the vehicle; , and All are parameters, and satisfy the following conditions: , , .

[0013] Optionally, the state space parameters include: ; , , ; in, Indicates the distance between the front of one vehicle and the rear of the vehicle in front; This represents the time constant of the chassis system; The construction rule model includes: Based on H ∞ Robust control is achieved by constructing the following linear inequalities for the state space parameters and calculating the state feedback control gain matrix from these linear inequalities.

[0014] in, Indicates based on H ∞ The robust performance condition presupposes a positive definite symmetric matrix that satisfies ; Represents the state feedback control gain matrix; = ; Indicates based on H ∞ Robust performance conditions are preset positive scalars; Using the state feedback control gain matrix and the state variable matrix, the following rule model is constructed:

[0015] in, This represents the expected acceleration of the vehicle as output by the rule-based model. Represents the state feedback control gain matrix; This represents the state variable matrix.

[0016] Optionally, the assisted driving method further includes: Based on the vehicle state information, preceding vehicle information, and surrounding environment information included in the training data, the state space set and action space set are determined for the deep deterministic policy gradient model. Using the state space set and the action space set, a deep deterministic policy gradient model is trained to obtain the data-driven model.

[0017] Optionally, the state space set includes: the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the vehicle in front's actual speed, the vehicle's actual acceleration, and the vehicle in front's actual acceleration. The set of motion spaces includes the vehicle's desired acceleration.

[0018] Optionally, the deep deterministic policy gradient model includes a policy network and a value network. The policy network is used to generate policies and output actions. The value network is a fully connected network. The hidden layers of the fully connected network use modified linear units as activation functions, and the hyperbolic tangent function is used as the activation function for the final output. The value network is used to fit the state-action value function and evaluate the actions output by the policy network. The value network includes two input layers, which are respectively input to the state space set and the action space set.

[0019] Optionally, the assisted driving method further includes: constructing a reward function for the deep deterministic policy gradient model with the following: , in, Indicates the total reward; Indicates the weighting coefficient of the security reward; Indicates the weighting coefficient of efficiency rewards; The weighting coefficient representing the comfort reward; Indicates a safety reward; Indicates a reward for efficiency; This indicates a comfort reward.

[0020] Optionally, the assisted driving method also includes: The security reward is calculated using the following formula: Formula 1:

[0021] in, Indicates a safety reward; This indicates the preset collision reward safety parameters. This represents the preset compensation factor. ; Indicates the expected following distance of the vehicle; This indicates the actual following distance of the vehicle.

[0022] Optionally, the assisted driving method also includes: The efficiency bonus is calculated using the following formula (Formula 2); Formula 2:

[0023] in, Indicates a reward for efficiency; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; This represents the preset compensation factor. .

[0024] Optionally, the assisted driving method also includes: The comfort reward is calculated using the following formula three; Formula 3: Indicates a comfort reward; jerk Indicates judder; This indicates the vehicle's actual acceleration; This represents the preset compensation factor. .

[0025] Optionally, the assisted driving method further includes: determining a current speed limit for the vehicle; If the actual current vehicle speed, as included in the vehicle's driving data, is greater than the current speed limit, the vehicle will be controlled to output deceleration. If the current actual vehicle speed is less than the current speed limit, the vehicle is adjusted according to the following acceleration.

[0026] Optionally, determining the current speed limit for the vehicle includes: The current speed limit is determined by the smallest of the following: road speed limit, curve speed limit, visibility speed limit, and the vehicle's set maximum cruise speed.

[0027] Optionally, the road speed limit is calculated using the following formula:

[0028] in, Indicates the road speed limit; This indicates the speed limit indicated by the speed limit sign; Indicates the maximum speed limit corresponding to the road's classification level; This indicates that there are speed limit signs on the road.

[0029] Optionally, the speed limit for the curve is calculated using the following formula: Indicates the speed limit for the curve; Indicates the radius of curvature of the road. , and These represent the parameter settings.

[0030] Optionally, the visibility speed limit is calculated using the following formula: Indicates visibility speed limit; This indicates the visible distance of the vehicle's onboard sensor system. Indicates reaction time.

[0031] Secondly, embodiments of the present invention provide an assisted driving device for longitudinal following scenarios, comprising: a model building module, a generation module, and an auxiliary control module, wherein, The model building module is used to dynamically fuse a rule-based model and a data-driven model based on the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario. This results in the construction of a dynamic fusion control law for the vehicle in a longitudinal following scenario. The rule-based model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on a deep deterministic policy gradient model; The generation module is used to generate a following acceleration relative to the preceding vehicle for the vehicle using the dynamic fusion control law; The auxiliary control module is used to adjust the vehicle according to the following acceleration.

[0032] Thirdly, embodiments of the present invention provide an electronic device for assisted driving in longitudinal following scenarios, the electronic device comprising: One or more processors; 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 assisted driving method for longitudinal following scenarios provided by the first aspect embodiments and related embodiments described above.

[0033] Fourthly, embodiments of the present invention provide a vehicle, characterized in that it includes the driver assistance device for longitudinal following scenarios provided in the second aspect embodiment above or the electronic device for driver assistance for longitudinal following scenarios provided in the third aspect embodiment above.

[0034] Fifthly, embodiments of the present invention provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the assisted driving method for longitudinal following scenarios as described in the above embodiments.

[0035] One embodiment of the above invention has the following advantages or beneficial effects: In the process of constructing a dynamic fusion control law for the vehicle in a longitudinal following scenario, the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario are used as references to dynamically fuse the rule model and the data-driven model. This can make up for the defects of a single model, so that the constructed dynamic fusion control law can better match the environment in which the vehicle is located, so as to accurately control the vehicle to follow the vehicle in front, improve the reliability of the vehicle's longitudinal following, and improve the user's driving experience and vehicle traffic efficiency. In particular, the technical solution provided by the embodiment of the present invention can adapt well to low-speed congestion scenarios and improve vehicle traffic efficiency in low-speed congestion scenarios.

[0036] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0037] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a schematic diagram of the main process of an assisted driving method for a longitudinal following scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a first scenario for which the assisted driving method provided according to an embodiment of the present invention is applied; Figure 3 This is a schematic diagram of a second scenario for the assisted driving method provided according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a fuzzy regular surface corresponding to the weight coefficients of a regular model according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the architecture of an Actor network according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the architecture of a Critic network according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the system architecture upon which the assisted driving method according to an embodiment of the present invention depends; Figure 8 This is a schematic diagram of the main process flow of the system architecture implementation upon which the assisted driving method according to an embodiment of the present invention depends; Figure 9 This is a schematic diagram of the main modules of the driver assistance device according to an embodiment of the present invention; Figure 10 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 11 This is a schematic diagram of the structure of a computer system suitable for implementing embodiments of the present invention. Detailed Implementation

[0038] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0039] It should be noted that, unless otherwise specified, the embodiments of the present invention and the technical features thereof can be combined with each other.

[0040] The longitudinal following scenario involved in this invention mainly refers to a vehicle following the vehicle in front of it (hereinafter referred to as the "leading vehicle"). Specifically, in the scenario where the vehicle remains in its current lane, the leading vehicle generally refers to a vehicle in the same lane as the vehicle and located in front of it; in the scenario where the vehicle changes lanes, the leading vehicle generally refers to a vehicle in the same lane as the vehicle and located in front of it, as well as a vehicle traveling in the lane the vehicle is about to change to and located in front of it. It should be noted that the vehicle and the leading vehicle are adjacent vehicles, and there is no obstruction between them.

[0041] The vehicle referred to in the embodiments of the present invention generally refers to a vehicle that applies the assisted driving method for longitudinal following scenarios provided in the embodiments of the present invention or is equipped with the assisted driving device or system for longitudinal following scenarios provided in the embodiments of the present invention.

[0042] The vehicle involved in this embodiment of the invention can refer to any vehicle traveling in the lane. It can be the vehicle itself, the vehicle in front of it, or other vehicles unrelated to the vehicle itself. This vehicle mainly provides training data for training models of assisted driving methods, devices, or systems for longitudinal following scenarios. It can be understood that this vehicle generally provides the training data required for training the rule model and data-driven model during the construction or updating of the rule model and data-driven model.

[0043] The dynamic fusion control law involved in the embodiments of the present invention generally refers to an algorithm that is constructed in real time for the vehicle to control the vehicle to follow the vehicle in front, based on the environment and scenario in which the vehicle is located, and outputs control commands to the vehicle control layer to control the vehicle to follow the vehicle in front.

[0044] Specifically, such as Figure 1 As shown, an embodiment of the present invention provides an assisted driving method for longitudinal following scenarios, which may include the following steps: Step S101: Based on the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario, dynamically fuse the rule model and the data-driven model to construct a dynamic fusion control law for the vehicle in the longitudinal following scenario. The rule model is based on H... ∞ Robust control was constructed, and the data-driven model was constructed based on the Deep Deterministic Policy Gradient (DDPG) model (hereinafter referred to as the DDPG model).

[0045] The dynamic fusion of rule models and data-driven models involved in the embodiments of the present invention generally refers to the real-time and variable fusion of rule models and data-driven models. That is, when any one or more of the current vehicle spacing, the vehicle's driving data, and the vehicle's driving scenario change, the fusion of rule models and data-driven models will also change accordingly, thereby causing the dynamic fusion control law to change accordingly.

[0046] Among them, the application scenarios of the assisted driving method for longitudinal following scenarios provided in the embodiments of the present invention are as follows: Figure 2 and Figure 3 As shown. In Figure 2 and Figure 3 The diagram shows V1 representing this vehicle, V2 representing the vehicle in front, and V3 representing other vehicles traveling in the lane. Figure 2 The following scenario is shown: This vehicle V1 follows the vehicle V2 traveling in the same lane. Figure 3 The following scenario is as follows: following the lane change path K1, the vehicle changes lanes and follows the vehicle in front V2. Before changing lanes, the vehicle in front of vehicle V2 is the vehicle in front of vehicle V1 in its current lane. After vehicle V1 changes lanes to the adjacent lane, the vehicle in front of vehicle V1 and adjacent to vehicle V1 is in the adjacent lane.

[0047] The current distance between this vehicle and the vehicle in front generally refers to the distance from the front of this vehicle to the rear of the vehicle in front. For example, in the case where this vehicle and the vehicle in front are in the same lane, such as... Figure 2 In the scenario shown, the distance L1 between the front of this car and the rear of the car in front is the length of the line connecting the front of this car and the rear of the car in front. For example... Figure 3 In the scenario shown, the distance L1 between the front of this vehicle and the rear of the vehicle in front can also be the distance between the extension line of the front of this vehicle V1 in the lane width direction and the extension line of the rear of the vehicle in front V2 in the lane width direction.

[0048] The driving data of the vehicle involved in the embodiments of the present invention generally comes from the vehicle's sensing system, such as the vehicle's actual speed and actual acceleration.

[0049] The driving scenarios of the vehicle involved in this invention embodiment may include low-speed driving scenarios and medium-to-high-speed driving scenarios, which are mainly determined based on the environmental information around the vehicle and the vehicle's driving data. For example, if there are few vehicles around the vehicle and the vehicle's speed exceeds a preset speed threshold, the vehicle is determined to be in a medium-to-high-speed driving scenario. Conversely, if the vehicle is in a congested area and its speed is below a preset speed threshold, the vehicle is determined to be in a congested low-speed driving scenario. Generally, in congested low-speed driving scenarios, the reliability requirements for vehicle control are relatively high, so the rule model in the dynamic fusion control law is used as the primary model, and the data-driven model as the auxiliary model. In medium-to-high-speed driving scenarios, driving behavior tends to be more stable and dynamic complexity is higher, so the data-driven model in the dynamic fusion control law is used as the primary model, and the rule model as the auxiliary model, giving the dynamic fusion control law better adaptability and generalization ability. The preset speed threshold is generally pre-configured for the vehicle and can be determined based on the vehicle's historical driving data; the specific value of the preset speed threshold is not limited here.

[0050] Among them, based on H ∞ Robust control, through the construction of a rule-based model, enables vehicles to operate under optimized rules, thereby achieving robustness in vehicle control. The rule-based model construction process incorporates various rules constraining vehicle operation, such as system errors (e.g., longitudinal following distance limits, following distance errors, longitudinal speed errors, acceleration errors, etc.) and uncertainties related to external environmental disturbances. ∞ Robust control has good anti-interference ability and can suppress uncertainties in the system or environment. Based on H ∞ The rule-based model for robust control can control the longitudinal kinematics of a vehicle, thereby improving the robustness of longitudinal following control.

[0051] The DDPG model refers to the Deep Deterministic Policy Gradient (DDPG) algorithm. DDPG is a deep reinforcement learning algorithm that incorporates neural networks from deep learning into reinforcement learning. In this embodiment of the invention, an Actor-Critic structure (AC) is built for the DDPG model during its construction, and the DDPG model is constructed within this AC.

[0052] Step S102: Using the dynamic fusion control law, generate the following acceleration of the vehicle relative to the vehicle in front.

[0053] The dynamic fusion control law in this step can serve as a new model, directly generating the vehicle's following acceleration relative to the vehicle in front. Alternatively, the following acceleration can be obtained by weighted summing of the first acceleration output by the rule-based model and the second acceleration output by the data-driven model. For example, the first acceleration output by the rule-based model is... The second acceleration output by the data-driven model is The weight of the first acceleration is The following acceleration obtained in this step is: .

[0054] Step S103: Adjust the vehicle according to the following vehicle acceleration.

[0055] Understandably, after generating the following acceleration, the driver assistance system converts the following acceleration into a corresponding control signal, and outputs the control signal to the execution layer such as the chassis domain controller, motor controller and powertrain to control the acceleration of the vehicle.

[0056] The technical solution provided by this invention, in the process of constructing a dynamic fusion control law for longitudinal following scenarios, uses the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario as references to dynamically fuse rule models and data-driven models. This can compensate for the deficiencies of a single model, enabling the constructed dynamic fusion control law to better match the vehicle's environment, thereby accurately controlling the vehicle's following of the vehicle in front, improving the reliability of longitudinal following, and enhancing the user's driving experience and vehicle traffic efficiency. In particular, the technical solution provided by this invention can adapt well to low-speed congestion scenarios and improve vehicle traffic efficiency in low-speed congestion scenarios.

[0057] In this embodiment of the invention, to achieve dynamic fusion of the rule model and the data-driven model, the above-mentioned assisted driving method may further include: constructing a first mapping relationship for the rule model, wherein in the first mapping relationship, a combination of a vehicle distance and a vehicle speed is mapped to a weight coefficient of the corresponding rule model; constructing a second mapping relationship between at least one driving scenario and a weight coefficient corresponding to the rule model; based on this, a specific implementation of step S101 may include: determining a target weight coefficient for the rule model according to the first mapping relationship and the second mapping relationship, wherein the target weight coefficient satisfies the current vehicle distance and the vehicle's driving data and / or satisfies the vehicle's driving scenario; and fusing the rule model and the data-driven model using the target weight coefficient. That is, the target weight coefficient may only satisfy the current vehicle distance and the vehicle's driving data, or only satisfy the vehicle's driving scenario, or simultaneously satisfy the current vehicle distance, the vehicle's driving data, and the vehicle's driving scenario. Preferably, the target weight coefficient simultaneously satisfies the current vehicle distance, the vehicle's driving data, and the vehicle's driving scenario.

[0058] Among them, the first mapping relationship can be shown in Table 1 below.

[0059] Table 1

[0060] Among them, α represents the weight coefficient corresponding to the rule model; v ego represents the actual vehicle speed of this vehicle; d represents the actual vehicle distance between this vehicle and the vehicle in front. v ego takes values from {VS, S, M, L, VL}; d takes values from {CD, MD, FD}; the value range of α is {VSK, SK, MK, LK, VLK}.

[0061] That is, the first mapping relationship can be stored through the structure shown in Table 1 above. Among them, d the value of v ego and the value of d the value of CD of v ego and the value of VS of d the value of MD of v ego and the value of M of d the value of FD of v ego and the value of M of

[0062] Among them, for the vehicle distance d , CD represents a defined smaller vehicle distance; MD represents a defined medium vehicle distance; FD represents a defined larger vehicle distance; generally CD < MD < FD, where CD, MD and FD can be assigned a specific value or a value range according to actual needs. For the actual vehicle speed v ego, VS represents a defined relatively low vehicle speed of the host vehicle; S represents a defined low vehicle speed of the host vehicle; M represents a defined medium vehicle speed of the host vehicle; L represents a defined high vehicle speed of the host vehicle; VL represents a defined relatively high vehicle speed of the host vehicle. Generally, VS < S < M < L < VL. Among them, VS, S, M, L, and VL can also be assigned a specific value or a value range according to actual requirements. For the weight coefficient α corresponding to the rule model, VSK represents a defined relatively small weight coefficient; SK represents a defined small weight coefficient; MK represents a defined medium weight coefficient; LK represents a defined large weight coefficient; VLK represents a defined relatively large weight coefficient. Generally, VSK < SK < MK < LK < VLK. Among them, VSK, SK, MK, LK, and VLK can also be assigned a specific value according to actual requirements. Exemplarily, VSK is assigned 0.1, SK is assigned 0.3, MK is assigned 0.5, LK is assigned 0.7, and VLK is assigned 0.9.

[0063] It should be noted that the first mapping relationship shown in Table 1 above is only an example. For those skilled in the art, more mapping dimensions and mappable data can be designed based on the structure given in Table 1.

[0064] In addition, for the first mapping relationship, in addition to the form shown in Table 1 above, it can also be determined through Figure 4 the fuzzy rule surface of the weight coefficient corresponding to the rule model shown, Figure 4 The fuzzy rule surface of the weight coefficient corresponding to the rule model given shows the actual inter-vehicle distance d and the actual vehicle speed of the host vehicle v ego The mapping relationship between and the weight coefficient α corresponding to the rule model. From a combination of an actual inter-vehicle distance d and an actual vehicle speed of the host vehicle v ego a corresponding weight coefficient corresponding to the rule model, that is, the above first weight coefficient, can be uniquely determined in the fuzzy rule surface.

[0065] In addition, the second mapping relationship between at least one driving scenario and the corresponding weight coefficient of the rule model may include: a low-speed driving scenario and its corresponding weight coefficient, and / or a medium-to-high-speed driving scenario and its corresponding weight coefficient, wherein the weight coefficient corresponding to the low-speed driving scenario is smaller than the weight coefficient corresponding to the medium-to-high-speed driving scenario. In low-speed driving scenarios, the weight coefficient is generally larger; for example, a single weight coefficient (e.g., 0.8) can be set, or multiple weight coefficients can be set according to the degree of congestion (the greater the congestion, the larger the weight coefficient). In medium-to-high-speed driving scenarios, a single weight coefficient (e.g., 0.1) can be set, or multiple weight coefficients can be set according to the vehicle speed (the higher the speed, the smaller the corresponding weight coefficient).

[0066] By integrating the first and second mapping relationships to determine the target weight coefficients for the rule model, the dynamic fusion control law constructed by fusing the rule model and the data-driven model can better fit the driving scenario of the vehicle and the driving relationship between the vehicle and the vehicle in front, so that the dynamic fusion control law can control the vehicle more accurately.

[0067] More specifically, a specific implementation plan for determining the target weight coefficient for the rule model may include: finding a first weight coefficient from a first mapping relationship that corresponds to the current vehicle speed and the current vehicle distance and driving data of the current vehicle; finding a second weight coefficient from a second mapping relationship that corresponds to the driving scenario of the current vehicle; if the first weight coefficient and the second weight coefficient are inconsistent, if the driving scenario of the current vehicle is a low-speed driving scenario, determining the larger of the first weight coefficient and the second weight coefficient as the target weight coefficient; if the driving scenario of the current vehicle is a medium-to-high-speed driving scenario, determining the smaller of the first weight coefficient and the second weight coefficient as the target weight coefficient. If the first weight coefficient and the second weight coefficient are consistent, directly determining the first weight coefficient and the second weight coefficient as the target weight coefficient.

[0068] For example, a first weight coefficient of VLK is determined based on a first mapping relationship, and a second weight coefficient of 0.8 is determined based on a low-speed driving scenario. Where VLK is greater than 0.8, the target weight coefficient is determined to be VLK. A first weight coefficient of SK is determined based on a first mapping relationship, and a second weight coefficient of 0.1 is determined based on a medium-to-high-speed driving scenario. Where SK is greater than 0.1, the target weight coefficient is determined to be 0.1.

[0069] Among them, the specific implementation plan for dynamically integrating rule models and data-driven models may include: integrating them using the following calculation formula (1).

[0070] (1)

[0071] in, This represents the dynamic fusion control law; Representational rule model; This represents a data-driven model; This represents the target weight coefficient corresponding to the rule-based model. .

[0072] because It will adjust based on the vehicle's driving scenario and changes in data related to the vehicle's driving, i.e. The dynamic fusion control law will change dynamically, and thus the dynamic fusion control law will also change dynamically according to the driving scenario of the vehicle and the changes in data related to the driving of the vehicle, so as to improve the reliability and accuracy of the dynamic fusion control law.

[0073] The rule model and data-driven model involved in the embodiments of the present invention can be trained using existing training methods, that is, the rule model can be based on H ∞ Robust control is obtained using existing training methods, and data-driven models can be obtained based on DDPG models using existing training methods.

[0074] Furthermore, as can be seen from the above technical solutions, the foundation for constructing the dynamic fusion control law in the embodiments of the present invention is the rule model and the data-driven model. Therefore, the accuracy and reliability of the rule model and the data-driven model will also affect the dynamic fusion control law. The technical solutions provided in the embodiments of the present invention also innovatively design the construction process of the rule model and the data-driven model to obtain rule models and data-driven models with higher accuracy and reliability, thereby further improving the reliability and accuracy of the dynamic fusion control law.

[0075] The following describes the construction process of the rule-based model and the construction process of the data-driven model provided in the embodiments of the present invention.

[0076] Specifically, regarding the construction process of the rule model, the technical solution provided by this embodiment of the invention includes: based on the vehicle state information, preceding vehicle information, and surrounding environment information included in the training data, for H... ∞ Robust control determines the state variable matrix and state space parameters; it then compares the state variable matrix and state space parameters with H... ∞ Robust control is combined to construct a rule-based model. By integrating vehicle state information, preceding vehicle information, and surrounding environment information, the state variable matrix and state space parameters are determined. This allows the state variable matrix and state space parameters to more accurately and realistically reflect the relationship between the vehicle and the preceding vehicle, as well as the scenario in which the vehicle is located. Therefore, the constructed rule-based model is more closely aligned with real-world scenarios. The training data can be historical driving data of various vehicles provided by the vehicle management platform, which includes at least vehicle state information, preceding vehicle information, and surrounding environment information. Alternatively, the training data can be driving data of various vehicles monitored under multiple test scenarios.

[0077] More specifically, for H ∞ A specific implementation scheme for robust control to determine the state variable matrix may include: determining the vehicle's actual following distance (which is the actual distance between the vehicle and the preceding vehicle), the vehicle's desired following distance (which is the ideal distance between the vehicle and the preceding vehicle under control conditions), the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment period, and the desired acceleration of the current adjustment period, based on the vehicle's state information and the preceding vehicle's information; and constructing the following state variable matrix corresponding to the current adjustment period using the vehicle's actual following distance, the vehicle's desired following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment period, and the desired acceleration of the current adjustment period. :

[0078] in, ; ; ; Indicates the following distance error of the vehicle; Indicates the actual following distance of the vehicle; Indicates the expected following distance of the vehicle; This indicates the longitudinal speed error of the vehicle; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; T Indicates the time constant of the sampling step; This indicates the vehicle's actual acceleration; Indicates the vehicle's expected acceleration; This represents the time constant of the chassis system; Indicates the first One adjustment cycle. Among them, the specific implementation plan for determining the expected following distance of the vehicle may include: using the following constant inter-vehicle distance model (i.e., calculation formula (2)) to calculate the expected following distance of the vehicle; Constant workshop time-distance model: (2) in, This indicates the minimum static inter-vehicle distance set for the vehicle. This indicates the upper limit of the longitudinal following distance set for the vehicle; This indicates the lower limit of the vehicle's set longitudinal following distance. Indicates the vehicle's actual speed; This indicates the relative speed between the two workshops; The actual acceleration of the vehicle in front; Indicates the mass of the vehicle; , and All are parameters, and satisfy the following conditions: , , .

[0079] In a preferred embodiment, the state space parameters of the constructed rule model may include: ; , , ; in, Indicates the distance between the front of one vehicle and the rear of the vehicle in front; This represents the time constant of the chassis system.

[0080] Accordingly, specific implementation plans for constructing rule models may include: Based on H ∞ Robust control is achieved by constructing the following linear inequality (3) for the state-space parameters, and calculating the state feedback control gain matrix from the following linear inequality (3). ; (3) in, Indicates based on H ∞ The robust performance condition presupposes a positive definite symmetric matrix that satisfies ; Represents the state feedback control gain matrix; = ; Indicates based on H ∞ Robust performance conditions are preset positive scalars; Using the state feedback control gain matrix and the state variable matrix, the following rule model (4) is constructed: (4) in, This represents the expected acceleration of the vehicle as output by the rule-based model. Represents the state feedback control gain matrix; This represents the state variable matrix.

[0081] For the above rule model training process, the state space description formula (5) for vehicle closed-loop control is defined as follows: (5) Among them, the state variable matrix in vehicle closed-loop control ; Indicates corresponding to time State-space parameters (such as position, velocity, yaw angle, and sideslip angle, or one or more of these). The derivative of the state-space parameters; This indicates the output parameters.

[0082] Next, based on this formula (5), the vehicle closed-loop control uses H ∞ Robust control methods and rule models are designed.

[0083] Specifically, for a given vehicle closed-loop control, if the following conditions T1 and T2 are satisfied, then the vehicle closed-loop control is said to have H. ∞ performance: Condition T1: When hour, ,in, This indicates external disturbance. It can be regarded as a random variable that follows a normal distribution, that is: ,in, and All of these are preset constants.

[0084] Condition T2: A positive scalar exists. And positive definite symmetric matrices P and Q, such that the following linear inequality (3) holds: (3) in, Indicates based on H ∞ The robust performance condition presupposes a positive definite symmetric matrix that satisfies ; Represents the state feedback control gain matrix; = ; Indicates based on H ∞ The robust performance conditions are presupposed by positive scalars. This leads to the rule model: .

[0085] More specifically, for the above inequality (3), it can be guaranteed that H ∞ The performance is demonstrated as follows: Using Schur's complement theorem, we assume the existence of a positive scalar. This enables the vehicle closed-loop control to have H ∞ Performance, i.e.: ,in, Represents a positive definite matrix P The largest eigenvalue, The Lyapunov function is designed as follows: The derivative of this Lyapunov function is:

[0086] To describe the closed-loop system H ∞ Performance, introducing evaluation index functions :

[0087] in,

[0088] If satisfied This will ensure system H ∞ Performance. Multiply by both sides According to Schur's complement theorem, we can obtain:

[0089] Then the evaluation index function It can be equivalently described as : ,in, .

[0090] The state variable matrix of vehicle closed-loop control Substituting into the above equation and performing elementary transformations, we get:

[0091] set up Substituting this into the above equation, we get:

[0092] From the above derivation, we can conclude that as long as the evaluation index function is guaranteed... If it is established, then It also holds true, and thus the system of H ∞ Performance is guaranteed.

[0093] As can be seen from the above, the embodiments of the present invention can realize H-type closed-loop control of the vehicle based on the above linear inequality (3). ∞ Robust control.

[0094] Furthermore, regarding the construction of a data-driven model, in a preferred embodiment of the present invention, based on the vehicle state information, preceding vehicle information, and surrounding environment information included in the training data, a state space set and an action space set are determined for the DDPG model; using the state space set and action space set, the DDPG model is trained to obtain the data-driven model. The training data can be historical driving data of various vehicles provided by a vehicle management platform, which includes at least vehicle state information, preceding vehicle information, and surrounding environment information. Alternatively, the training data can be driving data of various vehicles monitored under multiple test scenarios.

[0095] Specifically, the state space set required to construct the data-driven model may include: the vehicle's actual following distance (which is the actual distance between the vehicle and the vehicle in front), the vehicle's expected following distance (which is the ideal distance between the vehicle and the vehicle in front, and can also be calculated using the constant vehicle-time distance model mentioned above), the vehicle's actual speed, the vehicle in front's actual speed, the vehicle's actual acceleration, and the vehicle in front's actual acceleration. The vehicle's actual following distance, actual speed, actual speed of the vehicle in front, actual acceleration of the vehicle, and actual acceleration of the vehicle in front can all be acquired through onboard sensors.

[0096] In addition, the action space set required to build a data-driven model generally includes the vehicle's expected acceleration.

[0097] In constructing the data-driven model, this embodiment of the invention also features a special design for the DDPG model. Specifically, based on a policy network-value network architecture, the DDPG model involved in this embodiment includes a policy network and a value network. The policy network generates policies and outputs actions. The policy network is a fully connected network, where each hidden layer uses a rectified linear unit (ReLU) as the activation function, and the output uses a hyperbolic tangent function (tanh). The value network fits the state-action value function and evaluates the actions output by the policy network. The value network includes two input layers, which take the state space set and the action space set as inputs, respectively. The state-action value function fitted by the value network can evaluate the quality of the actions output by the policy network in the current state.

[0098] The number of hidden layers in the policy network can be set according to requirements. For example, if there are 3 hidden layers, the modified linear unit is used as the activation function between the hidden layers, and the hyperbolic tangent function is used as the activation function in the 3rd hidden layer to connect the output policy and action.

[0099] In this system, both input layers of the value network are hidden layers, and a modified linear unit (MRU) is used as the activation function between the two hidden layers. That is, one hidden layer uses a MRU as the activation function to activate the next hidden layer, and the output of these two input layers is further input into the next connected hidden layer. For the last hidden layer, the MRU is used as the activation function to output the evaluation result. The parameters of the value network and the policy network are continuously adjusted throughout the training process.

[0100] For example, such as Figure 5As shown, the input (G1) of the policy network is the state space set (P1). A modified linear unit (MLU) is used as the activation function (H1) to activate the hidden layer (G2). The hidden layer (G2) processes the state space set (P1), and then another modified linear unit (MLU) is used as the activation function (H1) to activate the next hidden layer (G3). After the next hidden layer (G3) activates the hyperbolic tangent function (H2) as the activation function, it outputs the policy (Y) and action (Q). Further, as... Figure 6 As shown, the policy (Y) and action (Q) are input through the first input layer (F1) and the second input layer (F2) in the value network, respectively. Then, the hidden layer (G4) corresponding to the first input layer (F1) and the hidden layer (G5) corresponding to the second input layer (F2) are activated by the modified linear unit (H1) as the activation function. The hidden layer (G4) corresponding to the first input layer (F1) and the hidden layer (G5) corresponding to the second input layer (F2) are in the policy (Y) and action (Q), respectively. Then, the processing results of the hidden layer (G4) corresponding to the first input layer (F1) and the processing results of the hidden layer (G5) corresponding to the second input layer (F2) are input together into the next hidden layer (G6) activated by the modified linear unit (H1) as the activation function. Finally, the next hidden layer (G6) outputs the evaluation result (G7) by the hyperbolic tangent function as the activation function.

[0101] Furthermore, in order to ensure the accuracy of the training results of the DDPG model, the present invention also constructs the following reward function (6) for the DDPG model, and iteratively trains the DDPG model through the reward function (6).

[0102] (6)

[0103] in, Indicates the total reward; Indicates the weighting coefficient of the security reward; Indicates the weighting coefficient of efficiency rewards; The weighting coefficient representing the comfort reward; Indicates a safety reward; Indicates a reward for efficiency; This indicates a comfort reward.

[0104] Among them, the safety reward is calculated using the following formula (7).

[0105] (7)

[0106] in, Indicates a safety reward; This indicates the preset collision reward safety parameters. This represents the preset compensation factor. ; Indicates the expected following distance of the vehicle; This indicates the actual following distance of the vehicle.

[0107] The efficiency reward is calculated using the following formula (8).

[0108] (8)

[0109] in, Indicates a reward for efficiency; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; This represents the preset compensation factor. .

[0110] The comfort reward is calculated using the following formula (9).

[0111] (9) Indicates a comfort reward; jerk This indicates a jump (the jump is the rate of change of acceleration); This indicates the vehicle's actual acceleration; This represents the preset compensation factor. .

[0112] It is worth noting that the preset compensation factors involved in the above calculation formulas (7) to (9) can be the same or different.

[0113] The security reward, efficiency reward, comfort reward, and reward function constructed above can effectively improve the reliability and accuracy of the DDPG model.

[0114] Furthermore, to more accurately control the vehicle and reduce the longitudinal following risk, the technical solution provided in this embodiment of the invention further includes: determining a current speed limit for the vehicle. When the actual vehicle speed, as included in the vehicle's driving data, is greater than the current speed limit, the vehicle outputs deceleration; when the actual vehicle speed is less than the current speed limit, the vehicle is controlled according to the following acceleration. By introducing a current speed limit, the driving risk of the vehicle can be further reduced, and the driving safety of the vehicle can be improved.

[0115] Specifically, the implementation plan for determining the current speed limit for this vehicle may include: determining the minimum of the following: the road speed limit, the curve speed limit, the visibility speed limit, and the vehicle's set maximum cruising speed. Understandably, for non-curve sections, the minimum of the following can be used directly as the current speed limit: the road speed limit, the visibility speed limit, and the vehicle's set maximum cruising speed. Furthermore, if the current environment is not in abnormal weather conditions such as rain, snow, or fog, the visibility speed limit is not considered.

[0116] More specifically, the road speed limit is calculated using the following formula (10): (10) in, Indicates the road speed limit; This indicates the speed limit indicated by the speed limit sign; Indicates the maximum speed limit corresponding to the road's classification level; This indicates that the road has speed limit signs. Specifically, it refers to speed limits identified by cameras from speed limit signs posted alongside the lanes. Then, the road speed limit is determined to be the speed limit indicated by the speed limit sign. If the speed limit sign beside the lane is not detected by the camera, the road speed limit is determined to be the maximum speed corresponding to the road's classification level. For example, for a lane without a center line, the road classification is urban road, and the corresponding maximum speed is 30 km / h; for a highway, the corresponding maximum speed is 40 km / h. For a road with only one lane in the same direction, the road classification is urban road, and the corresponding maximum speed is 50 km / h; for a highway, the corresponding maximum speed is 70 km / h. For a road with two or more lanes in the same direction, the road classification is urban road, and the corresponding maximum speed is 70 km / h; for a highway, the corresponding maximum speed is 80 km / h.

[0117] In addition, the speed limit for curves is calculated using the following formula (11): (11) Indicates the speed limit for the curve; Indicates the radius of curvature of the road. , and These represent the parameter settings.

[0118] The visibility speed limit is calculated using the following formula (12): (12) Indicates visibility speed limit; This indicates the visible distance of the vehicle's onboard sensor system. Indicates reaction time.

[0119] In particular, the technical solutions provided by the embodiments of the present invention can be implemented on the basis of existing L2 level assisted driving.

[0120] Furthermore, the blind spot assisted driving methods provided in the above embodiments can be implemented through interaction between multiple functional layers. Specifically, such as... Figure 7 and Figure 8 As shown, the functional layers provided in this embodiment of the invention are: a perception fusion and prediction layer Ly1, a planning and control layer Ly2, and an execution layer Ly3. Specifically, in the perception fusion and prediction layer Ly1, sensor data D12, vehicle status D13, and environmental information D14 provided by external sensors Se, ECU, and high-precision map Ma, respectively, are fused and preprocessed (A1) to obtain the current distance between the vehicle and the preceding vehicle, the vehicle's driving data, and the vehicle's driving scenario as described in step S101. The perception fusion and prediction layer Ly1 then transmits the results of the information fusion and preprocessing (A1) (the current distance between the vehicle and the preceding vehicle, the vehicle's driving data, and the vehicle's driving scenario) to the planning and control layer Ly2. Within the planning and control layer Ly2, a dynamic fusion control law (A4) is constructed, and a speed limit (A2) is determined for the vehicle. Specifically, for constructing the dynamic fusion control law (A4), the rule model (RM) built based on rules (D111) and the data-driven model (DM) built based on data-driven (D112) are dynamically fused to form the dynamic fusion control law (IM). The dynamic fusion control law (IM) then generates acceleration for the vehicle. The process of determining the speed limit (A2) for this vehicle is based on road and environmental information (D15), using the minimum value among the road speed limit, visibility speed limit, curve speed limit, and the vehicle's set maximum cruising speed as the vehicle's speed limit. Based on the vehicle's current actual speed ( The speed limit of this vehicle is ( ). The dynamic fusion control law (IM) generates acceleration for this vehicle. The target acceleration (a) is output to the execution layer Ly3, and the execution layer Ly3 controls the chassis domain controller DC, the motor controller MC and the powertrain PT based on the target acceleration (a).

[0121] In addition, such as Figure 8As shown, the construction process for the data-driven model (DM) is as follows: A state space set and an action space set are selected from the training data (A01), and a training network (A02) and a reward function (A03) are designed for the data-driven model (DM). The model is then trained by combining the state space set, action space set, training network, and reward function (A04), thus obtaining the data-driven model (DM). The construction process for the rule-based model (RM) is as follows: A fixed state variable matrix and state space parameters are selected (A05), and H is determined. ∞ Robust control of H ∞ Robust performance condition (A06), combined with the fixed state variable matrix, state space parameters and H ∞ The robust performance condition (A07) is used to construct the rule model (RM).

[0122] The vehicle's acceleration is generated through a dynamic fusion control law (IM). Specifically, the first acceleration can be obtained from the rule model (RM). The second acceleration model obtained from the data-driven model (DM) Acceleration is obtained through weighted calculation. ).

[0123] In addition, the determination of the target acceleration (a) mainly involves: at the actual vehicle speed ( () is less than the vehicle's speed limit () In the case of ), the target acceleration (a) is equal to the weighted acceleration (a). At actual vehicle speed ( () is greater than or equal to the vehicle's speed limit ( In the case of ), the target acceleration (a) is a set negative value, that is, it becomes deceleration.

[0124] In summary, the technical solution provided by the embodiments of the present invention retains the flexibility and generalization ability of the data-driven vehicle-following model while ensuring a certain degree of interpretability and controllability. Therefore, the technical solution provided by the embodiments of the present invention can effectively improve the conservatism of rule-based models and the uncontrollability risks of data-driven models, meet regulatory requirements and driver needs, achieve a balanced optimization of safety, comfort, and traffic efficiency in the vehicle-following process under various road conditions, and effectively suppress uncertain interference in complex environments, exhibiting strong robustness and adaptability.

[0125] Furthermore, the technical solution provided by the embodiments of the present invention is particularly effective in low-speed congested following and medium-to-high-speed following cruise scenarios, significantly reducing driver workload, improving safety performance, and enhancing traffic efficiency. In addition, the technical solution provided by the embodiments of the present invention, by fusing two control models (rule-based model and data-driven model), possesses the dual advantages of the controllability attribute of the rule-based model and the high-precision attribute of the data-driven model, offering valuable insights for high-level ADAS systems that balance safety and flexibility in auxiliary control.

[0126] Furthermore, embodiments of the present invention provide an assisted driving device for longitudinal following scenarios. For example... Figure 9 As shown, the driver assistance device 900 may include: a model building module 901, a generation module 902, and an auxiliary control module 903, wherein, The model building module 901 is used to dynamically fuse rule-based models and data-driven models based on the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario. This results in the construction of a dynamic fusion control law for the vehicle in a longitudinal following scenario. The rule-based model is based on H... ∞ Robust control was constructed, and the data-driven model was constructed based on the Deep Deterministic Policy Gradient (DDPG) model. The generation module 902 is used to generate the following acceleration relative to the preceding vehicle for the vehicle using a dynamic fusion control law; The auxiliary control module 903 is used to adjust the vehicle according to the following vehicle's acceleration.

[0127] In this embodiment of the invention, the model building module 901 is further configured to build a first mapping relationship for the rule model, wherein in the first mapping relationship, a combination of a vehicle spacing and a vehicle speed should be provided with a weight coefficient corresponding to the rule model; build a second mapping relationship between at least one driving scenario and the weight coefficient corresponding to the rule model; determine a target weight coefficient for the rule model based on the first mapping relationship and the second mapping relationship, wherein the target weight coefficient satisfies the current vehicle spacing and the vehicle's driving data and / or the target weight coefficient satisfies the vehicle's driving scenario; and fuse the rule model and the data-driven model using the target weight coefficient.

[0128] In this embodiment of the invention, the model building module 901 is further configured to: find a first weight coefficient corresponding to the current vehicle speed and the current vehicle distance and driving data of the vehicle from the first mapping relationship; find a second weight coefficient corresponding to the driving scenario of the vehicle from the second mapping relationship; and determine the smaller value between the first weight coefficient and the second weight coefficient as the target weight coefficient if the first weight coefficient and the second weight coefficient are inconsistent.

[0129] In this embodiment of the invention, the model building module 901 is further used to, based on the vehicle state information, preceding vehicle information, and surrounding environment information included in the training data, provide H ∞ Robust control determines the state variable matrix and state space parameters; it then compares the state variable matrix and state space parameters with H... ∞ Robust control is combined to construct a rule model.

[0130] In this embodiment of the invention, the model building module 901 is further configured to determine, based on vehicle state information and preceding vehicle information, the actual following distance of the vehicle, the expected following distance of the vehicle, the actual speed of the vehicle, the actual speed of the preceding vehicle, the first acceleration of the current adjustment cycle, and the expected acceleration of the current adjustment cycle; and to construct, using the actual following distance, the expected following distance, the actual speed of the vehicle, the actual speed of the preceding vehicle, the first acceleration of the current adjustment cycle, and the expected acceleration of the current adjustment cycle, the following state variable matrix corresponding to the current adjustment cycle is constructed. :

[0131] in, ; ; ; Indicates the following distance error of the vehicle; Indicates the actual following distance of the vehicle; Indicates the expected following distance of the vehicle; This indicates the longitudinal speed error of the vehicle; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; T Indicates the time constant of the sampling step; This indicates the vehicle's actual acceleration; Indicates the vehicle's expected acceleration; This represents the time constant of the chassis system; Indicates the first One adjustment cycle.

[0132] In this embodiment of the invention, the model building module 901 is further used to calculate the expected following distance of the vehicle using the following constant inter-vehicle time distance model; Constant workshop time-distance model:

[0133] in, This indicates the minimum static inter-vehicle distance set for the vehicle. This indicates the upper limit of the longitudinal following distance set for the vehicle; This indicates the lower limit of the vehicle's set longitudinal following distance. Indicates the vehicle's actual speed; This indicates the relative speed between the two workshops; The actual acceleration of the vehicle in front; Indicates the mass of the vehicle; , and All are parameters, and satisfy the following conditions: , , .

[0134] In this embodiment of the invention, the state space parameters include: ; , , ; in, Indicates the distance between the front of one vehicle and the rear of the vehicle in front; This represents the time constant of the chassis system.

[0135] Based on these state-space parameters, model building module 901 is further used for H-based... ∞ Robust control is achieved by constructing the following linear inequalities for the state-space parameters and calculating the state feedback control gain matrix from these linear inequalities.

[0136] in, Indicates based on H ∞ The robust performance condition presupposes a positive definite symmetric matrix that satisfies ; Represents the state feedback control gain matrix; = ; Indicates based on H ∞ Robust performance conditions are preset positive scalars; Using the state feedback control gain matrix and the state variable matrix, the following rule model is constructed:

[0137] in, This represents the expected acceleration of the vehicle as output by the rule-based model. Represents the state feedback control gain matrix; This represents the state variable matrix.

[0138] In this embodiment of the invention, the model building module 901 is further used to determine the state space set and action space set for the DDPG model based on the vehicle state information, the preceding vehicle information and the surrounding environment information included in the training data; and to train the DDPG model using the state space set and action space set to obtain the data-driven model.

[0139] The state space set includes: the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the vehicle's actual acceleration, and the preceding vehicle's actual acceleration. The action space set includes: the vehicle's expected acceleration.

[0140] The DDPG model includes a policy network and a value network. The policy network generates policies and outputs actions. It is a fully connected network, with modified linear units used as activation functions between hidden layers and a hyperbolic tangent function as the final output. The value network fits the state-action value function and evaluates the actions output by the policy network. It has two input layers, which take the state space set and the action space set as inputs, respectively.

[0141] In this embodiment of the invention, the model building module 901 is further configured to build a reward function for the DDPG model with the following: , in, Indicates the total reward; Indicates the weighting coefficient of the security reward; Indicates the weighting coefficient of efficiency rewards; The weighting coefficient representing the comfort reward; Indicates a safety reward; Indicates a reward for efficiency; This indicates a comfort reward.

[0142] In this embodiment of the invention, the model building module 901 is further used to calculate the security reward using the following calculation formula one; Formula 1:

[0143] in, Indicates a safety reward; This indicates the preset collision reward safety parameters. This represents the preset compensation factor. ; Indicates the expected following distance of the vehicle; This indicates the actual following distance of the vehicle.

[0144] In this embodiment of the invention, the model building module 901 is further used to calculate the efficiency reward using the following calculation formula two. Formula 2:

[0145] in, Indicates a reward for efficiency; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; This represents the preset compensation factor. .

[0146] In this embodiment of the invention, the model building module 901 is further used to calculate the comfort reward using the following calculation formula three. Formula 3: Indicates a comfort reward; jerk Indicates judder; This indicates the vehicle's actual acceleration; This represents the preset compensation factor. .

[0147] In this embodiment of the invention, the auxiliary control module 903 is further used to determine the current speed limit for the vehicle; when the actual vehicle speed in the vehicle's driving data is greater than the current speed limit, it controls the vehicle to output deceleration; when the actual vehicle speed is less than the current speed limit, it performs vehicle acceleration control according to following vehicle acceleration.

[0148] In this embodiment of the invention, the auxiliary control module 903 is further used to determine the current speed limit as the smallest of the road speed limit, curve speed limit, visibility speed limit and the vehicle's set maximum cruise speed.

[0149] In this embodiment of the invention, the auxiliary control module 903 is further configured to calculate the road speed limit using the following calculation formula:

[0150] in, Indicates the road speed limit; This indicates the speed limit indicated by the speed limit sign; Indicates the maximum speed limit corresponding to the road's classification level; This indicates that there are speed limit signs on the road.

[0151] In this embodiment of the invention, the auxiliary control module 903 is further used to calculate the speed limit for curves using the following calculation formula: Indicates the speed limit for the curve; Indicates the radius of curvature of the road. , and These represent the parameter settings.

[0152] In this embodiment of the invention, the auxiliary control module 903 is further configured to calculate the visibility speed limit using the following calculation formula: Indicates visibility speed limit; This indicates the visible distance of the vehicle's onboard sensor system. Indicates reaction time.

[0153] The modules in the aforementioned driver assistance devices can be integrated into the vehicle's ADAS to implement driver assistance methods for longitudinal following scenarios.

[0154] In one embodiment of the present invention, the rule model building process and the data-driven model building process in the model building module of the above-mentioned driver assistance device can be implemented by a server or the cloud, and the trained rule model and data-driven model can be distributed to each vehicle by the server or the cloud.

[0155] In one embodiment of the present invention, the dynamic fusion of the rule model and the data-driven model in the above-mentioned model building module to build a dynamic fusion control law for the vehicle in the longitudinal following scenario can be implemented on the vehicle side through ADAS, or through a server or the cloud.

[0156] Furthermore, embodiments of the present invention provide an electronic device for assisted driving in a longitudinal following scenario, which is applied to a vehicle. Specifically, the electronic device for assisted driving in a longitudinal following scenario may include: One or more processors; 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 assisted driving method for longitudinal following scenarios provided in any of the above embodiments.

[0157] Furthermore, embodiments of the present invention provide a vehicle. This vehicle may include the driver assistance device for longitudinal following scenarios provided in the above embodiments, or the electronic device for driver assistance for longitudinal following scenarios provided in the above embodiments.

[0158] In addition, the aforementioned driver assistance devices for longitudinal following scenarios can also be applied to servers.

[0159] Figure 10 An exemplary vehicle system architecture 1000 is shown, to which the assisted driving method or assisted driving device for longitudinal following scenarios can be applied according to embodiments of the present invention.

[0160] like Figure 10 As shown, the vehicle system architecture 1000 may include various systems, such as a driving control system 1001, a power system 1002, a sensor system 1003, a control system 1004, one or more peripheral devices 1005, a power supply 1006, a computer system 1007, and a user interface 1008. The assisted driving method for longitudinal following scenarios provided in this embodiment can be implemented through interaction with the aforementioned systems. Optionally, the vehicle system architecture 1000 may include more or fewer systems, and each system may include multiple components. Furthermore, each system and component of the vehicle system architecture 1000 can be interconnected via wired or wireless means.

[0161] The vehicle system architecture 1000 includes a driving control system 1001, which can be in a fully or partially automated driving mode. For example, the driving control system 1001 can automatically control the vehicle to follow the vehicle longitudinally without human interaction, based on an assisted driving method for longitudinal following scenarios.

[0162] The powertrain 1002 may include components that provide power to the vehicle. For example, the powertrain 1002 may include an engine, an energy source, a transmission, wheels, tires, etc. The engine may be an internal combustion engine, an electric motor, an air-compressed engine, or other combinations of engines, such as a hybrid engine consisting of a gasoline engine and an electric motor, or a hybrid engine consisting of an internal combustion engine and an air-compressed engine. The engine converts the energy source into mechanical energy to supply the transmission. Examples of energy sources may include gasoline, diesel, other petroleum-based fuels, propane, other compressed gas-based fuels, ethanol, solar panels, batteries, and other electrical sources. The energy source may also provide energy to other systems of the vehicle. Furthermore, the transmission may include a gearbox, a differential, a drive shaft, and a clutch, etc.

[0163] The sensor system 1003 may include sensors for sensing the vehicle's surrounding environment and pressure sensors for sensing whether there are passengers in the seats. Examples include a positioning system (which may be a Global Positioning System (GPS), BeiDou Navigation Satellite System, or other positioning systems), radar, a laser rangefinder, an inertial measurement unit (IMU), a camera, and metal detection devices such as capacitive sensors. The positioning system can be used to determine the vehicle's geographical location. The IMU is used to sense changes in the vehicle's position and orientation based on inertial acceleration. In one embodiment, the IMU may be a combination of an accelerometer and a gyroscope. The radar can use radio signals to sense objects in the vehicle's surrounding environment. In some embodiments, in addition to sensing objects, the radar can also be used to sense the speed and / or direction of travel of objects.

[0164] To detect environmental information and objects outside the vehicle, cameras can be configured at appropriate locations on the vehicle's exterior. For example, to acquire environmental images of the vehicle's sides, a camera can be mounted on the side mirror. The camera can be a still or video camera.

[0165] The control system 1004 may include software systems for implementing vehicle driving control, such as systems for analyzing the vehicle's surrounding environment, pretensioning seat belts, route planning, obstacle avoidance, and image analysis. The control system 1004 may also include hardware systems such as an accelerator, steering wheel system, seat belt system, and airbag system. Furthermore, the control system 1004 may add or replace components other than those shown and described. Alternatively, some of the components shown above may be omitted.

[0166] Furthermore, as described above, the control system 1004 can also be used as part of the aforementioned assisted driving method execution system for longitudinal following scenarios, utilizing a dynamic fusion control law to generate a following acceleration relative to the preceding vehicle for the vehicle itself; and adjusting the vehicle according to the following acceleration.

[0167] Additionally, the control system 1004 can interact with external sensors, other autonomous driving devices, other computer systems, or users via peripheral devices 1005. Peripheral devices 1005 may include wireless communication systems, on-board computers, microphones, and / or speakers.

[0168] In some embodiments, peripheral device 1005 provides a means for user interaction with the control system 1004 via a user interface. For example, an onboard computer may provide information to a user of the vehicle. The user interface may also operate the onboard computer to receive user input. The onboard computer may be operated via a touchscreen. In other cases, peripheral device may provide a means for communicating with other devices located within the vehicle. For example, a microphone may receive audio (e.g., voice commands or other audio input) from a user of the control system 1004. Similarly, a speaker may output audio to a user of the control system 1004.

[0169] Wireless communication systems can communicate wirelessly with one or more devices, either directly or via a communication network. For example, wireless communication systems can use networks such as cellular networks, WiFi, and wireless local area networks (WLANs), or they can use infrared links, Bluetooth, or ZigBee to communicate directly with devices. Other wireless protocols include those used in various autonomous driving communication systems.

[0170] The power source 1006 can provide power to various components of the vehicle. The power source 1006 can be a rechargeable lithium-ion or lead-acid battery.

[0171] The computer system 1007 controls some or all of the functions of the assisted driving method for longitudinal following scenarios. The computer system 1007 may include at least one processor that executes instructions stored in a non-transitory computer-readable medium such as memory. The computer system 1007 provides the aforementioned control system with execution code for implementing the assisted driving method for longitudinal following scenarios.

[0172] The processor can be any conventional processor, such as a commercially available central processing unit (CPU). Alternatively, the processor can be a dedicated device such as an application-specific integrated circuit (ASIC) or other hardware-based processor. Those skilled in the art will understand that the processor, computer, or memory can actually include multiple processors, computers, or memories that may or may not be stored in the same physical housing. For example, memory can be a hard disk drive or other storage media located in a housing different from that of a computer. Therefore, references to processors or computers will be understood to include references to a collection of processors or computers or memories that may or may not operate in parallel. Unlike using a single processor to perform the steps described herein, some components, such as steering and deceleration components, may each have their own processor that performs calculations only related to the component's specific function.

[0173] User interface 1008 is used to provide information to or receive information from users of the vehicle. Optionally, user interface 1008 may include one or more input / output devices within a set of peripheral devices 1005, such as wireless communication systems, on-board computers, microphones, and speakers.

[0174] It should be understood that the components described above are merely an example. In actual applications, components in the various modules or systems mentioned above may be added or removed as needed. Figure 10 This should not be construed as a limitation on the embodiments of this application.

[0175] The following is for reference. Figure 11 It shows a schematic diagram of the structure of a computer system 1100 suitable for implementing an assisted driving method for longitudinal following scenarios in accordance with embodiments of the present invention. Figure 11 The computer system shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0176] like Figure 11 As shown, the computer system 1100 includes a central processing unit (CPU) 1101, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 1102 or programs loaded from storage section 1108 into random access memory (RAM) 1103. The RAM 1103 also stores various programs and data required for the operation of the system 1100. The CPU 1101, ROM 1102, and RAM 1103 are interconnected via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0177] The following components are connected to I / O interface 1105: an input section 1106; an output section 1107 including devices such as cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; a storage section 1108 including devices such as hard disks; and a communication section 1109 including network interface cards such as LAN cards and modems. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to I / O interface 1105 as needed. Removable media 1111, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 1110 as needed so that computer programs read from them can be installed into storage section 1108 as needed.

[0178] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1109, and / or installed from removable medium 1111. When the computer program is executed by central processing unit (CPU) 1101, it performs the functions defined above in the system of this invention.

[0179] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0180] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0181] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor can be described as including the aforementioned model building module, generation module, and auxiliary control module. The names of these modules or units do not necessarily limit the module or unit itself; for example, the generation module can also be described as "a module or unit that generates the following acceleration relative to the preceding vehicle for this vehicle."

[0182] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to: dynamically fuse a rule model and a data-driven model based on the current distance between the vehicle and the preceding vehicle, the vehicle's driving data, and the vehicle's driving scenario, to construct a dynamic fusion control law for the vehicle in a longitudinal following scenario, wherein the rule model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on the deep deterministic policy gradient model. The dynamic fusion control law is used to generate the following acceleration of the vehicle relative to the vehicle in front. The vehicle is adjusted according to the following acceleration.

[0183] According to the technical solution of the present invention, in the process of constructing a dynamic fusion control law for the vehicle in a longitudinal following scenario, the current distance between the vehicle and the preceding vehicle, the vehicle's driving data, and the vehicle's driving scenario are used as references to dynamically fuse the rule model and the data-driven model. This can compensate for the defects of a single model, enabling the constructed dynamic fusion control law to better match the vehicle's environment, thereby accurately controlling the vehicle to follow the preceding vehicle, improving the reliability of longitudinal following, and enhancing the user's driving experience and vehicle traffic efficiency. In particular, the technical solution provided by the embodiments of the present invention can adapt well to low-speed congestion scenarios and improve vehicle traffic efficiency in low-speed congestion scenarios.

[0184] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A driving assistance method for longitudinal following scenarios, characterized in that, include: Based on the current distance between this vehicle and the vehicle in front, this vehicle's driving data, and this vehicle's driving scenario, a rule-based model and a data-driven model are dynamically fused to construct a dynamic fusion control law for this vehicle in longitudinal following scenarios. The rule-based model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on a deep deterministic policy gradient model; Using the aforementioned dynamic fusion control law, a following acceleration relative to the preceding vehicle is generated for this vehicle; Adjust the vehicle's acceleration according to the stated following vehicle acceleration.

2. The assisted driving method according to claim 1, characterized in that, Also includes: A first mapping relationship is constructed for the rule model, wherein, in the first mapping relationship, a combination of a vehicle spacing and a vehicle speed is mapped to a weight coefficient corresponding to the rule model; Construct a second mapping relationship between at least one driving scenario and the weight coefficients corresponding to the rule model; The dynamic fusion of rule-based models and data-driven models includes: Based on the first mapping relationship and the second mapping relationship, a target weight coefficient is determined for the rule model, wherein the target weight coefficient satisfies the current vehicle spacing and the vehicle's driving data and / or the target weight coefficient satisfies the vehicle's driving scenario; The target weight coefficients are used to fuse the rule model and the data-driven model.

3. The assisted driving method according to claim 2, characterized in that, Determining the target weight coefficients for the rule model includes: From the first mapping relationship, find the first weighting coefficient corresponding to the current vehicle speed included in the current vehicle distance and the vehicle's driving data; From the second mapping relationship, find the second weight coefficient corresponding to the driving scenario of the vehicle; If the first weighting coefficient and the second weighting coefficient are inconsistent, and the driving scenario of the vehicle is a low-speed driving scenario, the larger of the first weighting coefficient and the second weighting coefficient is determined to be the target weighting coefficient; if the driving scenario of the vehicle is a medium-high speed driving scenario, the smaller of the first weighting coefficient and the second weighting coefficient is determined to be the target weighting coefficient.

4. The assisted driving method according to any one of claims 1 to 3, characterized in that, Also includes: Based on the training data, including vehicle status information, preceding vehicle information, and surrounding environment information, for H ∞ Robust control determines the state variable matrix and state space parameters; The state variable matrix and the state space parameters are compared with H. ∞ Robust control is combined to construct a rule model.

5. The assisted driving method according to claim 4, characterized in that, For H ∞ Robust control determines the state variable matrix, which includes: Based on the vehicle status information and the preceding vehicle information, determine the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment cycle, and the expected acceleration of the current adjustment cycle. Using the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the preceding vehicle's actual speed, the first acceleration of the current adjustment period, and the expected acceleration of the current adjustment period, the following state variable matrix corresponding to the current adjustment period is constructed. : in, ; ; ; Indicates the following distance error of the vehicle; Indicates the actual following distance of the vehicle; Indicates the expected following distance of the vehicle; This indicates the longitudinal speed error of the vehicle; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; T Indicates the time constant of the sampling step; This indicates the vehicle's actual acceleration; Indicates the vehicle's expected acceleration; This represents the time constant of the chassis system; Indicates the first One adjustment cycle.

6. The assisted driving method according to claim 5, characterized in that, Determine the desired following distance for the vehicle, including: The following constant interval model is used to calculate the expected following distance of the vehicle; Constant workshop time-distance model: in, This indicates the minimum static inter-vehicle distance set for the vehicle. This indicates the upper limit of the longitudinal following distance set for the vehicle; This indicates the lower limit of the vehicle's set longitudinal following distance. Indicates the vehicle's actual speed; This indicates the relative speed between the two workshops; The actual acceleration of the vehicle in front; Indicates the mass of the vehicle; , and All are parameters, and satisfy the following conditions: , , .

7. The assisted driving method according to claim 4, characterized in that, The state space parameters include: ； ， ， ； in, Indicates the distance between the front of one vehicle and the rear of the vehicle in front; This represents the time constant of the chassis system; The construction rule model includes: Based on H ∞ Robust control is achieved by constructing the following linear inequalities for the state space parameters and calculating the state feedback control gain matrix from these linear inequalities. in, Indicates based on H ∞ The robust performance condition presupposes a positive definite symmetric matrix that satisfies ; Represents the state feedback control gain matrix; = ; Indicates based on H ∞ Robust performance conditions are preset positive scalars; Using the state feedback control gain matrix and the state variable matrix, the following rule model is constructed: in, This represents the expected acceleration of the vehicle as output by the rule-based model. Represents the state feedback control gain matrix; This represents the state variable matrix.

8. The assisted driving method according to any one of claims 1 to 3, characterized in that, Also includes: Based on the vehicle state information, preceding vehicle information, and surrounding environment information included in the training data, the state space set and action space set are determined for the deep deterministic policy gradient model. Using the state space set and the action space set, a deep deterministic policy gradient model is trained to obtain the data-driven model.

9. The assisted driving method according to claim 8, characterized in that, The state space set includes: the vehicle's actual following distance, the vehicle's expected following distance, the vehicle's actual speed, the vehicle in front's actual speed, the vehicle's actual acceleration, and the vehicle in front's actual acceleration. The set of motion spaces includes the vehicle's desired acceleration.

10. The assisted driving method according to claim 8 or 9, characterized in that, The deep deterministic policy gradient model includes a policy network and a value network. The policy network is used to generate policies and output actions. The policy network is a fully connected network. The hidden layers of the fully connected network use modified linear units as activation functions, and the hyperbolic tangent function is used as the activation function for the final output. The value network is used to fit the state-action value function and evaluate the actions output by the policy network. The value network includes two input layers, which are respectively input to the state space set and the action space set.

11. The assisted driving method according to claim 8 or 9, characterized in that, Also includes: The following reward function is constructed for the deep deterministic policy gradient model: ， in, Indicates the total reward; Indicates the weighting coefficient of the security reward; Indicates the weighting coefficient of efficiency rewards; The weighting coefficient representing the comfort reward; Indicates a safety reward; Indicates a reward for efficiency; This indicates a comfort reward.

12. The assisted driving method according to claim 11, characterized in that, Also includes: The security reward is calculated using the following formula: Formula 1: in, Indicates a safety reward; This indicates the preset collision reward safety parameters. This represents the preset compensation factor. ; Indicates the expected following distance of the vehicle; Indicates the actual following distance of the vehicle; And / or, The efficiency bonus is calculated using the following formula (Formula 2); Formula 2: in, Indicates a reward for efficiency; Indicates the actual speed of the vehicle in front; Indicates the vehicle's actual speed; This represents the preset compensation factor. ; And / or, The comfort reward is calculated using the following formula three; Formula 3: Indicates a comfort reward; jerk Indicates judder; This indicates the vehicle's actual acceleration; This represents the preset compensation factor. .

13. The assisted driving method according to any one of claims 1 to 12, characterized in that, The assisted driving method further includes: determining a current speed limit for the vehicle; If the actual current vehicle speed, as included in the vehicle's driving data, is greater than the current speed limit, the vehicle will be controlled to output deceleration. If the current actual vehicle speed is less than the current speed limit, the vehicle is adjusted according to the following acceleration.

14. The assisted driving method according to claim 13, characterized in that, Determining the current speed limit for the vehicle includes: The current speed limit is determined by the smallest of the following: road speed limit, curve speed limit, visibility speed limit, and the vehicle's set maximum cruise speed.

15. The assisted driving method according to claim 14, characterized in that, The speed limit for the road is calculated using the following formula: in, Indicates the road speed limit; This indicates the speed limit indicated by the speed limit sign; Indicates the maximum speed limit corresponding to the road's classification level; This indicates that there are speed limit signs on the road; And / or, The speed limit for the curve is calculated using the following formula: Indicates the speed limit for the curve; Indicates the radius of curvature of the road. , and These represent the parameter settings; And / or, The visibility speed limit is calculated using the following formula: Indicates visibility speed limit; This indicates the visible distance of the vehicle's onboard sensor system. Indicates reaction time.

16. A driver assistance device for longitudinal following scenarios, characterized in that, include: The model consists of a model building module, a generation module, and an auxiliary control module. The model building module is used to dynamically fuse a rule-based model and a data-driven model based on the current distance between the vehicle and the vehicle in front, the vehicle's driving data, and the vehicle's driving scenario. This results in the construction of a dynamic fusion control law for the vehicle in a longitudinal following scenario. The rule-based model is based on H... ∞ Robust control is constructed, and the data-driven model is constructed based on a deep deterministic policy gradient model; The generation module is used to generate a following acceleration relative to the preceding vehicle for the vehicle using the dynamic fusion control law; The auxiliary control module is used to adjust the vehicle according to the following acceleration.

17. An electronic device for assisting driving in longitudinal following scenarios, characterized in that, The electronic device includes: One or more processors; 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 assisted driving method for longitudinal following scenarios as described in claims 1-15.

18. A vehicle, characterized in that, The electronic device includes the driver assistance device for longitudinal following scenarios as described in claim 16 or the driver assistance device for longitudinal following scenarios as described in claim 17.

Images