Intelligent driving continuous domain decision level space optimization method for extended game

Abstract

The application discloses an intelligent driving continuous domain decision level space optimization method for extended game, and comprises the following steps: an intelligent driving vehicle respectively realizes information acquisition and sharing with surrounding traffic vehicles and a traffic environment; an extended game decision model of the intelligent driving vehicle is established; each vehicle participating in the game predicts the reaction of other vehicles to the driving behavior of the vehicle, respectively obtains an expected utility function, an equality constraint and an inequality constraint of a preceding vehicle and a following vehicle, solves the above model, and outputs an acceleration strategy of each vehicle at T moment to execute a chassis control instruction; information of the intelligent driving vehicle, the surrounding traffic vehicles and the traffic environment is updated at T+1 moment, and each vehicle re-predicts the reaction of other vehicles to the driving behavior of the vehicle; the above steps are repeatedly executed until each vehicle eliminates the risk and exits the game interaction process. The method improves the interactivity of multi-vehicle decision, overcomes the problem of rough solution in a discrete domain, and improves the accuracy of optimization solution.

Description

Technical Field

[0001] This invention belongs to the technical field of interactive behavior decision-making and optimization solutions for intelligent driving vehicles, and in particular, a hierarchical space optimization method for continuous domain decision-making in intelligent driving oriented towards extended game theory. Background Technology

[0002] As research into intelligent technologies continues to deepen, intelligent driving vehicles have garnered widespread attention. They represent a cutting-edge trend and an inevitable direction for future automotive development. Decision-making is a crucial hub and key link in enhancing the intelligence and human-like qualities of vehicle driving, acting as a central nervous system similar to the human driver's brain. Simultaneously, intelligent driving vehicles, surrounding traffic vehicles, and traffic environment factors are dynamically coupled, exhibiting strong interactivity. Extended game theory is a powerful technical tool for describing the interactivity of decision-making; however, its inherent sequential characteristics of decision actions, the ability of the initiator to accurately predict the optimal response strategy of the followers and thus make decisions in advance, and the non-simultaneous nature of decision information stamps all present technical challenges to decision-making and optimization solutions.

[0003] Existing solutions describe decision-making using specific task definitions, incorporating numerous inductive reasoning functions. They often sample action intervals as a finite number of discrete variables, neglecting the fact that these are finite continuous variables. Discrete actions are also unfavorable for accurately determining the optimal strategy. In extended game theory, the optimization objective functionals of the forerunners and followers are simultaneously related to themselves and the environment, forming a hierarchically intertwined relationship. The actual game equilibrium point of a specific model does not always fall within the set of discrete action strategies; rather, it can exist in any of the continuous action domains. Only when the forerunners and followers can very precisely adopt strategies that satisfy the perfect Nash equilibrium of the extended game subgames can the accuracy, stability, and reliability of multi-vehicle decision-making results be achieved. At this point, no participating vehicle has a unilateral incentive to change its own strategy. Discrete action domains struggle to achieve such goals, often accompanied by decision bias and the motivation to deviate from equilibrium.

[0004] Therefore, a decision optimization method that can effectively adapt to hierarchical sequential driving behavior while also considering the accuracy of optimization solutions and the stability of interactive decision-making is a key technical challenge that urgently needs to be addressed to truly achieve human-like driving. Solving these problems will improve the rationality of intelligent driving vehicle decisions and is of great significance for enhancing decision-making accuracy and machine intelligence. Summary of the Invention

[0005] In view of the above problems, the present invention provides a hierarchical space optimization method for continuous domain decision-making in intelligent driving oriented towards extended game theory, which at least solves some of the above technical problems. The method fully considers the hierarchical characteristics of extended game theory, improves the interactivity of multi-vehicle decision-making, overcomes the problem of coarse solutions in discrete domain, and further improves the accuracy of optimization solutions.

[0006] This invention provides a hierarchical space optimization method for continuous domain decision-making in intelligent driving, oriented towards extended game theory, including:

[0007] S1. Intelligent driving vehicles acquire and share information with surrounding traffic vehicles and the traffic environment;

[0008] S2. Based on the acquired information, establish an extended game decision-making model for intelligent driving vehicles within a continuous action domain;

[0009] S3. Each car participating in the game predicts the reactions of other cars to its own driving behavior and obtains the expected utility functions, equality constraints and inequality constraints of the leading car and the following car respectively.

[0010] S4. Based on the expected utility functions, equality constraints, and inequality constraints of the leading and following vehicles, solve the extended game decision model of the intelligent driving vehicle using the hierarchical space optimization method, and output the acceleration strategy of each vehicle participating in the game at time T.

[0011] S5. The intelligent driving vehicle executes its own acceleration strategy at time T;

[0012] S6. At time T+1, update the information of the intelligent driving vehicle, surrounding traffic vehicles and traffic environment, and each car participating in the game re-predicts the reactions of other cars to its own driving behavior.

[0013] S7. Repeat steps S1-S6 until each vehicle resolves its risk and exits the game interaction process.

[0014] Furthermore, S1 also includes: when the acquired information is incomplete, the intelligent driving vehicle uses the collected historical information of the traffic environment, including surrounding traffic vehicles, and combines it with its understanding and cognition of the surrounding traffic vehicles and traffic environment to supplement the incomplete information.

[0015] Furthermore, in S2, the intelligent driving vehicle extended game decision model involves multi-objective vehicle driving needs and traffic performance needs embedded with driving styles;

[0016] The multi-objective vehicle driving requirements include the safety requirements, comfort requirements, and spatial requirements of intelligent driving vehicles.

[0017] The traffic performance requirements include the efficiency requirements, energy-saving requirements, and coordination requirements of the traffic system.

[0018] Furthermore, the security model corresponding to the security requirement is represented as follows:

[0019]

[0020]

[0021]

[0022] in, This represents the remaining time difference between vehicle p and vehicle q passing through the target area at time T; and v represents the remaining time for vehicles p and q to pass through the target area at time T, respectively; p (T) and v q (T) represent the instantaneous velocities of vehicle p and vehicle q at time T, respectively; a p (τ) and a q (τ) represent the instantaneous accelerations of vehicle p and vehicle q at time τ, respectively; O p (T+T m ) and O q (T+T m ) represent the location coordinates of the target area; O p (t) and O q (t) represent the position coordinates of vehicle p and vehicle q at the current time, respectively; Indicates the solution in T m It is a function of the dependent variable;

[0023] The comfort model corresponding to the aforementioned comfort requirements is represented as follows:

[0024]

[0025]

[0026] in, and These represent the cumulative squared amplitudes of adjacent time-domain acceleration fluctuations of vehicles p and q, respectively, from the initial time to time T; a p (τ) and a q (τ) represents the instantaneous acceleration of vehicle p and vehicle q at time τ, respectively; a p (τ-1) and a q (τ-1) represents the instantaneous acceleration of vehicle p and vehicle q at time τ-1, respectively;

[0027] The spatial model corresponding to the spatial requirements is represented as follows:

[0028]

[0029] O p (T)=[X p (T),Y p (T)],O q(T)=[X q (T),Y q (T)]

[0030] D p (Θ)=D p (α,β,γ;T),D q (Ξ)=D q (δ,η,θ;T)

[0031] Among them, O p (T) and O q (T) represent the position coordinates of vehicle p and vehicle q at time T, respectively; [X p (T),Y p (T)] and [X q (T),Y q [(T)] represents the horizontal and vertical coordinates of vehicle p and vehicle q at time T, respectively; D p (Θ) and D q (Ξ) represents the psychological acceptance of vehicle p and vehicle q, respectively; α, β, γ and δ, η, θ represent factors related to the driving style of vehicle p and vehicle q, respectively, aiming to distinguish the differences in the psychological acceptance of the surrounding drivable space by different driving styles.

[0032] Furthermore, the efficiency model corresponding to the efficiency requirement is expressed as follows:

[0033]

[0034] in, and These represent the remaining time for vehicles p and q to pass through the target area at time T, respectively.

[0035] The energy-saving model corresponding to the energy-saving requirement is represented as follows:

[0036]

[0037] Where, N c (a p (T)|a p (T-1)) and N c (a q (T)|a q (T-1)) represent the number of acceleration / deceleration mode transitions in the game decision outputs of vehicles p and q at adjacent time points; Represents an energy consumption model; a p (T) and a q (T) represent the instantaneous accelerations of vehicle p and vehicle q at time T, respectively; a p (T-1) and a q(T-1) represents the instantaneous acceleration of vehicle p and vehicle q at time T-1, respectively; v p (T) and v q (T) represent the instantaneous velocities of vehicle p and vehicle q at time T, respectively;

[0038] The collaborative model corresponding to the collaborative requirement is represented as follows:

[0039]

[0040] Among them, P e and S g R represents the duration of the game interaction and the number of stages in the stage game evolution, respectively; r This indicates the conditions under which each vehicle resolves its risk and exits the game interaction process.

[0041] Furthermore, in S3, the expected utility function, equality constraints, and inequality constraints of the preceding vehicle are expressed as follows:

[0042]

[0043]

[0044]

[0045] V1(a p ,a q ) = a p -a max ≤0

[0046] V2(a p ,a q ) = v p -v max ≤0

[0047]

[0048] V4(a p ,a q ) = a p -a min ≥0

[0049] V5(a p ,a q ) = v p -v min ≥0

[0050] in, Describe the expected utility function of the preceding vehicle; a p a represents the instantaneous acceleration of vehicle p at the current moment; q This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle p and vehicle q as they pass through the target area. This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle p; The distance between vehicles p and q represents the degree of psychological acceptance of their integration; ε pq (a p ,a q E represents a traffic efficiency model for a traffic system including vehicles p and q; pq (a p ,a q This represents an energy-saving model of a transportation system that includes vehicles p and q. This represents a collaborative model of a transportation system that includes vehicles p and q. Let U1 represent the Nth additional performance function involved in the extended game decision model of the intelligent driving vehicle; U1 and U2 both represent the equality constraints required for the kinematics of the preceding vehicle; V1, V2, V3, V4, and V5 all represent the inequality constraints required for the kinematics of the preceding vehicle; O p (T) represents the position coordinates of vehicle p at time T; O p (T-1) represents the position coordinates of vehicle p at time T-1; v p (T) represents the instantaneous velocity of vehicle p at time T; v p (T-1) represents the instantaneous velocity of vehicle p at time T-1; a p (τ) represents the instantaneous acceleration of vehicle p at time τ; a p (T) represents the instantaneous acceleration of vehicle p at time T; a p (T-1) represents the instantaneous acceleration of vehicle p at time T-1; This represents the time interval between time T-1 and time T; a max Indicates the upper limit of acceleration; v min Indicates the lower limit of speed; j max Indicates the upper limit of jerk.

[0051] Furthermore, in S3, the expected utility function, equality constraints, and inequality constraints of the following vehicle are expressed as follows:

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in, The expected utility function for following vehicles; a p a represents the instantaneous acceleration of vehicle p at the current moment; q This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle q and vehicle p as they pass through the target area; This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle q; The vehicle spacing ε represents the degree of psychological acceptance of vehicle q and vehicle p; qp (a p ,a q E represents a traffic efficiency model for a traffic system including vehicles q and p; qp (a p ,a q This represents an energy-saving model of a transportation system that includes vehicles q and p. This represents a collaborative model of a transportation system that includes vehicles q and p. This represents the Nth additional performance function involved in the extended game decision-making model of intelligent driving vehicles; and Both represent the equality constraints required to follow the vehicle's kinematics; and Both represent the inequality constraints required to follow the vehicle's kinematics; O q (T) represents the position coordinates of vehicle q at time T; O q (T-1) represents the position coordinates of vehicle q at time T-1; v q (T) represents the instantaneous velocity q of vehicle at time T; v q (T-1) represents the instantaneous velocity of vehicle q at time T-1; a q (τ) represents the instantaneous acceleration of vehicle q at time τ; a q (T) represents the instantaneous acceleration of vehicle q at time T; a q (T-1) represents the instantaneous acceleration of vehicle q at time T-1; This represents the time interval between time T-1 and time T; a max Indicates the upper limit of acceleration; v min Indicates the lower limit of speed; j max Indicates the upper limit of jerk.

[0061] Furthermore, S4 specifically includes:

[0062] S41. Treat the leading vehicles as the upper-level space of the hierarchical optimization task, and the following vehicles as the lower-level space of the hierarchical optimization task.

[0063] S42. Based on the expected utility functions, equality constraints, and inequality constraints of the leading and following vehicles, obtain the optimality conditions that satisfy the upper and lower spaces.

[0064] S43. Based on the aforementioned optimality conditions, the expected utility functions, equality constraints, and inequality constraints of the leading vehicle and the following vehicle are respectively transformed into the corresponding augmented expected utility functions of the upper space and the lower space.

[0065] S44. Based on the augmented expected utility function of the upper space and the augmented expected utility function of the lower space, obtain the acceleration strategy of each car participating in the game at time T.

[0066] Furthermore, in S42, the optimality conditions for the upper and lower spaces are as follows:

[0067]

[0068]

[0069]

[0070]

[0071] in, express For a p,q The gradient; The function representing the expected utility function of the leading vehicle in an extended game and the regularization of its equality and inequality constraints. express For a p,q The gradient; A function representing the expected utility function of a following vehicle in an extended game, and regularized by its equality and inequality constraints; a p,q This represents the combination of the acceleration of the leading vehicle and the acceleration of the following vehicle; i represents one type of inequality constraint; j represents another type of inequality constraint. This represents the acceleration of the preceding vehicle in the optimality condition. This represents the following vehicle acceleration oriented towards the optimality condition; The gradient represents the i-th type of inequality constraint for the preceding vehicle; The gradient represents the j-type inequality constraint of the preceding vehicle; The gradient of the equality constraint for the preceding vehicle; α i+jV represents the multiplier of type i and type j inequality constraints for the preceding vehicle; i+j Represents the type i and type j inequality constraints for the preceding vehicle; This represents the gradient of the i-th type inequality constraint following the vehicle; This represents the gradient of the j-type inequality constraint following the vehicle; The gradient of the equality constraint for the following vehicle; π i+j The multiplier represents the multiplier for the i-type and j-type inequality constraints following the vehicle; Represent the type i and type j inequality constraints for the following vehicle; if there exists a multiplier α i+j ,β k ,π i+j ,ρ k If the optimality condition formula holds, then the optimality conditions of the upper and lower spaces are satisfied.

[0072] Furthermore, S44 specifically includes:

[0073] S441. Find the partial derivative of the augmented expected utility function of the lower space with respect to the acceleration of the following vehicle; set the partial derivative to zero to obtain the characterization equation of the acceleration of the following vehicle.

[0074] S442. Find the partial derivative of the augmented expected utility function of the upper space with respect to the acceleration of the preceding vehicle; set the partial derivative to zero to obtain the characterization equation of the acceleration of the preceding vehicle.

[0075] S443. Combine the characterization equations of the following vehicle's acceleration and the leading vehicle's acceleration into a system of equations; solve the system of equations to obtain the acceleration strategy output by the leading vehicle at time T.

[0076] S444. Substitute the acceleration strategy output by the leading vehicle at time T into the characterization equation of the following vehicle's acceleration to obtain the acceleration strategies output by each car participating in the game at time T.

[0077] Compared with existing technologies, the intelligent driving continuous domain decision-making hierarchical space optimization method for extended game theory described in this invention has the following beneficial effects: the method fully considers the hierarchical characteristics of extended game theory, improves the interactivity of multi-vehicle decision-making, overcomes the problem of coarse solutions in the discrete domain, and further improves the accuracy of optimization solutions.

[0078] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0079] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0080] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0081] Figure 1 This is a schematic diagram of the process for optimizing the decision hierarchy space of intelligent driving in the continuous domain for extended game theory, as provided in an embodiment of the present invention.

[0082] Figure 2 This is a schematic diagram illustrating the spatial characteristics of the decision hierarchy in an extended game, as provided in an embodiment of the present invention.

[0083] Figure 3 This is a schematic diagram of the hierarchical space optimization method provided in an embodiment of the present invention. Detailed Implementation

[0084] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0085] See Figure 1 As shown, this embodiment of the invention provides a hierarchical space optimization method for decision-making in the continuous domain of intelligent driving oriented towards extended game theory, specifically including the following steps:

[0086] S1. Intelligent driving vehicles acquire and share information with surrounding traffic vehicles and the traffic environment;

[0087] S2. Based on the acquired information, establish an extended game decision-making model for intelligent driving vehicles within a continuous action domain;

[0088] S3. Each car participating in the game predicts the reactions of other cars to its own driving behavior and obtains the expected utility functions, equality constraints and inequality constraints of the leading car and the following car respectively.

[0089] S4. Based on the expected utility functions, equality constraints, and inequality constraints of the leading and following vehicles, solve the extended game decision model of the intelligent driving vehicle using the hierarchical space optimization method, and output the acceleration strategy of each vehicle participating in the game at time T.

[0090] S5. The intelligent driving vehicle executes its own acceleration strategy at time T;

[0091] S6. At time T+1, update the information of the intelligent driving vehicle, surrounding traffic vehicles and traffic environment, and each car participating in the game re-predicts the reactions of other cars to its own driving behavior.

[0092] S7. Repeat steps S1-S6 until each vehicle resolves its risk and exits the game interaction process.

[0093] The following sections will provide a detailed explanation of each of the above steps.

[0094] In step S1 above, the intelligent driving vehicle uses onboard sensors and vehicle-to-vehicle communication equipment to perceive, acquire, and share information with surrounding vehicles and the traffic environment. In cases of incomplete information, the intelligent driving vehicle utilizes historical traffic environment information, including surrounding vehicles, combined with its understanding and perception of the surrounding traffic and environment, to supplement the incomplete information and ensure its completeness.

[0095] In step S2 above, based on the information obtained in step S1, an extended game-theoretic decision-making model for intelligent driving vehicles is established within a continuous action domain. This model involves multi-objective vehicle driving needs and traffic performance needs embedded with driving styles. Multi-objective vehicle driving needs include, but are not limited to, the safety, comfort, and spatiality of intelligent driving vehicles; traffic performance needs include, but are not limited to, the efficiency, energy conservation, and coordination of the traffic system. The established extended game-theoretic decision-making model for intelligent driving vehicles allows the acceleration and deceleration behaviors and acceleration of intelligent driving vehicles to continuously take arbitrary values ​​within the upper and lower limits that conform to the vehicle's physical constraints.

[0096] The aforementioned safety requirements for vehicle driving are based on the incentive of the intelligent driving vehicle and surrounding traffic vehicles participating in the game interaction process each arbitrarily taking consecutive driving actions from their respective game strategy sets. The remaining time at time T under the specific actions of the vehicle and surrounding vehicles is calculated, and the difference in remaining time is used as a safety measure. In other words, the larger the remaining time difference, the higher the level of vehicle safety and the lower the probability of conflict; the smaller the remaining time difference, the lower the level of vehicle safety and the higher the probability of conflict. The safety model corresponding to this safety requirement is expressed as follows:

[0097]

[0098]

[0099]

[0100] in, This represents the remaining time difference between vehicle p and vehicle q passing through the target area at time T; and v represents the remaining time for vehicles p and q to pass through the target area at time T, respectively; p (T) and v q (T) represent the instantaneous velocities of vehicle p and vehicle q at time T, respectively; a p (τ) and a q (τ) represent the instantaneous accelerations of vehicle p and vehicle q at time τ, respectively; O p (T+T m ) and O q (T+T m ) represent the location coordinates of the target area; O p (T) and O q (T) represent the position coordinates of vehicle p and vehicle q at the current time, respectively; Indicates the solution in T m It is a function of the dependent variable.

[0101] The aforementioned vehicle driving comfort requirements aim to limit the unpleasant driving experience caused by frequent fluctuations in acceleration, allowing the vehicle to travel as smoothly and comfortably as possible. The corresponding comfort model is represented as follows:

[0102]

[0103]

[0104] in, and These represent the cumulative squared amplitudes of adjacent time-domain acceleration fluctuations of vehicles p and q, respectively, from the initial time to time T; a p (τ) and a q (τ represents the instantaneous acceleration of vehicle p and vehicle q at time τ, respectively; a) p (τ-1) and a q (τ-1) represents the instantaneous acceleration of vehicle p and vehicle q at time τ-1, respectively.

[0105] The aforementioned spatial requirements for vehicle driving specify the required distance between vehicles, simulating the psychological acceptance of surrounding drivable space by human drivers. The spatial model corresponding to these requirements is represented as follows:

[0106]

[0107] O p (T)=[X p (T),Y p (T)],O q (T)=[X q (T),Y q (T)]

[0108] D p (Θ)=D p (α,β,γ;T),D q (Ξ)=D q (δ,η,θ;T)

[0109] Among them, O p (T) and O q (T) represent the position coordinates of vehicle p and vehicle q at time T, respectively; [X p (T),Y p (T)] and [X q (T),Y q [(T)] represents the horizontal and vertical coordinates of vehicle p and vehicle q at time T, respectively; D p (Θ) and D q (Ξ) represents the psychological acceptance of vehicle p and vehicle q, respectively; α, β, γ and δ, η, θ represent factors related to the driving style of vehicle p and vehicle q, respectively, aiming to distinguish the differences in the psychological acceptance of the surrounding drivable space by different driving styles.

[0110] The aforementioned efficiency requirements for traffic performance aim to maximize traffic flow efficiency and fully utilize traffic capacity and saturation. The efficiency model corresponding to these requirements is expressed as follows:

[0111]

[0112] in, and Let T represent the remaining time for vehicles p and q to pass through the target area at time T.

[0113] The aforementioned energy-saving requirements for traffic performance, from the perspective of the overall transportation system, aim to reduce the system's overall energy consumption, including fuel and electricity, and improve driving economy and energy efficiency. The corresponding energy-saving model for these requirements is as follows:

[0114]

[0115] Where, N c (a p (T)|a p (T-1)) and N c (a q (T)|a q (T-1)) represent the number of acceleration / deceleration mode transitions in the game decision outputs of vehicles p and q at adjacent time points; This represents an energy consumption model, including but not limited to fuel consumption and electricity consumption; a p (T) and a q(T) represent the instantaneous accelerations of vehicle p and vehicle q at time T, respectively; a p (T-1) and a q (T-1) represents the instantaneous acceleration of vehicle p and vehicle q at time T-1, respectively; v p (T) and v q (T) represent the instantaneous velocities of vehicle p and vehicle q at time T, respectively. The model is affected by the acceleration / deceleration mode transitions and the instantaneous velocities of the vehicles, as well as by driving resistance and gear selection. The model can be refined according to the types and penetration rates of vehicles with different power sources in the traffic system.

[0116] The aforementioned coordination requirements for traffic performance are related to the duration of interaction among various vehicles and the number of stages in the phased game evolution. The corresponding coordination model for this requirement is expressed as follows:

[0117]

[0118] Among them, P e and S g R represents the duration of the game interaction and the number of stages in the stage game evolution, respectively; r This indicates the conditions under which each vehicle resolves its risk and exits the game interaction process.

[0119] The extended game theory decision-making model for intelligent driving vehicles includes, but is not limited to, the multi-objective vehicle driving requirements and traffic performance requirements mentioned above. Other performance functions can also be incorporated into the modeling process, denoted as...

[0120] In step S3 above, the expected utility function, equality constraints, and inequality constraints of the leading vehicle in the extended game can be integrated as follows:

[0121]

[0122]

[0123]

[0124] V1(a p ,a q ) = a p -a max ≤0

[0125] V2(a p ,a q ) = v p -v max ≤0

[0126]

[0127] V4(ap ,a q ) = a p -a min ≥0

[0128] V5(a p ,a q ) = v p -v min ≥0

[0129] in, Describe the expected utility function of the preceding vehicle; a p a represents the instantaneous acceleration of vehicle p at the current moment; q This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle p and vehicle q as they pass through the target area. This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle p; The distance between vehicles p and q represents the degree of psychological acceptance of their integration; ε pq (a p ,a q E represents a traffic efficiency model for a traffic system including vehicles p and q; pq (a p ,a q This represents an energy-saving model of a transportation system that includes vehicles p and q. This represents a collaborative model of a transportation system that includes vehicles p and q. Let U1 represent the Nth additional performance function involved in the extended game decision model of the intelligent driving vehicle; U1 and U2 both represent the equality constraints required for the kinematics of the preceding vehicle; V1, V2, V3, V4, and V5 all represent the inequality constraints required for the kinematics of the preceding vehicle; O p (T) represents the position coordinates of vehicle p at time T; O p (T-1) represents the position coordinates of vehicle p at time T-1; v p (T) represents the instantaneous velocity of vehicle p at time T; v p (T-1) represents the instantaneous velocity of vehicle p at time T-1; a p (τ) represents the instantaneous acceleration of vehicle p at time τ; a p (T) represents the instantaneous acceleration of vehicle p at time T; a p (T-1) represents the instantaneous acceleration of vehicle p at time T-1; This represents the time interval between time T-1 and time T; a max Indicates the upper limit of acceleration; v min Indicates the lower limit of speed; j max Indicates the upper limit of jerk.

[0130] In extended game theory, the expected utility function, equality constraints, and inequality constraints for following vehicles can be integrated as follows:

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139] in, The expected utility function for following vehicles; a p a represents the instantaneous acceleration of vehicle p at the current moment; q This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle q and vehicle p as they pass through the target area; This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle q; The vehicle spacing ε represents the degree of psychological acceptance of vehicle q and vehicle p; qp (a p ,a q E represents a traffic efficiency model for a traffic system including vehicles q and p; qp (a p ,a q This represents an energy-saving model of a transportation system that includes vehicles q and p. This represents a collaborative model of a transportation system that includes vehicles q and p. This represents the Nth additional performance function involved in the extended game decision-making model of intelligent driving vehicles; and Both represent the equality constraints required to follow the vehicle's kinematics; and Both represent the inequality constraints required to follow the vehicle's kinematics; O q (T) represents the position coordinates of vehicle q at time T; O q (T-1) represents the position coordinates of vehicle q at time T-1; v q (T) represents the instantaneous velocity q of vehicle at time T; v q(T-1) represents the instantaneous velocity of vehicle q at time T-1; a q (τ) represents the instantaneous acceleration of vehicle q at time τ; a q (T) represents the instantaneous acceleration of vehicle q at time T; a q (T-1) represents the instantaneous acceleration of vehicle q at time T-1; This represents the time interval between time T-1 and time T; a max Indicates the upper limit of acceleration; v min Indicates the lower limit of speed; j max Indicates the upper limit of jerk;

[0140] In the above content, and This is reflected in the expected utility functions of different leading and following vehicles, respectively. This indicates the specific manifestation of safety requirements in the expected utility function of the preceding vehicle. This indicates the specific manifestation of safety requirements in the expected utility function of following vehicles.

[0141] In step S4 above, see Figure 2 As shown, the lead vehicle, acting proactively, is considered the upper-level space of the hierarchical optimization task, while the follower vehicles, acting reactively, are considered the lower-level space. Both upper and lower spaces satisfy their respective constraint criteria. The upper level must make optimal decisions based on accurate predictions of the future behavior of the lower level. The lower level reactively optimizes its equations by following the optimal decisions made by the upper level. This approach is applicable to high-dimensional problems and can be extended to multi-agent game systems. Using the hierarchical space optimization method proposed in this invention to solve the extended game decision model of intelligent driving vehicles in a continuous action domain, the acceleration strategies that each participating vehicle should adopt at time T can be output.

[0142] For details, please refer to [link / details]. Figure 3 As shown. The expected utility functions of the leading and following vehicles, along with their equality and inequality constraints, are rewritten as follows:

[0143]

[0144]

[0145] in, The function representing the expected utility function of the leading vehicle in an extended game and the regularization of its equality and inequality constraints. α represents the expected utility function of the following vehicle in an extended game, and a function regularized by its equality and inequality constraints; i V represents the multiplier of type i inequality constraints for the preceding vehicle; i Represents the type i inequality constraint for the preceding vehicle; αj V represents the multiplier of the j-type inequality constraint for the preceding vehicle; j Represents the j-type inequality constraint for the preceding vehicle; β k U represents the multiplier of the equality constraint for the preceding vehicle; k Represents the equality constraints for the preceding vehicle; π i Represents the multiplier of type i inequality constraints following the vehicle; Represents the type i inequality constraint for the following vehicle; π j Represents the multiplier of the j-type inequality constraint following the vehicle; Represents the j-type inequality constraint for the following vehicle; ρ k The multiplier representing the equality constraint following the vehicle; The equation constraint represents the following vehicle;

[0146] Calculate the following formulas respectively:

[0147]

[0148]

[0149]

[0150]

[0151] in, and They represent and For a p,q The gradient; a p,q This represents the combination of the acceleration of the leading vehicle and the acceleration of the following vehicle; i represents one type of inequality constraint; j represents another type of inequality constraint. This represents the acceleration of the preceding vehicle in the optimality condition. This represents the following vehicle acceleration oriented towards the optimality condition; The gradient represents the i-th type of inequality constraint for the preceding vehicle; The gradient represents the j-type inequality constraint of the preceding vehicle; The gradient of the equality constraint for the preceding vehicle; α i+j V represents the multiplier of type i and type j inequality constraints for the preceding vehicle; i+j Represents the type i and type j inequality constraints for the preceding vehicle; This represents the gradient of the i-th type inequality constraint following the vehicle; This represents the gradient of the j-type inequality constraint following the vehicle; The gradient of the equality constraint for the following vehicle; π i+jThe multiplier represents the multiplier for the i-type and j-type inequality constraints following the vehicle; Represent the type i and type j inequality constraints for the following vehicle; if there exists a multiplier α i+j ,β k ,π i+j ,ρ k If the above equation holds true, then the optimality conditions of the upper and lower spaces are satisfied.

[0152] Under the optimality conditions of the upper and lower spaces described above, this embodiment of the invention comprehensively transforms the expected utility functions of the leading and following vehicles, along with their equality and inequality constraints, into augmented expected utility functions using two different methods; wherein:

[0153] The augmented expected utility function of the upper space is achieved by using an auxiliary function. It is constructed, specifically represented as:

[0154]

[0155]

[0156] in, This represents an augmented constraint function, used to integrate equality and inequality constraints through specific logical operations, while penalizing violations of the constraints.

[0157] The augmented expected utility function of the lower space is achieved by using three auxiliary functions. and It is constructed, specifically represented as:

[0158]

[0159]

[0160]

[0161]

[0162] In this invention, the embodiments fully consider the following vehicles in the lower space. The cumulative effect of their optimal acceleration strategy error can induce the optimal acceleration strategies obtained by the leading and following vehicles in the extended game to deviate from the steady-state subgame perfect Nash equilibrium. This invention proposes four ways to construct auxiliary functions, namely, in the above formula... and The appropriate option can be chosen by weighing the specific refined expected utility function and constraints.

[0163] For the augmented expected utility functions of the upper and lower spaces after the above transformation, the acceleration strategies output by each participating vehicle at time T are obtained through the following calculation process. First, the partial derivative of the augmented expected utility function of the lower space with respect to the acceleration of the following vehicle is calculated. Right now Setting this partial derivative to zero yields the characterization equation for the following vehicle's acceleration. Then, the partial derivative of the augmented expected utility function in the upper space with respect to the preceding vehicle's acceleration is calculated. Right now Setting the partial derivative to zero yields the characterization equation for the acceleration of the leading vehicle; combining the characterization equations for the acceleration of the following vehicle and the leading vehicle's acceleration into a system of equations; solving this system of equations yields the acceleration strategy 'a' output by the leading vehicle at time T. p * (T). Finally, substituting the acceleration strategy output by the leading vehicle at time T into the characterization equation of the following vehicle's acceleration, we can obtain the acceleration strategies (a) output by all participating vehicles at time T. p * (T),a q * (T)) can solve the problem of solving the extended game decision-making model of intelligent driving vehicles with continuous domain actions mentioned above.

[0164] In step S5 above, the chassis controller of the intelligent driving vehicle executes the acceleration strategy of the intelligent driving vehicle at time T, which is based on the output of the top-level driving decision system.

[0165] In step S6 above, at time T+1, the information of the intelligent driving vehicle, surrounding traffic vehicles and traffic environment is updated, and each car participating in the game re-predicts the reactions of other cars to its own driving behavior.

[0166] In step S7 above, all the above steps are repeated. By rolling game decision-making and optimization solution, the subgame perfect Nash equilibrium of the game in each time domain stage is repeatedly calculated to obtain the stable acceleration strategy in each time domain until each vehicle eliminates the risk and exits the game interaction process.

[0167] To effectively address the hierarchical characteristics of sequential driving behavior in extended game theory and solve the problem of coarse solutions in the discrete domain, this invention extends to the general case of continuous domain decision-making. Aiming at the challenging problem of optimizing driving decisions in hierarchical space, it proposes a hierarchical space optimization method for intelligent driving continuous domain decision-making in extended game theory. This method addresses the hierarchical challenges and optimization difficulties of extended game models in driving scenarios involving strong interactions among multiple agents. It effectively improves the inclusiveness of intelligent driving decision-making methods towards heterogeneous, complementary, mutually exclusive, and personalized driving performance optimization objectives, ensuring the accuracy of interactive decision results. This, in turn, enhances the optimality of decisions and the stability of the game, fully leveraging the advantages of high-precision control by machine intelligence to achieve highly safe, robust, and acceptable intelligent driving continuous domain decision-making.

[0168] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A hierarchical space optimization method for decision-making in the continuous domain of intelligent driving, oriented towards extended game theory, characterized in that... include: S1. Intelligent driving vehicles acquire and share information with surrounding traffic vehicles and the traffic environment; S2. Based on the acquired information, establish an extended game decision-making model for intelligent driving vehicles within a continuous action domain. S3. Each car participating in the game predicts the reactions of other cars to its own driving behavior and obtains the expected utility functions, equality constraints and inequality constraints of the leading car and the following car respectively. S4. Based on the expected utility functions, equality constraints, and inequality constraints of the leading and following vehicles, solve the extended game decision model of the intelligent driving vehicle using the hierarchical space optimization method, and output the acceleration strategy of each vehicle participating in the game at time T. S5. The intelligent driving vehicle executes its own acceleration strategy at time T; S6. At time T+1, update the information of the intelligent driving vehicle, surrounding traffic vehicles and traffic environment, and each car participating in the game re-predicts the reactions of other cars to its own driving behavior. S7. Repeat steps S1-S6 until each vehicle eliminates the risk and exits the game interaction process. S4 specifically includes: S41. Treat the leading vehicles as the upper-level space of the hierarchical optimization task, and the following vehicles as the lower-level space of the hierarchical optimization task. S42. Based on the expected utility functions, equality constraints, and inequality constraints of the leading and following vehicles, obtain the optimality conditions that satisfy the upper and lower spaces. S43. Based on the aforementioned optimality conditions, the expected utility functions, equality constraints, and inequality constraints of the leading vehicle and the following vehicle are respectively transformed into the corresponding augmented expected utility functions of the upper space and the lower space. S44. Based on the augmented expected utility function of the upper space and the augmented expected utility function of the lower space, obtain the acceleration strategy of each car participating in the game at time T.

2. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 1, characterized in that, The S1 further includes: when the acquired information is incomplete, the intelligent driving vehicle uses the historical information of the traffic environment collected, including surrounding traffic vehicles, and combines it with its understanding and cognition of the surrounding traffic vehicles and traffic environment to supplement the incomplete information.

3. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 1, characterized in that, In S2, the intelligent driving vehicle extended game decision model involves multi-objective vehicle driving needs and traffic performance needs embedded with driving styles; The multi-objective vehicle driving requirements include the safety requirements, comfort requirements, and spatial requirements of intelligent driving vehicles. The traffic performance requirements include the efficiency requirements, energy-saving requirements, and coordination requirements of the traffic system.

4. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 3, characterized in that: The security model corresponding to the security requirements is represented as follows: in, Indicates the vehicle at time T and vehicles Using the remaining time difference in the target area; and They represent the vehicles at time T, respectively. and vehicles Remaining time to reach the target area; and They represent the vehicles at time T, respectively. and vehicles Instantaneous velocity; and They represent Time vehicle and vehicles Instantaneous acceleration; and These represent the location coordinates of the target area; and These represent the vehicles at the current time. and vehicles Position coordinates; Indicates the solution to It is a function of the dependent variable; The comfort model corresponding to the aforementioned comfort requirements is represented as follows: in, and These represent the vehicles from the start time to time T. and vehicles The cumulative sum of the squared amplitudes of adjacent time-domain acceleration fluctuations; and They represent Time vehicle and vehicles Instantaneous acceleration; and They represent Time vehicle and vehicles Instantaneous acceleration; The spatial model corresponding to the spatial requirements is represented as follows: in, and They represent the vehicles at time T, respectively. and vehicles Position coordinates; and They represent the vehicles at time T, respectively. and vehicles The horizontal and vertical coordinates; and Representing vehicles and vehicles The level of psychological acceptance; and Respectively represent vehicles and vehicles Driving style-related factors are designed to differentiate the differences in psychological acceptance of the surrounding drivable space among different driving styles.

5. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 3, characterized in that: The efficiency model corresponding to the efficiency requirement is represented as follows: in, and They represent the vehicles at time T, respectively. and vehicles Remaining time to reach the target area; The energy-saving model corresponding to the energy-saving requirement is represented as follows: + in, and These represent vehicles at adjacent times. and vehicles The number of acceleration / deceleration mode transitions in the game decision output; Represents an energy consumption model; and Representing vehicles and vehicles The instantaneous acceleration at time T; and Representing vehicles and vehicles Instantaneous acceleration at time T-1; and Representing vehicles and vehicles The instantaneous velocity at time T; The collaborative model corresponding to the collaborative requirement is represented as follows: in, and These represent the duration of the game interaction and the number of stages in the game evolution, respectively. This indicates the conditions under which each vehicle resolves its risk and exits the game interaction process.

6. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 1, characterized in that, In S3, the expected utility function, equality constraints, and inequality constraints of the preceding vehicle are expressed as follows: ) in, This represents the expected utility function of the preceding vehicle; This represents the instantaneous acceleration of vehicle p at the current moment; This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle p and vehicle q as they pass through the target area. This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle p; The distance between vehicles represents the degree of psychological acceptance of vehicle p and vehicle q. This represents a traffic efficiency model for a traffic system that includes vehicles p and q. This represents an energy-saving model for a transportation system that includes vehicle p and vehicle q. This represents a collaborative model of a transportation system that includes vehicles p and q. This represents the Nth additional performance function involved in the extended game decision-making model of intelligent driving vehicles; and All of these represent the equality constraints required for the kinematics of the preceding vehicle; and All of these represent the inequality constraints required for the kinematics of the preceding vehicle; This represents the position coordinates of vehicle p at time T; This represents the position coordinates of vehicle p at time T-1; Let represent the instantaneous velocity of vehicle p at time T; Let represent the instantaneous velocity of vehicle p at time T-1; express The instantaneous acceleration of vehicle p at time p; Let represent the instantaneous acceleration of vehicle p at time T; Let represent the instantaneous acceleration of vehicle p at time T-1; This represents the time interval between time T-1 and time T; Indicates the upper limit of acceleration; Indicates the lower limit of speed; Indicates the upper limit of jerk.

7. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 1, characterized in that, In S3, the expected utility function, equality constraints, and inequality constraints of the following vehicle are expressed as follows: ) in, The expected utility function for following vehicles; This represents the instantaneous acceleration of vehicle p at the current moment; This represents the instantaneous acceleration of vehicle q at the current moment; This represents the remaining time difference between vehicle q and vehicle p as they pass through the target area; This represents the cumulative sum of the squared amplitudes of the adjacent time-domain acceleration fluctuations of vehicle q; The distance between vehicles represents the degree of psychological acceptance of the integration of vehicles q and p; This represents a traffic efficiency model for a traffic system that includes vehicles q and p. This represents an energy-saving model for a transportation system that includes vehicle q and vehicle p. This represents a collaborative model of a transportation system that includes vehicles q and p. This represents the Nth additional performance function involved in the extended game decision-making model of intelligent driving vehicles; and Both represent the equation constraints required to follow the vehicle's kinematics; and All of these represent the inequality constraints required to follow the vehicle's kinematics. This represents the position coordinates of vehicle q at time T; This represents the position coordinates of vehicle q at time T-1; Let q represent the instantaneous velocity of vehicle q at time T; Let q represent the instantaneous velocity of vehicle q at time T-1; express The instantaneous acceleration of the vehicle at time q; Let q represent the instantaneous acceleration of vehicle q at time T; Let q represent the instantaneous acceleration of vehicle q at time T-1; This represents the time interval between time T-1 and time T; Indicates the upper limit of acceleration; Indicates the lower limit of speed; Indicates the upper limit of jerk.

8. The method for hierarchical space optimization of decision-making in the continuous domain for extended game theory in intelligent driving as described in claim 1, characterized in that, In S42, the optimality conditions for the upper and lower spaces are as follows: in, express right The gradient; The function representing the expected utility function of the leading vehicle in an extended game and the regularization of its equality and inequality constraints. express right The gradient; The function representing the expected utility function of the following vehicle in an extended game, and the function regularized by its equality and inequality constraints. This represents the combination of the acceleration of the leading vehicle and the acceleration of the following vehicle; i represents one type of inequality constraint; j represents another type of inequality constraint. This represents the acceleration of the preceding vehicle in relation to the optimality condition. This represents the following vehicle acceleration oriented towards the optimality condition; The gradient represents the i-th type of inequality constraint for the preceding vehicle; The gradient represents the j-type inequality constraint of the preceding vehicle; This represents the gradient of the equality constraint for the preceding vehicle; The multiplier represents the multiplier for the i-type and j-type inequality constraints of the preceding vehicle; Represents the type i and type j inequality constraints for the preceding vehicle; This represents the gradient of the i-th type inequality constraint following the vehicle; This represents the gradient of the j-type inequality constraint following the vehicle; This represents the gradient of the equality constraint following the vehicle. The multiplier represents the multiplier for the i-type and j-type inequality constraints following the vehicle; Represent the type i and type j inequality constraints for the following vehicle; if multipliers exist If the optimality condition formula holds, then the optimality conditions of the upper and lower spaces are satisfied.

9. The method for hierarchical space optimization of decision-making in the continuous domain for intelligent driving oriented towards extended game theory as described in claim 1, characterized in that, S44 specifically includes: S441. Find the partial derivative of the augmented expected utility function of the lower space with respect to the acceleration of the following vehicle; set the partial derivative to zero to obtain the characterization equation of the acceleration of the following vehicle. S442. Find the partial derivative of the augmented expected utility function of the upper space with respect to the acceleration of the preceding vehicle; set the partial derivative to zero to obtain the characterization equation of the acceleration of the preceding vehicle. S443. Combine the characterization equations of the following vehicle's acceleration and the characterization equations of the leading vehicle's acceleration into a system of equations; solve the system of equations to obtain the acceleration strategy output by the leading vehicle at time T. S444. Substitute the acceleration strategy output by the leading vehicle at time T into the characterization equation of the following vehicle's acceleration to obtain the acceleration strategies output by each car participating in the game at time T.

Images