Multi-task multi-target optimization decision planning processing method and system in autonomous driving

Abstract

The application discloses a kind of multi-task multi-target optimization decision planning processing methods in automatic driving, including steps: in the process of automatic driving, real-time acquisition car information, target vehicle information and driving environment information;Determine the target decision domain corresponding to vehicle in desired time;When determining that decision domain switching needs to be carried out, the control amount corresponding to target decision domain is obtained according to the optimization controller preset in target decision domain;Soft switching control strategy of dynamic buffer field is used, and the control amount corresponding to target decision domain is used to control vehicle, and vehicle is switched from the driving mode corresponding to current decision domain to the driving mode corresponding to target decision domain.The application also discloses corresponding system.Implementation of the present application can realize the optimization decision planning under complex working conditions such as unprotected intersection, and improve the success rate of decision and driving safety.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method and system for multi-task, multi-objective optimization decision planning in autonomous driving. Background Technology

[0002] Decision-making and planning during vehicle operation is a core component of autonomous driving (such as driverless driving), directly impacting vehicle safety, economy, and comfort. Commonly used decision-making and planning solutions in existing technologies can be categorized as follows: rule-based driving task decision-making, random sampling-based decision-making and planning, machine learning-based intelligent decision-making and planning, and optimization theory-based decision-making and planning.

[0003] As the level of intelligence increases, autonomous vehicles need to perform more diverse driving tasks, adapt to a wider range of driving conditions, and consider more performance requirements. Especially at unprotected intersections, the road topology is complex, traffic dynamics are constantly changing, and the behavior trajectories of traffic participants are random and involve adversarial game dynamics, thus presenting the challenge of optimizing driving decisions for multiple tasks and objectives. Particularly at unprotected intersections, the multi-task nature of driving leads to a game-like dynamic between exclusive driving functions (lane changing, lane keeping) and exacerbates the dynamic coupling of coexisting functions (lane keeping, cruise control). Furthermore, changes in driving tasks can cause abrupt changes in the dimension, order, and parameters of the decision-making and planning objects. On the other hand, different performance requirements are constrained (power, economy, safety, comfort) and dynamically change with traffic conditions. Moreover, the rigid (double yellow lane lines, red lights, etc.) and flexible (white dashed lane lines, etc.) constraints of traffic rules on the driving process in terms of operating area and motion trends make real-time optimization decision-making and planning for diverse performance requirements under integrated driving tasks extremely difficult. The existing decision-making and planning methods mentioned above cannot adequately address these problems.

[0004] Therefore, how to carry out autonomous driving in complex traffic scenarios is a bottleneck and difficult problem that urgently needs to be solved. The currently commonly used method of manually designing scheduling strategies to decompose the driving decision-making process is no longer adequate for complex traffic environments. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a multi-task, multi-objective optimization decision-making and planning processing method and system for autonomous driving, which can realize optimization decision-making and planning in complex working conditions such as unprotected intersections, thereby improving the success rate of decision-making and driving safety.

[0006] To address the aforementioned technical problems, as one aspect of the present invention, a multi-task, multi-objective optimization decision-making and planning processing method for autonomous driving is provided, applied to an autonomous driving system, and includes at least the following steps:

[0007] During autonomous driving, real-time information is acquired about the vehicle's own driving conditions, the target vehicle's information, and the driving environment.

[0008] Determine the target decision domain of the vehicle within the expected time frame based on the information obtained;

[0009] Whether a decision domain switch is needed is determined based on the priority of the target decision domain and the current decision domain. When a decision domain switch is needed, the control quantity corresponding to the target decision domain is calculated based on the optimization controller preset in the target decision domain.

[0010] A soft-switching control strategy with a dynamic buffer domain is adopted, which uses the control quantity corresponding to the target decision domain to control the vehicle and switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain.

[0011] The vehicle's driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed; the target vehicle information includes at least the relative distance and relative speed between each target vehicle and the vehicle; the driving environment information includes at least lane markings, wind speed, and gradient; and the decision domain includes, in descending order of priority, the dynamic stability domain, the collision safety domain, the regulatory domain, and the comfort domain.

[0012] The step of determining whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain is as follows:

[0013] When the priority of the target decision domain is higher than the priority of the current decision domain, it is determined that a decision domain switch is required.

[0014] The step of calculating the control quantity corresponding to the target decision domain based on the pre-set optimization controller of the target decision domain is as follows:

[0015] The control quantity u of the target decision domain is calculated using the following formula. j :

[0016]

[0017] Where, x i =[v x v y ,ω] T The state variables include longitudinal vehicle speed, lateral vehicle speed, and yaw rate; P is a pre-obtained diagonalized Lyapunov function, and n is the number of targets; u j The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle.

[0018] The soft-switching control strategy employing a dynamic buffer domain, which uses the control quantity corresponding to the target decision domain to control the vehicle and switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain, specifically involves the following steps:

[0019] Establish a buffer zone between the current decision domain and the target decision domain;

[0020] Obtain the current decision domain control variable u k The control quantity for the current buffer domain output is calculated periodically using the following formula:

[0021]

[0022] Among them, u * It is the control quantity output by the buffer domain, u j It is the control quantity of the decision domain, and α is the weighting coefficient, which is determined according to the switching completion degree of the decision domain;

[0023] The vehicle is controlled using the control quantity output from the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

[0024] This further includes the step of pre-obtaining the optimization controller corresponding to each decision domain, including:

[0025] Based on the time domain and state space covered by driving needs, driving needs are divided into global needs, dynamic-related real-time needs, and trend indicator needs.

[0026] Obtain the characterization indicators and constraints for each type of driving demand, and establish different decision domains;

[0027] For each decision domain, constraint equations between variables are constructed based on driving requirements. Then, the vehicle dynamics state equations and the relative motion state equations between the vehicle and the road and between the vehicle and the target vehicle are associated with the vehicle speed and yaw rate. Different decision domains are mapped to a three-dimensional state space consisting of yaw rate, longitudinal acceleration and vehicle speed.

[0028] By establishing an affine nonlinear state-space model that considers uncertainty, the state variables and first-order information of the state variables corresponding to each decision domain are calculated. The uncertainty includes modeling error, nonlinear characteristics and external disturbance factors.

[0029] Hamilton-Jacobi inequalities are established based on the state variables and first-order information of the state variables corresponding to each decision domain to construct optimal control problems for different decision domains. The corresponding diagonalized Lyapunov functions are obtained by solving them through offline numerical methods.

[0030] Based on the diagonalized Lyapunov function, the calculation formulas for the control quantities of each decision domain are obtained, forming an optimized controller for each decision domain.

[0031] Accordingly, another aspect of the present invention also provides a multi-task, multi-objective optimization decision-making and planning processing system for autonomous driving, applied in an autonomous driving system, comprising at least:

[0032] The information acquisition unit is used to acquire real-time driving information of the vehicle itself, target vehicle information, and driving environment information during the autonomous driving process.

[0033] The target decision domain determination unit is used to determine the target decision domain corresponding to the vehicle within the expected time based on the acquired information.

[0034] The decision processing unit is used to determine whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain. When a decision domain switch is needed, the unit calculates the control quantity corresponding to the target decision domain based on the preset optimization controller of the target decision domain.

[0035] The switching processing unit is used to employ a soft switching control strategy with a dynamic buffer domain, and to control the vehicle using the control quantity corresponding to the target decision domain, switching the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain.

[0036] The vehicle's driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed; the target vehicle information includes at least the relative distance and relative speed between each target vehicle and the vehicle; the driving environment information includes at least lane markings, wind speed, and gradient; and the decision domain includes, in descending order of priority, the dynamic stability domain, the collision safety domain, the regulatory domain, and the comfort domain.

[0037] The decision processing unit further includes:

[0038] The switching determination unit is used to determine that a decision domain switch is required when the priority of the target decision domain is higher than the priority of the current decision domain.

[0039] The decision processing unit further includes:

[0040] The target decision domain control quantity acquisition unit is used to calculate the control quantity u of the target decision domain according to the following formula. j :

[0041]

[0042] Where, x i =[v x v y ,ω] TThe state variables include longitudinal vehicle speed, lateral vehicle speed, and yaw rate; P is a pre-obtained diagonalized Lyapunov function, and n is the number of targets; u j The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle.

[0043] The switching processing unit further includes:

[0044] The buffer domain establishment unit is used to establish a buffer domain between the current decision domain and the target decision domain.

[0045] The buffer domain control quantity acquisition unit is used to obtain the current decision domain control quantity u. k The control quantity for the current buffer domain output is calculated periodically using the following formula:

[0046]

[0047] Among them, u * It is the control quantity output by the buffer domain, u j It is the control quantity of the decision domain, and α is the weighting coefficient, which is determined according to the switching completion degree of the decision domain;

[0048] The vehicle control unit is used to control the vehicle using the control quantity output from the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

[0049] This further includes:

[0050] An optimization controller generation unit is used to pre-obtain the optimization controller corresponding to each decision domain, and the optimization controller generation unit further includes:

[0051] The classification processing unit is used to classify driving needs into global needs, dynamic-related real-time needs, and trend indicator needs based on the time domain and state space covered by the driving needs.

[0052] The representation processing unit is used to obtain the representation indicators and constraints for each type of driving demand, and to establish different decision domains.

[0053] The mapping processing unit constructs constraint equations between variables for each decision domain based on driving requirements. Then, it maps different decision domains to a three-dimensional state space composed of yaw rate, longitudinal acceleration, and vehicle speed by associating vehicle speed and yaw rate with vehicle dynamics state equations and relative motion state equations between vehicle-road and vehicle-target vehicle.

[0054] The spatial model calculation unit is used to calculate the state variables and first-order information of the state variables corresponding to each decision domain by adopting an affine nonlinear state-space model that considers uncertainties. The uncertainties include modeling errors, nonlinear characteristics and external disturbance factors.

[0055] The inequality calculation unit is used to establish Hamilton-Jacobi inequalities based on the state variables and first-order information of the state variables corresponding to each decision domain to construct optimal control problems in different decision domains. The problem is solved through offline numerical calculation to obtain the corresponding diagonalized Lyapunov function.

[0056] The forming unit is used to obtain the calculation formula of the control quantity of each decision domain according to the diagonalized Lyapunov function, and form an optimized controller for each decision domain.

[0057] Implementing the embodiments of the present invention has the following beneficial effects:

[0058] This invention provides a multi-task, multi-objective optimization decision-making and planning method and system for autonomous driving. It simultaneously considers multiple objectives, including power, economy, safety, and comfort, for decision-making and planning. A multi-domain hierarchical approach is used to resolve conflicts and contradictions between different indicators in the multi-objective optimization solution. Furthermore, it considers the uncertainties caused by environmental disturbances such as obstacle dimensional changes, vehicle mass, and wind speed on the decision-making and planning problem, thereby enabling optimized decision-making and planning in complex conditions such as unprotected intersections and improving the success rate of decision-making.

[0059] By implementing the embodiments of the present invention, the applicable road conditions of intelligent vehicles can be effectively expanded, the level of automation can be improved, driving safety, stability and fuel economy can be comprehensively improved, and the bottleneck problems faced by high-level autonomous driving decision-making and planning can be overcome. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0061] Figure 1 This is a schematic diagram of the main flow of an embodiment of a multi-task, multi-objective optimization decision-making and planning processing method for autonomous driving provided by the present invention;

[0062] Figure 2 This is a more detailed flowchart illustrating the process of obtaining the optimization controller corresponding to each decision domain in advance, as described in this invention.

[0063] Figure 3 for Figure 2 A schematic diagram illustrating the representation and constraints of driving demands involved in the process;

[0064] Figure 4 for Figure 2 A schematic diagram illustrating the driving mode strategies and coordination switching principles involved.

[0065] Figure 5 for Figure 2 A schematic diagram illustrating the solution process of a robust optimization control problem that considers modeling errors, nonlinear characteristics, and external disturbances.

[0066] Figure 6 This is a schematic diagram of the structure of an embodiment of a multi-task, multi-objective optimization decision-making and planning processing system for autonomous driving provided by the present invention;

[0067] Figure 7 for Figure 6 A schematic diagram of the structure of the decision-making processing unit in the middle;

[0068] Figure 8 for Figure 6 A schematic diagram of the switching processing unit in the middle;

[0069] Figure 9 for Figure 6 A schematic diagram of the structure of the optimization controller generation unit;

[0070] Figure 10 This is a schematic diagram comparing the decision-making effects of the method of the present invention and existing methods in a typical unprotected intersection scenario;

[0071] Figure 11 This diagram illustrates a comparison of the decision planning success rates of the method of the present invention with those of existing methods, in a specific example. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0073] like Figure 1 As shown, this invention illustrates a multi-task, multi-objective optimization decision-making and planning processing method for autonomous driving. (In conjunction with...) Figures 2 to 5 As shown, in this embodiment, the method is applied to an autonomous driving system and includes at least the following steps:

[0074] Step S10: During the autonomous driving process, real-time vehicle driving information, target vehicle information, and driving environment information are acquired; the vehicle driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed; the target vehicle information includes at least the relative distance and relative speed between each target vehicle and the vehicle; the driving environment information includes at least lane lines, wind speed, and slope value.

[0075] Step S11: Determine the target decision domain of the vehicle within the expected time based on the acquired information; the decision domains are divided into dynamic stability domain, collision safety domain, regulatory domain and comfort domain according to priority from high to low.

[0076] Step S12: Determine whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain. If a decision domain switch is needed, calculate the control quantity corresponding to the target decision domain based on the preset optimization controller of the target decision domain.

[0077] Specifically, in step S12, in one example, when the priority of the target decision domain is higher than the priority of the current decision domain, it is determined that a decision domain switch is required. When the priority of the target decision domain is lower than the priority of the current decision domain, control can proceed according to the normal sequence control strategy.

[0078] The step of calculating the control quantity corresponding to the target decision domain based on the pre-set optimization controller of the target decision domain is as follows:

[0079] The control quantity uj in the target decision domain is calculated using the following formula:

[0080]

[0081] Where, x i =[v x v y ,ω] T The state variables include longitudinal vehicle speed, lateral vehicle speed, and yaw rate; P is a pre-obtained diagonalized Lyapunov function, and n is the number of targets; u j The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle.

[0082] Step S13: A soft-switching control strategy with a dynamic buffer domain is adopted. The vehicle is controlled by the control quantity corresponding to the target decision domain, and the vehicle is switched from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain.

[0083] The soft-switching control strategy employing a dynamic buffer domain, which uses the control quantity corresponding to the target decision domain to control the vehicle and switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain, specifically involves the following steps:

[0084] Establish a buffer zone between the current decision domain and the target decision domain;

[0085] Obtain the current decision domain control variable u k The control quantity for the current buffer domain output is calculated periodically using the following formula:

[0086]

[0087] Among them, u * It is the control quantity output by the buffer domain, u j It is the control quantity of the decision domain, and α is the weighting coefficient, which is determined according to the switching completion degree of the decision domain; Let b be the diagonal element corresponding to the transpose of P. T (x i ) is a predetermined value about x i The transpose of the function, b(x) i This will be introduced later;

[0088] The vehicle is controlled using the control quantity output from the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

[0089] In this invention, it is necessary to predetermine each decision domain and pre-obtain the corresponding optimization controller for each decision domain. For example... Figure 2 As shown, the process specifically includes the following steps:

[0090] Step S20: Based on the time domain covered by driving requirements and the state space involved, driving requirements are divided into global requirements, dynamic-related real-time requirements, and trend indicator requirements.

[0091] Specifically, in this embodiment of the invention, driving needs are categorized into global needs (economy and travel time), dynamic-related real-time needs (stability and comfort), and trend indicators (safety and traffic rules) based on the time domain (the entire journey, driving time, or a period of time) and the state space involved (vehicle longitudinal and lateral dynamics, relative motion relationships, etc.).

[0092] Step S21: Obtain the characterization index and constraints for each type of driving demand, and establish different decision domains;

[0093] like Figure 2As shown in this embodiment of the invention, for global requirements, since future traffic conditions are unpredictable and global optimization is not possible, it is proposed to indirectly improve global performance by optimizing the index q(t) (fuel consumption or travel time per unit distance) within the prediction time period T:

[0094]

[0095] Where J is the objective function.

[0096] Real-time requirements related to dynamics necessitate a focus on the constraints between different needs and longitudinal and lateral dynamics. From the perspective of tire grip characteristics and passenger perception of vehicle motion, a dynamic stability domain D is constructed in the vehicle's dynamic state space. s (x d (u) and driving comfort domain D c (x d ,u), where x d Let be the vehicle dynamics state variables (yaw rate, velocity, and acceleration / deceleration), and u be the control quantity consisting of acceleration / deceleration and steering angle. Applying vehicle dynamics principles, numerical simulation analysis is used to study the constraints, specifically including: the constraints between different dynamic state variables within the dynamic stability domain and driving comfort domain; the influence of the intelligent vehicle's control quantities on the boundaries of the stability and comfort domains; and the relative relationship between the dynamic stability domain and the driving comfort domain in the vehicle dynamics state space.

[0097] Furthermore, due to limitations imposed by vehicle motion trends, collisions and traffic violations are influenced by the spatiotemporal state of the system within a local timeframe preceding the event. Since future changes in road users and traffic conditions are unpredictable, accurately assessing the likelihood of future collisions and violations at the present moment is extremely difficult. This paper proposes a combined approach using static and dynamic indicators to characterize driving safety and traffic rule compliance requirements. The former reflects the intelligent vehicle's adaptability to environmental changes in its current state of motion, while the latter represents the reaction time allocated to the intelligent vehicle. Taking collision safety as an example, the corresponding static indicator J... ss and dynamic index J sd The calculation formula is:

[0098]

[0099] Where d is the distance vector between the main vehicle and the target workshop; s The desired safe following distance is a function of the vehicle's speed v and the included angle θ (the angle between v and d); v o It is the velocity vector of the target vehicle, θ r It is the angle between the relative velocity and distance vectors.

[0100] The following challenges exist in constructing multi-objective optimization decision-making and programming problems for multi-task applications in intelligent vehicles: the mathematical forms of the object equations differ, and their dimensions and orders are uncertain; the real-time performance and robustness of general numerical optimization methods are difficult to meet practical requirements; and the continuous approximation of discrete optimization indices can easily lead to singularities and deadlocks in the numerical solution process.

[0101] Therefore, in the method of this invention, a hierarchical multi-domain driving mode decision-making and scheduling strategy is constructed based on the priority and mathematical representation characteristics of driving needs. This hierarchical approach decomposes the autonomous driving process into an optimization control problem. Specifically, the system is divided into four domains according to priority: dynamic stability domain, collision safety domain, regulatory domain, and comfort domain. The optimization control objectives differ across these domains. If the vehicle cannot currently meet the requirements of the dynamic stability domain (which has the highest priority), the vehicle needs to meet these requirements as quickly as possible, ignoring the needs of other domains. Therefore, in actual decision-making, the system determines which decision domain the vehicle needs to switch to based on priority and actual road conditions, and then calls the corresponding optimization controller for optimization processing.

[0102] Step S22: For each decision domain, construct constraint equations between variables based on driving requirements, and then map different decision domains to a three-dimensional state space composed of yaw rate, longitudinal acceleration and vehicle speed by associating vehicle speed and yaw rate with vehicle dynamics state equations, relative motion state equations between vehicle and road and between vehicle and target vehicle.

[0103] Specifically, different driving needs involve different state spaces. To achieve the mapping of different needs to a unified decision space, firstly, constraint equations between variables are constructed based on driving needs. Then, by associating vehicle speed and yaw rate with vehicle dynamics state equations (longitudinal and lateral speeds, yaw rate) and relative motion state equations between vehicle-road and vehicle-object (relative speed and distance), different domains are mapped to a three-dimensional state space composed of yaw rate, longitudinal acceleration, and vehicle speed.

[0104] Taking the dynamic stability domain as an example, the longitudinal and lateral tire friction constraint relationship is first transformed into longitudinal and lateral acceleration constraints based on the tire-road adhesion characteristics:

[0105]

[0106] Among them, a x and a y These are the longitudinal and lateral accelerations of the vehicle, a, respectively. x.Max and a y.Max These represent the maximum allowable longitudinal and lateral accelerations / decelerations to ensure dynamic stability, respectively, where μ is the road adhesion coefficient. Lateral acceleration a y The relationship between the yaw rate and the vehicle steering dynamics equations is constrained.

[0107] When switching between different domains, the focus is on control smoothness, and a soft-switching control strategy based on a dynamic buffer domain is proposed. Assuming the vehicle is currently in domain A and the target domain is C, a buffer domain B is constructed on the domain boundary, and the control quantity u output during the switching process is...

[0108]

[0109] Among them, u A For the current domain control variable, u C α is the target domain control variable, and α is the weighting coefficient, which is determined based on the completion rate of driving mode switching.

[0110] Step S23: By establishing an affine nonlinear state-space model that considers uncertainty, calculate the state variables and first-order information of the state variables corresponding to each decision domain. The uncertainty includes modeling error, nonlinear characteristics and external disturbance factors.

[0111] In addition to nonlinear elements (such as tire slip, powertrain fuel consumption characteristics, suspension damping, and steering mechanism), real-world vehicles also exhibit parameter uncertainties (such as changes in occupant numbers), environmental disturbances (such as gradient and wind resistance), and high-order unmodeled dynamics (such as braking and steering actuators). Therefore, this invention establishes the following affine nonlinear perturbation model for each decision domain:

[0112]

[0113]

[0114] Wherein, state variable x i =[v x v y ,ω] T The parameters represent longitudinal vehicle speed, lateral vehicle speed, and yaw rate, respectively. The initial state variables are the inputs to the decision-making, planning, or control algorithm; while the control variable u = [F T δ f ] T The parameters in a(x) represent the longitudinal driving force and the front wheel steering angle, respectively, and are used as the output of decision-making, planning, or control algorithms. i ), b(x i ), g(x i ) for x i A nonlinear function, where d is the external disturbance, h(x) i z is the disturbance vector used to evaluate optimization performance and model uncertainty. i The output vector is used to evaluate optimization performance and model uncertainty.

[0115] Step S24: Based on the state variables and first-order information of the state variables corresponding to each decision domain, Hamilton-Jacobi inequalities are established to construct optimal control problems for different decision domains. The problems are solved through offline numerical methods to obtain the corresponding diagonalized Lyapunov functions.

[0116] In the method of this invention, predictive control can be employed, and based on dissipation theory, the following Hamilton-Jacobi inequality can be established to achieve suboptimal optimization within the prediction interval:

[0117]

[0118] Where P(x) is the Lyapunov function to be determined, g T (x), b T (x), P x T (x) represents g(x), b(x), and P. x The transpose of (x), where γ is a given supremum.

[0119] Understandably, due to the dynamic changes in the order of the decision object model for even a single driving task as the number of targets and traffic rules in the road scenario increase or decrease, the aforementioned inequality is essentially unsolvable. To address this problem, we propose utilizing the consistency of the analytical structure of the decision object equations. Specifically, taking the relative motion between vehicles as an example, the number of target vehicles affects the order of the equations, but the relative motion states between the main vehicle and each target vehicle are relative speed and distance, resulting in the same analytical form of the state equations. This allows us to construct a structurally universal diagonal Lyapunov function, achieving decomposition and dimensionality reduction of the Hamilton-Jacobi inequality, and numerically solving it to obtain a robust performance controller. We then calculate the specific numerical values ​​of the diagonalized Lyapunov function.

[0120] Step S25: Based on the diagonalized Lyapunov function obtained by solving, the calculation formula of the control quantity for each decision domain is obtained, and an optimized controller for each decision domain is formed.

[0121] The final control quantity u of the decision domain j and buffer domain control quantity u * The calculation formula is as follows:

[0122]

[0123] Among them, u * It is a control variable for the buffer domain, u j It is the control variable of the decision domain. Here, n is the diagonalized Lyapunov function, and n is the target number. Let be the diagonal element of the transpose of P.

[0124] The control quantity u of the decision domain j and buffer domain control quantity u * The formula can be used in autonomous driving of vehicles to switch target decision domains.

[0125] It is understood that, in the embodiments of the present invention, real-time decision planning in the autonomous driving process can be achieved by first evaluating, representing and constraining driving needs; then by coordinating driving mode scheduling strategies and switching, and by performing multi-objective robust real-time optimization of driving performance.

[0126] It is understood that the multi-task, multi-objective optimization decision-making and planning processing method for autonomous driving provided in this embodiment of the invention homogenizes different driving needs with heterogeneous manifestations, influence ranges, and modes of action, and determines the coupling, constraints, and inhibition of different needs; it adopts a hierarchical multi-domain decision-making and scheduling strategy that integrates multiple driving tasks and needs, as well as a smooth control method for the driving mode switching process; and it optimizes the robustness of the indeterminate-order system by considering modeling errors, nonlinear characteristics, and external disturbances. Therefore, it can adapt to more complex application scenarios, especially unprotected intersections, and can handle the complex situations arising from complex road topology, time-varying traffic dynamics, random and adversarial behavior trajectories of traffic participants in unprotected intersections, achieving a higher success rate in decision-making.

[0127] like Figure 6 The diagram shown illustrates a structural schematic of an embodiment of a multi-task, multi-objective optimization decision-making and planning processing system for autonomous driving provided by the present invention. (In conjunction with...) Figures 7-8 As shown, in this embodiment, system 1 is applied to an autonomous driving system, and it includes at least:

[0128] The information acquisition unit 10 is used to acquire, in real time, the vehicle's driving information, target vehicle information, and driving environment information during autonomous driving. The vehicle's driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed. The target vehicle information includes at least the relative distance and relative speed between each target vehicle and the vehicle. The driving environment information includes at least lane markings, wind speed, and slope.

[0129] The target decision domain determination unit 11 is used to determine the target decision domain corresponding to the vehicle within the expected time based on the acquired information; the decision domains are divided into dynamic stability domain, collision safety domain, regulatory domain and comfort domain according to priority from high to low.

[0130] The decision processing unit 12 is used to determine whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain. When a decision domain switch is needed, the control quantity corresponding to the target decision domain is calculated based on the optimization controller preset in the target decision domain.

[0131] The switching processing unit 13 is used to adopt a soft switching control strategy with a dynamic buffer domain, and use the control quantity corresponding to the target decision domain to control the vehicle, and switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain.

[0132] And an optimization controller generation unit 14, used to obtain the optimization controller corresponding to each decision domain in advance.

[0133] like Figure 7 As shown, the decision processing unit 12 further includes:

[0134] The switching determination unit 120 is used to determine that a decision domain switching is required when the priority of the target decision domain is higher than the priority of the current decision domain.

[0135] The target decision domain control quantity acquisition unit 121 is used to calculate the control quantity u of the target decision domain according to the following formula. j :

[0136]

[0137] Where, x i =[v x v y ,ω] T The state variables include longitudinal vehicle speed, lateral vehicle speed, and yaw rate; P is a pre-obtained diagonalized Lyapunov function, and n is the number of targets; u j The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle.

[0138] like Figure 8 As shown, in a specific example, the switching processing unit 13 further includes:

[0139] The buffer domain establishment unit 130 is used to establish a buffer domain between the current decision domain and the target decision domain;

[0140] Buffer domain control quantity acquisition unit 131 is used to obtain the current decision domain control quantity u. k The control quantity for the current buffer domain output is calculated periodically using the following formula:

[0141]

[0142] Among them, u* It is the control quantity output by the buffer domain, u j It is the control quantity of the decision domain, and α is the weighting coefficient, which is determined according to the switching completion degree of the decision domain;

[0143] The vehicle control unit 132 is used to control the vehicle using the control quantity output by the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

[0144] like Figure 9 As shown, in a specific example, the optimization controller generation unit 14 further includes:

[0145] The classification processing unit 140 is used to classify driving needs into global needs, dynamic-related real-time needs, and trend indicator needs based on the time domain covered by driving needs and the state space involved.

[0146] The representation processing unit 141 is used to obtain the representation indicators and constraints for each type of driving demand, and to establish different decision domains.

[0147] The mapping processing unit 142 constructs constraint equations between variables according to driving requirements for each decision domain, and then maps different decision domains to a three-dimensional state space composed of yaw rate, longitudinal acceleration and vehicle speed by associating vehicle speed and yaw rate with vehicle dynamics state equations, relative motion state equations between vehicle and road and between vehicle and target vehicle.

[0148] The spatial model calculation unit 143 is used to calculate the state variables and first-order information of the state variables corresponding to each decision domain by adopting an affine nonlinear state-space model that considers uncertainties. The uncertainties include modeling errors, nonlinear characteristics and external disturbance factors.

[0149] Inequality calculation unit 144 is used to establish Hamilton-Jacobi inequalities based on the state variables and first-order information of the state variables corresponding to each decision domain to construct optimal control problems in different decision domains. The corresponding diagonalized Lyapunov function is obtained by solving the problem through offline numerical calculation.

[0150] Forming unit 145 is used to obtain the calculation formula of the control quantity of each decision domain according to the diagonalized Lyapunov function, and form an optimized controller for each decision domain.

[0151] For more details, please refer to and combine with the above. Figures 1 to 5 The description of that will not be repeated here.

[0152] To verify the effectiveness of the algorithm proposed in this invention, in the following ways... Figure 10The verification was conducted in a scenario of an unprotected four-lane intersection. This scenario involved eight obstacle vehicles (OV1 to OV8) traveling in different directions under different conditions. Traffic rules such as dashed lane markings for lane changes and solid lines prohibiting lane changes were implemented. The dotted dashed lines represent the global navigation routes of the vehicles on a high-precision map. During the comparison, the vehicle's mass and the number of obstacles were treated as variable perturbations, and the results were compared with traditional algorithms that do not consider uncertainties. Figure 10 The diagram shows the vehicle's trajectory when there are 8 obstacles (target vehicles). Different shapes on each trajectory represent different time points. The trajectory shows that, in the trajectory generated by the traditional algorithm, the vehicle collides with obstacle OV8 at 15 seconds. However, the method used in this invention can successfully complete decision-making and planning in congested situations at intersections without traffic lights and safely navigate intersections with multiple dynamic obstacles.

[0153] At the same time, such as Figure 11 The diagram illustrates a comparison of the decision-planning success rates of the algorithm of this invention and a conventional algorithm in a typical unprotected intersection scenario. In one example, with a fixed vehicle mass and a random number of obstacles ranging from 1 to 10, 100 sets of data were compared. The decision-planning success rate of the conventional algorithm was 55%; while the success rate of decision-planning using the method provided by this invention was increased to 98%.

[0154] Implementing the embodiments of the present invention has the following beneficial effects:

[0155] This invention provides a multi-task, multi-objective optimization decision-making and planning method and system for autonomous driving. It simultaneously considers multiple objectives, including power, economy, safety, and comfort, for decision-making and planning. A multi-domain hierarchical approach is used to resolve conflicts and contradictions between different indicators in the multi-objective optimization solution. Furthermore, it considers the uncertainties caused by environmental disturbances such as obstacle dimensional changes, vehicle mass, and wind speed on the decision-making and planning problem, thereby enabling optimized decision-making and planning in complex conditions such as unprotected intersections and improving the success rate of decision-making.

[0156] By implementing the embodiments of the present invention, the applicable road conditions of intelligent vehicles can be effectively expanded, the level of automation can be improved, driving safety, stability and fuel economy can be comprehensively improved, and the bottleneck problems faced by high-level autonomous driving decision-making and planning can be overcome.

[0157] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0158] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0159] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A multi-task, multi-objective optimization decision-making and planning method for autonomous driving, applied to an autonomous driving system, characterized in that, It should include at least the following steps: During autonomous driving, real-time information is acquired about the vehicle's own driving conditions, the target vehicle's information, and the driving environment. Determine the target decision domain of the vehicle within the expected time frame based on the information obtained; Whether a decision domain switch is needed is determined based on the priority of the target decision domain and the current decision domain. When a decision domain switch is needed, the control quantity corresponding to the target decision domain is calculated based on the optimization controller preset in the target decision domain. A soft-switching control strategy with a dynamic buffer domain is adopted, which uses the control quantity corresponding to the target decision domain to control the vehicle and switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain. The step of determining whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain is as follows: When the priority of the target decision domain is higher than the priority of the current decision domain, it is determined that a decision domain switch is required; The step of calculating the control quantity corresponding to the target decision domain based on the pre-set optimization controller of the target decision domain is as follows: The control quantity of the target decision domain is calculated using the following formula. : in, It is the control quantity output by the buffer domain. These are the weighting coefficients. For the current decision domain control quantity, The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle; For state variables, These are longitudinal vehicle speed, lateral vehicle speed, and yaw rate information, respectively. For the pre-obtained diagonalized Lyapunov function, It refers to the number of targets; Let be the diagonal element corresponding to the transpose of P. For a pre-arranged agreement regarding The transpose of the function.

2. The method as described in claim 1, characterized in that, in: The vehicle's driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed; The target vehicle information includes at least the relative distance and relative speed between each target vehicle and the current vehicle; The driving environment information includes at least lane markings, wind speed, and gradient. The decision domains include: the dynamic stability domain, the collision safety domain, the regulatory domain, and the comfort domain, with priorities from high to low.

3. The method as described in claim 2, characterized in that, The soft-switching control strategy employing a dynamic buffer domain, which uses the control quantity corresponding to the target decision domain to control the vehicle, specifically involves the following steps to switch the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain: Establish a buffer zone between the current decision domain and the target decision domain; Obtain the current decision domain control quantity The control quantity for the current buffer domain output is calculated periodically using the following formula: in, It is the control quantity output by the buffer domain. It is the target decision domain control variable. The weighting coefficients are determined based on the completion rate of the decision domain switch. The vehicle is controlled using the control quantity output from the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

4. The method as described in claim 3, characterized in that, The process further includes the step of pre-obtaining the optimal controller corresponding to each decision domain, including: Based on the time domain and state space covered by driving needs, driving needs are divided into global needs, dynamic-related real-time needs, and trend indicator needs. Obtain the characterization indicators and constraints for each type of driving demand, and establish different decision domains; For each decision domain, constraint equations between variables are constructed based on driving requirements. Then, the vehicle dynamics state equations and the relative motion state equations between the vehicle and the road and between the vehicle and the target vehicle are associated with the vehicle speed and yaw rate. Different decision domains are mapped to a three-dimensional state space consisting of yaw rate, longitudinal acceleration and vehicle speed. By establishing an affine nonlinear state-space model that considers uncertainty, the state variables and first-order information of the state variables corresponding to each decision domain are calculated. The uncertainty includes modeling error, nonlinear characteristics and external disturbance factors. Hamilton-Jacobi inequalities are established based on the state variables and first-order information of the state variables corresponding to each decision domain to construct optimal control problems for different decision domains. The corresponding diagonalized Lyapunov functions are obtained by solving them through offline numerical methods. Based on the diagonalized Lyapunov function, the calculation formulas for the control quantities of each decision domain are obtained, forming an optimized controller for each decision domain.

5. A multi-task, multi-objective optimization decision-making and planning processing system for autonomous driving, applied in an autonomous driving system, characterized in that, At least including: The information acquisition unit is used to acquire real-time driving information of the vehicle itself, target vehicle information, and driving environment information during the autonomous driving process. The target decision domain determination unit is used to determine the target decision domain corresponding to the vehicle within the expected time based on the acquired information. The decision processing unit is used to determine whether a decision domain switch is needed based on the priority of the target decision domain and the current decision domain. When a decision domain switch is needed, the unit calculates the control quantity corresponding to the target decision domain based on the preset optimization controller of the target decision domain. The switching processing unit is used to employ a soft switching control strategy with a dynamic buffer domain, and to control the vehicle using the control quantity corresponding to the target decision domain, switching the vehicle from the driving mode corresponding to the current decision domain to the driving mode corresponding to the target decision domain. The decision processing unit further includes: A switching determination unit is used to determine that a decision domain switch is needed when the priority of the target decision domain is higher than the priority of the current decision domain. The target decision domain control quantity acquisition unit is used to calculate the control quantity of the target decision domain according to the following formula. : in, It is the control quantity output by the buffer domain. These are the weighting coefficients. For the current decision domain control quantity, The control quantity for the target decision domain includes information on longitudinal driving force and front wheel steering angle; For state variables, These are longitudinal vehicle speed, lateral vehicle speed, and yaw rate information, respectively. For the pre-obtained diagonalized Lyapunov function, It refers to the number of targets; Let be the diagonal element corresponding to the transpose of P. For a pre-arranged agreement regarding The transpose of the function.

6. The system as described in claim 5, characterized in that, in: The vehicle's driving information includes at least: yaw rate, longitudinal acceleration, and vehicle speed; The target vehicle information includes at least the relative distance and relative speed between each target vehicle and the current vehicle; The driving environment information includes at least lane markings, wind speed, and gradient. The decision domains include: the dynamic stability domain, the collision safety domain, the regulatory domain, and the comfort domain, with priorities from high to low.

7. The system as described in claim 6, characterized in that, The switching processing unit further includes: The buffer domain establishment unit is used to establish a buffer domain between the current decision domain and the target decision domain. The buffer domain control quantity acquisition unit is used to obtain the current decision domain control quantity. The control quantity for the current buffer domain output is calculated periodically using the following formula: in, It is the control quantity output by the buffer domain. It is the target decision domain control variable. The weighting coefficients are determined based on the completion rate of the decision domain switch. The vehicle control unit is used to control the vehicle using the control quantity output from the current buffer domain until it switches to the driving mode corresponding to the target decision domain.

8. The system as described in claim 7, characterized in that, Further includes: An optimization controller generation unit is used to pre-obtain the optimization controller corresponding to each decision domain, and the optimization controller generation unit further includes: The classification processing unit is used to classify driving needs into global needs, dynamic-related real-time needs, and trend indicator needs based on the time domain and state space covered by the driving needs. The representation processing unit is used to obtain the representation indicators and constraints for each type of driving demand, and to establish different decision domains. The mapping processing unit constructs constraint equations between variables for each decision domain based on driving requirements. Then, it maps different decision domains to a three-dimensional state space composed of yaw rate, longitudinal acceleration, and vehicle speed by associating vehicle speed and yaw rate with vehicle dynamics state equations and relative motion state equations between vehicle-road and vehicle-target vehicle. The spatial model calculation unit is used to calculate the state variables and first-order information of the state variables corresponding to each decision domain by adopting an affine nonlinear state-space model that considers uncertainties. The uncertainties include modeling errors, nonlinear characteristics and external disturbance factors. The inequality calculation unit is used to establish Hamilton-Jacobi inequalities based on the state variables and first-order information of the state variables corresponding to each decision domain to construct optimal control problems in different decision domains. The problem is solved through offline numerical calculation to obtain the corresponding diagonalized Lyapunov function. The forming unit is used to obtain the calculation formula of the control quantity of each decision domain according to the diagonalized Lyapunov function, and form an optimized controller for each decision domain.

Images