A method for realizing self-growth type cruise auxiliary driving

Abstract

The application discloses a method for realizing self-growth type cruise auxiliary driving, comprising the following steps: S1, establishing an intelligent vehicle cruise auxiliary driving system model and an adaptive cruise control model; S2, collecting human-machine interaction information representing human driving habits through the cruise auxiliary driving system model, combining real-time traffic environment information, and identifying the driving intention of a driver; S3, updating the cruise control model parameters according to the difference between the user personalized cruise trajectory estimated in step 2 and the actual performance of the cruise auxiliary driving system, updating the control law, so that the performance of the cruise auxiliary driving system in response to the external traffic environment gradually approaches the user personalized expectation, and realizing the self-growth of the capability. According to the application, the intelligent vehicle cruise performance gradually approaches and reaches the real expectation of the user driver, conforms to the personalized driving habits of the user driver, and provides technical support for expanding the intelligent auxiliary driving market and improving the trust level of the driver on the automatic driving vehicle.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent vehicle driving, and in particular to a method for realizing self-growing cruise-assisted driving. Background Technology

[0002] Adaptive cruise control and other driver assistance features have become significant innovations in the road traffic sector. However, these features currently face challenges such as low user adoption and acceptance. Most current driver assistance features are standardized and fail to match user preferences, leading to low user acceptance when using autonomous driving functions. Therefore, customization is a key issue for driver assistance technology.

[0003] With the rise of technologies such as machine learning and deep learning, learning-based driver assistance technologies are rapidly developing. By learning from expert demonstration behaviors (such as actual human driver trajectories), autonomous driving motion planning systems can generate driving trajectories that match the driver's individual driving habits, achieving customized autonomous driving and bringing tremendous opportunities for the practical promotion of autonomous driving functions. However, existing learning-based customized driver assistance system design methods still have limitations. First, current methods cannot adapt to users' real-time changing driving preferences. Existing learning-based methods collect users' personal driving data for offline training; however, users' driving preferences may change during driving, leading to poor performance and delayed response of offline training strategies in practical applications. Therefore, online self-learning capability is crucial for the customization of autonomous driving systems. Second, current methods lack the ability to respond to dynamic traffic environments. Existing learning-based methods essentially involve the vehicle motion planner imitating human driver behavior from the perspective of vehicle actions, but they do not understand the basic principles of human driving behavior, namely, response strategies to dynamic traffic environments. Therefore, how to learn and accumulate the experience of human drivers in responding to dynamic traffic environments and use it to update the vehicle motion planner to achieve self-learning of the autonomous driving system is also an urgent issue to be considered. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method for achieving self-evolving cruise assist, enabling intelligent vehicles to gradually approach and meet the actual expectations of users and drivers, conforming to the personalized driving habits of users, and providing technical support for expanding the intelligent assisted driving market and increasing drivers' trust in autonomous vehicles. To achieve the above-mentioned objectives and other advantages of the present invention, a method for achieving self-evolving cruise assist is provided, comprising the following steps:

[0005] S1. Establish a model for intelligent vehicle cruise assistance driving system and adaptive cruise control;

[0006] S2. Collect human-machine interaction information that represents human driving habits through the cruise assist driving system model, and combine it with real-time traffic environment information to identify the driver's driving intention;

[0007] S3. Based on the difference between the user's personalized cruise trajectory estimated in step 2 and the actual performance of the cruise assist driving system, update the cruise control model parameters and the control law, so that the performance of the cruise assist driving system in dealing with the external traffic environment gradually approaches the user's personalized expectations, and realizes self-growth capability.

[0008] Preferably, in step S1, an adaptive cruise control model is established based on the universal function objective of cruise driving to provide the cruise assistance system with the acceleration required when the vehicle is driving. This specifically includes the following steps:

[0009] S11. Based on empirical data, initialize the parameters of the driver's desired safe distance and desired acceleration / deceleration range. With the goal of maintaining the desired safe distance and consistent speed with the following vehicle, improving vehicle driving efficiency and fuel economy, and based on road alignment constraints, vehicle dynamics constraints, and vehicle performance constraints, establish an adaptive cruise control model using model predictive control methods.

[0010] S12. Using the current state of the intelligent vehicle as the initial state, solve the above adaptive cruise control model to obtain the optimal acceleration sequence of the vehicle in the future. The cruise assist driving system model uses the first control variable of the sequence as the vehicle's desired acceleration to execute the underlying control.

[0011] Preferably, the specific calculation process in step S2 includes the following steps:

[0012] S21. Using the relative position and relative speed of the main vehicle and surrounding vehicles as independent variables and the driver takeover probability as the dependent variable, establish a takeover probability distribution model.

[0013] S22. After the user activates the cruise assist driving system, collect the takeover data of the user driver during the actual driving process, including whether the accelerator or brake control is triggered, the status of the main vehicle and surrounding vehicles at the time of takeover, and fit the parameters of the probability distribution model by the information representing driving habits, thereby constructing a takeover probability distribution model for personalized users.

[0014] S23. Calculate the takeover probability of the main vehicle in each state using the takeover probability distribution model, extract the states with the smaller takeover probability as the expected driving state set of the main vehicle for cruise function, use this state set as a constraint, and optimize the driving trajectory of the main vehicle by using model predictive control and sequential dynamic programming methods to minimize the overall takeover probability, as the user's personalized cruise trajectory.

[0015] Preferably, in step S3, the user's personalized cruise trajectory is used as the expert trajectory, and the trajectory that the master vehicle has already traveled when it was last taken over is used as the actual trajectory. A series of trajectory feature evaluation functions are constructed to characterize the multi-dimensional features of the trajectory. Through the apprenticeship learning mechanism, the trajectory feature evaluation index is used as the optimization variable to maximize the difference between the expert trajectory and the actual trajectory while ensuring the optimality of the expert trajectory. This helps to identify the key features that led to the takeover of the actual trajectory and their weight distribution.

[0016] Preferably, the objective function of the adaptive cruise control model is updated based on the obtained feature and weight distribution results;

[0017] Based on the desired driving state set obtained in step 2.3, update the constraints of the adaptive cruise control model, solve the adaptive cruise control model, replan the expected acceleration sequence of the main vehicle, and use the first control variable of the expected acceleration sequence as the desired acceleration of the vehicle to execute the underlying control.

[0018] Compared with the prior art, the beneficial effects of this invention are: by solving and updating the desired driving state set, the vehicle driving state is constrained within the range of low takeover rate, and the vehicle action is optimized with the goal of reducing the takeover rate, thereby improving the driver's perceived safety level in different traffic environments and reducing the frequency of driver takeover.

[0019] The online self-evolving adjustable driver assistance system adapts to the unique driving styles of each driver, improving the driver's acceptance of the driver assistance system: by building a human-machine interaction system with driver takeover in the loop, it accumulates the driver's interaction response to background vehicles during autonomous vehicle cruise, and uses this experience to update the objective function, constraints and other parameters of the driver assistance system controller, so that the system performance gradually approaches the level expected by the driver.

[0020] Self-evolving driver assistance systems can adapt to various dynamic traffic environments and are universally applicable: through a driver-in-the-loop human-machine interface, the driver's responses to dynamic traffic environments (whether to take over, the state of takeover, etc.) are collected and used to train expert strategy models. As interactive data accumulates, the model is continuously improved, and the response strategies are gradually perfected. The system can cope with different dynamic traffic environments, even those it has never experienced before. This capability greatly facilitates the practical implementation of cruise assist systems. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] A method for implementing self-growing cruise-assisted driving includes the following steps:

[0023] S1. Establish a model for the intelligent vehicle cruise assist driving system and an adaptive cruise control model; In step S1, based on the universal functional objectives of cruise driving, an adaptive cruise control model is established to provide the cruise assist driving system with the acceleration required when the vehicle is driving. This specifically includes the following steps:

[0024] S11. Based on empirical data, initialize the parameters of the driver's desired safe distance and desired acceleration / deceleration range. With the goal of maintaining the desired safe distance and consistent speed with the following vehicle, improving vehicle driving efficiency and fuel economy, and based on road alignment constraints, vehicle dynamics constraints, and vehicle performance constraints, establish an adaptive cruise control model using model predictive control methods.

[0025] The model state variables of the adaptive cruise control model are Where K is the total number of time steps in the control model, and X i Let s be the system state variable at time step i. i v represents the distance traveled by the main vehicle along the centerline of the road at time step i. i Let l be the speed of the main vehicle at time step i. i Let be the distance the main vehicle deviates along the direction perpendicular to the road centerline at time step i. Let i be the heading angle of the main vehicle at time step i;

[0026] The model control variable of the adaptive cruise control model is U i =[a i δ i ] T , i∈[1,K], where, U i Let a be the system control variable at time step i. i Let δ be the acceleration of the main vehicle at time step i. i Let i be the deflection angle of the front wheels of the main vehicle at time step i;

[0027] The model system dynamics of the adaptive cruise control model are as follows:

[0028]

[0029]

[0030]

[0031]

[0032] Among them, A i Let B be the system matrix at time step i. iLet C be the control matrix at time step i. i This is the supplementary matrix for the i-th time step, where Δt is the control time step iteration size, l fr Where R is the front and rear wheelbase of the main vehicle, and J is the radius of curvature of the road. Q represents the desired system state at time step i, which is related to the state of the cruising vehicle. i R i Let be the weight matrix at time step i.

[0033] S12. Using the current state of the intelligent vehicle as the initial state, solve the above adaptive cruise control model to obtain the optimal acceleration sequence of the vehicle in the future. The cruise assist driving system model uses the first control variable of the sequence as the vehicle's desired acceleration to execute the underlying control.

[0034] S2. Collect human-machine interaction information representing human driving habits through the cruise assist driving system model, and combine it with real-time traffic environment information to identify the driver's driving intention; the specific calculation process in step S2 includes the following steps:

[0035] S21. Using the relative position and relative speed of the main vehicle and surrounding vehicles as independent variables, and the driver takeover probability as the dependent variable, establish a takeover probability distribution model. The specific formula is as follows:

[0036]

[0037] x=[Δs p Δl p dist p Δs a Δl a dist a ] T

[0038] Where x represents the vehicle and traffic state used as classification evaluation indicators, p(y|x) is the driver takeover probability under feature x, y is whether the vehicle has been taken over by the driver, y=1 represents no takeover, and y=0 represents takeover. class(x) represents the category to which state x belongs. Δs p The relative distance Δl between the main vehicle and the vehicle in front of it in the S-direction (along the centerline of the road). p The relative distance between the main vehicle and the vehicle in front in the l-direction (perpendicular to the road centerline), dist p The straight-line relative distance between the main vehicle and the vehicle ahead in the interaction. Δs a The relative distance Δl between the main vehicle and the vehicles interacting on the side in the S-direction (along the road centerline). a The relative distance between the main vehicle and vehicles interacting on the side in the l-direction (perpendicular to the road centerline), dist aThe straight-line relative distance between the main vehicle and the vehicles interacting on the side;

[0039] S22. After the user activates the cruise assist driving system, data on the user's actual driving process during takeover is collected, including whether accelerator or brake control is triggered, and the state of the main vehicle and surrounding vehicles at the time of takeover. By fitting the parameters of the probability distribution model with information representing driving habits, a takeover probability distribution model for personalized users is constructed. The specific model parameter identification process is as follows:

[0040]

[0041] p(y;θ)=θ y (1-θ) 1-y y∈{0,1}

[0042]

[0043]

[0044]

[0045]

[0046] Where μ is the mean vector, ∑ is the covariance matrix, and n is the feature dimension. Specifically, when y = 1, the mean vector is μ1 (i.e., the mean vector of the non-intervention probability is μ1), and when y = 0, the mean vector is μ0 (i.e., the mean vector of the intervention probability is μ0), θ is the probability of y = 1, N is the total amount of training data, N1 is the amount of non-intervention data in the training data, and N0 is the amount of intervention data in the training data. i y i () represents the i-th set of data;

[0047] S23. Calculate the takeover probability of the main vehicle in each state using the takeover probability distribution model. Extract the states with lower takeover probabilities as the expected driving state set of the main vehicle for cruise control, i.e., the set of x that satisfies the requirement class(x; μ, ∑, p) = 1. Using this state set as a constraint, and minimizing the overall takeover probability as the objective, optimize the main vehicle's driving trajectory using model predictive control and sequential dynamic programming methods, which serves as the user's personalized cruise trajectory ψ. E .

[0048] S3. Based on the difference between the user's personalized cruise trajectory estimated in step 2 and the actual performance of the cruise assist driving system, update the cruise control model parameters and the control law, so that the performance of the cruise assist driving system in dealing with the external traffic environment gradually approaches the user's personalized expectations, and realizes self-growth capability.

[0049] In step S3, the user's personalized cruise trajectory is used as the expert trajectory, and the trajectory already traveled when the master vehicle was last taken over is used as the actual trajectory. A series of trajectory feature evaluation functions are constructed to characterize the multi-dimensional features of the trajectory. Through an apprenticeship learning mechanism, using trajectory feature evaluation indicators as optimization variables, the difference between the expert trajectory and the actual trajectory is maximized while ensuring the optimality of the expert trajectory. This identifies the key features that led to the takeover of the actual trajectory and their weight distribution. The specific formula for apprenticeship learning is as follows:

[0050]

[0051] Where, w∈R M×K Let M be the objective function parameters of the system control model, M be the number of features corresponding to the vehicle state in a single control step, and K be the total number of control steps. ψ represents the vehicle trajectory, Φ(ψ)∈R. M×K Φ(ψ) represents the desired feature corresponding to the vehicle's trajectory ψ. E )∈R M×K Personalized cruise trajectory for users ψ E The corresponding expected features.

[0052] Furthermore, based on the obtained feature and weight distribution results, the objective function of the adaptive cruise control model is updated;

[0053] Based on the desired driving state set obtained in step 2.3, update the constraints of the adaptive cruise control model, solve the adaptive cruise control model, replan the expected acceleration sequence of the main vehicle, and use the first control variable of the expected acceleration sequence as the desired acceleration of the vehicle to execute the underlying control.

[0054] This type of update is "self-growing": the vehicle collects and processes information, uses its own computing power to optimize model parameters, iteratively updates the calculation results, and continues to apply them to the vehicle.

[0055] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention, and applications, modifications and variations thereof will be apparent to those skilled in the art.

[0056] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. Other modifications can be easily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and shown herein.

Claims

1. A method for realizing self-growing cruise-assisted driving, characterized in that, Includes the following steps: S1. Establish a model for the intelligent vehicle cruise assist driving system and an adaptive cruise control model; specifically, in step S1, based on the universal functional objectives of cruise driving, an adaptive cruise control model is established to provide the cruise assist driving system with the acceleration required when the vehicle is driving. This includes the following steps: S11. Based on empirical data, initialize the parameters of the driver's desired safe distance and desired acceleration / deceleration range. With the goal of maintaining the desired safe distance and consistent speed with the following vehicle, improving vehicle driving efficiency and fuel economy, and based on road alignment constraints, vehicle dynamics constraints, and vehicle performance constraints, establish an adaptive cruise control model using model predictive control methods. S12. Taking the current state of the intelligent vehicle as the initial state, solve the above adaptive cruise control model to obtain the optimal acceleration sequence of the vehicle in the future period of time, and the cruise assist driving system model uses the first control variable of the sequence as the vehicle's desired acceleration to execute the underlying control. S2. By collecting human-machine interaction information representing human driving habits through the cruise assist driving system model, and combining it with real-time traffic environment information, the driver's driving intention is identified; the specific calculation process includes the following steps: S21. Using the relative position and relative speed of the main vehicle and surrounding vehicles as independent variables and the driver takeover probability as the dependent variable, establish a takeover probability distribution model. S22. After the user activates the cruise assist driving system, collect the takeover data of the user driver during the actual driving process, including whether the accelerator or brake control is triggered, the status of the main vehicle and surrounding vehicles at the time of takeover, and fit the parameters of the probability distribution model by the information representing driving habits, thereby constructing a takeover probability distribution model for personalized users. S23. Calculate the takeover probability of the main vehicle in each state through the takeover probability distribution model, extract the state with the smaller takeover probability as the expected driving state set of the main vehicle for cruise function, use this state set as a constraint, and optimize the driving trajectory of the main vehicle by using model predictive control and sequential dynamic programming methods to minimize the overall takeover probability, as the user's personalized cruise trajectory. S3. Based on the difference between the user-personalized cruise trajectory estimated in step 2 and the actual performance of the cruise assist system, update the cruise control model parameters and the control law, so that the performance of the cruise assist system in dealing with the external traffic environment gradually approaches the user's personalized expectations, achieving self-growth capability. Specifically, the user-personalized cruise trajectory is used as the expert trajectory, and the trajectory that the main vehicle has traveled when it was last taken over is used as the actual trajectory. A series of trajectory feature evaluation functions are constructed to characterize the multi-dimensional features of the trajectory. Through the apprenticeship learning mechanism, the trajectory feature evaluation index is used as the optimization variable to maximize the difference between the expert trajectory and the actual trajectory while ensuring the optimality of the expert trajectory, thereby identifying the key features that lead to the takeover of the actual trajectory and their weight distribution.

2. The method for realizing self-growing cruise-assisted driving as described in claim 1, characterized in that, Based on the obtained feature and weight distribution results, update the objective function of the adaptive cruise control model; Based on the desired driving state set obtained in step 2.3, update the constraints of the adaptive cruise control model, solve the adaptive cruise control model, replan the expected acceleration sequence of the main vehicle, and use the first control variable of the expected acceleration sequence as the desired acceleration of the vehicle to execute the underlying control.

Images