CN111897213A - A car adaptive cruise control method under model uncertainty - Google Patents

Abstract

The invention belongs to the technical field of vehicle control, and more particularly, relates to a method for vehicle adaptive cruise control under model uncertainty. Provides a method for adaptive cruise control that models vehicle dynamics through adaptive online Bayesian inference, based on a Lyapunov function to control the current vehicle for cruise control, and based on the The distance information and control barrier function are used to adjust the cruising speed of the vehicle in real time to ensure safety. The invention can adaptively adjust the cruising speed according to the current road conditions and the distance from the preceding vehicle, which is beneficial to ensure the safety of cruising.

Description

A car adaptive cruise control method under model uncertainty

technical field

The invention belongs to the technical field of vehicle control, and more particularly, relates to a method for vehicle adaptive cruise control under model uncertainty.

Background technique

With the development of the times, the automatic cruise control system of automobiles plays an important role in the field of driving. The cruise device can obtain the information of the preceding vehicle, such as distance, speed, etc., through the on-board sensors. Then the driver can set the corresponding desired cruise speed, and the automatic cruise system can replace the human to perform tasks efficiently for a long time, so it is widely used in the field of unmanned driving. However, in the real environment, there may be road gradient interference and changes in the friction coefficient of the road surface. It is difficult for the vehicle to obtain an accurate dynamic model. In addition, the vehicle ahead may have rapid acceleration and deceleration behaviors. Therefore, for vehicles with limited action space, in mission scenarios where safety is extremely critical, it is necessary to design an adaptive controller that can adaptively adjust the cruising speed of the vehicle through the workshop information to ensure safety.

In order to achieve the above goals, the method of autonomous cruise in the prior art requires an accurate vehicle dynamics model. For example, the patent 2017111478472 provides a cruise control method, device and system. During the modeling process, it is necessary to obtain accurate vehicle dynamics. However, when the road conditions change, such as when the road surface has undulating steep slopes and the road friction coefficient changes, the vehicle cannot stably reach the desired cruising speed, and at the same time, the safety of the vehicle may not be guaranteed, making it difficult to meet the driver's cruising needs.

SUMMARY OF THE INVENTION

In order to overcome the above-mentioned defects in the prior art, the present invention provides a vehicle adaptive cruise control method under model uncertainty, which realizes the stable operation and safe driving of the vehicle under the dynamic uncertainty environment.

In order to solve the above-mentioned technical problem, the technical solution adopted in the present invention is: a method for auto adaptive cruise control under model uncertainty, comprising the following steps:

S1. Build a vehicle nonlinear affine system model, and then model the uncertainty existing in the vehicle dynamics model based on a Gaussian process;

S2. Obtain the distance information between vehicles and the speed information of the vehicle based on the on-board sensors;

S3. Based on the current speed of the vehicle, construct a control Lyapunov function combined with the uncertainty of the vehicle model predicted by the Gaussian process, construct a quadratic programming problem with inequality constraints, and solve to obtain the optimal action u _clf to control the vehicle and let its speed Rapid and gradual convergence to the desired cruise speed, ensuring the stability of the cruise speed under high confidence;

S4. Based on the expected minimum safety distance between vehicles and the distance between vehicles obtained by on-board sensors, construct a control barrier function combined with the model uncertainty predicted by the Gaussian process to construct a safety constraint that constrains the robot exploration and action space at high confidence In the safe feasible domain; construct a quadratic programming problem with inequality constraints, correct the best action u _clf obtained in step 3 into the action set of the safe domain under the minimum correction, and obtain the best action u*;

S5. Apply the action u* obtained by the vehicle to the interaction of the current vehicle with the environment;

S6. Collect the state and action sequence of the interaction between the vehicle and the environment online, and update the Gaussian process;

S7. Repeat steps S1 to S6 until the interaction ends.

In the present invention, real-time modeling of vehicle dynamics is performed through adaptive online Bayesian inference, and a constrained safety optimization problem is constructed based on the control Lyapunov function and the control barrier function, so as to realize the stability of the vehicle under the dynamic uncertainty environment. Sex work and safe driving.

其中f(s)+g(s)a代表对系统动力学、运动学建模所得的先验模型，d(s)代表模型与真实环境的偏差。In one of the embodiments, in the step S1, the vehicle nonlinear affine system is modeled as:

where f(s)+g(s)a represents the prior model obtained by modeling system dynamics and kinematics, and d(s) represents the deviation of the model from the real environment.

In one embodiment, in the step S1, the specific form of the vehicle dynamics model is:

In the formula, p, v, a are the position, speed and acceleration state of the car, respectively; F _r is the aerodynamic resistance; F _f is the rolling friction coefficient; M is the mass of the car; g is the gravitational acceleration constant; Among them, F _r , F _f and Δθ are the uncertain parameters of vehicle dynamics that need to be modeled in real time.

其中s_t是t时刻车辆的状态，n为采集样本的数目；高斯过程回归通过贝叶斯推理得到状态s_*的下偏差d(s_*)的均值μ(s)和σ(s)；得到关于d(s)的高置信区间：D(s)＝{d|μ(s)-c_δσ(s)≤d≤μ(s)+c_δσ(s)}，其中μ(s)是偏差d(s)的均值，σ(s)是偏差d(s)的标准差，c_δ＞0是高斯分布中(1-δ)置信度对应的常数。In one of the embodiments, in the step S1, through the data set collected online

where s _t is the state of the vehicle at time t, and n is the number of samples collected; Gaussian process regression obtains the mean μ(s) and σ(s) of the lower deviation d(s _* ) of the state s _* through Bayesian inference; High confidence interval for d(s): D(s)={d|μ(s)-c _δ σ(s)≤d≤μ(s)+c _δ σ(s)}, where μ(s) is the mean of the deviation d(s), σ(s) is the standard deviation of the deviation d(s), and c _δ > 0 is the constant corresponding to the (1-δ) confidence in the Gaussian distribution.

In one embodiment, in the step S3, the control Lyapunov function is constructed and combined with the uncertainty of the vehicle model predicted by the Gaussian process, and the specific formula for constructing an inequality-constrained quadratic programming problem is:

stL _g V(s)u+L _f V(s)+L _μ V(s)+c _σ |L _σ V(s)|-κV(s)≤ε,

u _min ≤u≤u _max ,

In the formula, V(s) is the energy function, u is the vehicle action, u _max , u _min correspond to its upper and lower bounds, the diagonal matrix H(s) represents the weight of each dimension, ε represents the slack variable and K _ε is the corresponding coefficient, c _σ is the coefficient corresponding to the square term, the symbol L represents the Li's derivative, and κ is a k-type function.

In one embodiment, in the step S4, the specific formula of the constructed inequality-constrained quadratic programming problem is:

st-L _g h(s)uL _f h(s)-L _μ h(s)+c _σ |L _σ h(s)|-κh(s)≤η,

u _min ≤u≤u _max ,

In the formula, h(s) is the safety control barrier function, u is the vehicle action, u _max and u _min correspond to its upper and lower bounds, the diagonal matrix P(s) represents the weight of each dimension, and η represents the slack variable And K _η is the corresponding coefficient, c _σ is the coefficient corresponding to the square term, the symbol L stands for Li's derivative, and κ is a k-type function.

In one embodiment, the radial basis function kernel and the linear kernel are selected as Gaussian process kernel functions, and the corresponding Gaussian process model is initialized.

In one of the embodiments, there are four initialized Gaussian models for the vehicle, respectively corresponding to the position and speed of the vehicle and the position and speed of the preceding vehicle.

In one embodiment, the state dimension of the current vehicle and the preceding vehicle is selected as the model input, the actual interaction error between the vehicle and the environment is the output, and the Gaussian is updated by collecting the data set (s _t , at , s _{t +1} ) online. process model.

Compared with the prior art, the beneficial effects are as follows: the invention provides a vehicle adaptive cruise control method under model uncertainty, which solves the task of automatic cruise control at a limited state and action space of a nonlinear dynamic model vehicle. In the system, there are technical problems of unstable cruising speed and unsafe driving in the face of the uncertainty of its own system and the uncertainty of the external environment; the uncertainty of vehicle dynamics is modeled by Gaussian process, based on control Lyapunov and control The barrier function constructs a constrained quadratic programming problem and implements a method for adaptively adjusting the speed of dynamically imprecise vehicles cruising based on vehicle-to-vehicle (V2V) information. Based on the Gaussian process, the present invention realizes online self-adaptation to the environment through online collection of interactive data and Bayesian inference under high confidence, so as to model the uncertainty of the system dynamics model and the environment interaction, and to ensure the local Lipschitz continuity of the model. ; Based on the control Lyapunov function and Gaussian process, the stability of cruise control is guaranteed under high confidence; based on the control barrier function and Gaussian process, the vehicle speed is adjusted to constrain the vehicle action space in the safe and feasible region under high confidence, And the state in the unsafe domain can be gradually converged into the safe domain to ensure the safety of the vehicle during the cruise. .

Description of drawings

FIG. 1 is a schematic diagram of the overall flow of the method of the present invention.

FIG. 2 is a schematic flowchart of the algorithm of the present invention.

Detailed ways

The accompanying drawings are for illustrative purposes only, and should not be construed as limiting the present invention; in order to better illustrate the present embodiment, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art It is understandable to the artisan that certain well-known structures and descriptions thereof may be omitted from the drawings. The positional relationships described in the drawings are only for exemplary illustration, and should not be construed as limiting the present invention.

Example 1:

As shown in FIG. 1 , the present invention provides a method for vehicle adaptive cruise control under model uncertainty, including the following steps:

Step 1: This example builds a vehicle affine dynamics system model based on the prior knowledge of vehicle dynamics and actual task scenarios, and sets a desired cruise speed v. According to the task scenario, the expected cruise speed is selected as the balance point, and the corresponding control Lyapunov function (CLF) is designed: V(s _t ). Define the safe state space of the vehicle according to the requirements: C={x|h(x)≥0}, such as: keep a certain safety distance and tracking distance from the preceding vehicle, design the corresponding control barrier function (CBF): h(s _t ).

Step 2: In this example, the radial basis function (RBF) kernel and the linear kernel are selected as Gaussian process kernel functions, and the corresponding Gaussian process model is initialized. There are 4 Gaussian models for vehicle initialization, which correspond to the position and speed of the vehicle and the position and speed of the preceding vehicle. The present invention selects the state dimension of the current vehicle and the vehicle ahead as the model input, the actual interaction error between the vehicle and the environment as the output, and updates the Gaussian process model by collecting data sets (s _t , at , s _{t +1} ) online, as shown in the figure As shown in Fig. 2, the model of the vehicle dynamics is realized, and the Gaussian distribution of the interaction uncertainty is obtained.

Step 3: Obtain the current speed of the vehicle based on the on-board sensors, and construct a stable constraint for the vehicle through the defined CLF function, solve the above constraints by the Lagrange multiplier method, and obtain the control variable u _clf to make the vehicle's speed asymptotically converge to . Set desired cruise speed.

Step 4: Based on the expected minimum safety distance between vehicles and the distance between vehicles obtained by on-board sensors, construct safety constraints for the vehicle through the defined CBF function to correct the control variable u _clf in real time, and correct u _clf to the safety domain action under the minimum correction criterion In the set, the state space of the vehicle is maintained in the safe domain, and the optimal action u* is obtained.

Step 5: Apply the action u* obtained by the vehicle to the interaction of the current vehicle with the environment;

Step 6: Repeat steps 2 to 6 until the desired goal or maximum experimental set-up time is reached.

In this embodiment, in the automatic cruise system of the car, the controlled vehicle can adjust the expected cruise speed through the algorithm proposed by the present invention under the uncertainty caused by the road environment and the acceleration and deceleration of the preceding vehicle to ensure safety, Keep the controlled vehicle within a safe distance from the vehicle in front.

Obviously, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those of ordinary skill in the art, changes or modifications in other different forms can also be made on the basis of the above description. There is no need and cannot be exhaustive of all implementations here. Any modifications, equivalent replacements and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (10)

1. An automobile self-adaptive cruise control method under model uncertainty is characterized by comprising the following steps:

s1, constructing a vehicle nonlinear affine system model, and then modeling uncertainty existing in a vehicle dynamics model based on a Gaussian process;

s2, acquiring distance information between vehicles and speed information of the vehicle based on a vehicle-mounted sensor;

s3, constructing a quadratic programming problem of inequality constraint by constructing a control Lyapunov function and combining with the uncertainty of an automobile model predicted by a Gaussian process based on the current speed of the automobile, and solving to obtain the optimal action u_clfThe vehicle is controlled to quickly and gradually converge to the expected cruising speed, and the stability of the cruising speed is ensured under high confidence;

s4, constructing a control barrier function based on the expected minimum workshop safety distance and the vehicle distance obtained by the vehicle-mounted sensor, combining model uncertainty of Gaussian process prediction to construct safety constraint, and constraining the robot exploration and the action space in a safe feasible region under high confidence; constructing an inequality constrained quadratic programming problem, and correcting the optimal action u obtained in the step three under the minimum correction_clfModifying the operation into a security domain action set, and obtaining an optimal action u;

s5, applying the action u obtained by the vehicle to the interaction of the current vehicle and the environment;

s6, collecting the interaction state and action sequence of the vehicle and the environment on line, and updating a Gaussian process;

s7, repeating the steps S1 to S6 until the interaction is finished.

2. The model-uncertain vehicle adaptive cruise control method according to claim 1, wherein in step S1, the vehicle nonlinear affine system is modeled as:

wherein f(s) + g(s) a represents the movement of the systemAnd (d) representing the deviation of the model from the real environment.

3. The method for controlling adaptive cruise for an automobile under model uncertainty according to claim 2, wherein in step S1, the vehicle dynamics model is embodied in the form of:

wherein p, v, a are the position, velocity and acceleration states of the vehicle, respectively; f_rIs aerodynamic drag; f_fIs the rolling friction coefficient; m is the mass of the automobile; g is a gravitational acceleration constant; delta theta is the disturbance of unmodeled environmental road gradient and the like; wherein, F_r、F_fAnd Δ θ are the vehicle dynamics uncertainty parameters that need to be modeled in real time.

4. The model-uncertain vehicle adaptive cruise control method according to claim 3, wherein in step S1, the data set is collected on-line

Wherein s is_tIs the state of the vehicle at time t, and n is the number of collected samples; gaussian process regression through Bayesian inference to obtain state s_*Lower deviation d(s) of_*) The mean values μ(s) and σ(s); a high confidence interval for d(s) is obtained.

5. The model-uncertain vehicle adaptive cruise control method according to claim 4, wherein the high confidence interval for d(s) is:

D(s)＝{d|μ(s)-cσ(s)≤d≤μ(s)+cσ(s) }; where μ(s) is the mean of the deviations d(s), σ(s) is the standard deviation of the deviations d(s), c> 0 is a constant corresponding to the (1-) confidence in the Gaussian distribution.

6. The model-uncertain automobile adaptive cruise control method according to claim 4, wherein in the step S3, a quadratic programming problem is constructed by combining the uncertainty of the automobile model predicted by the gaussian process and controlling the lyapunov function, and the specific formula of the inequality constraint quadratic programming problem is as follows:

s.t.L_gV(s)u+L_fV(s)+L_μV(s)+c_σ|L_σV(s)|-κV(s)≤,

u_min≤u≤u_max,

where V(s) is a function of energy, u is vehicle motion, u is_max,u_minThe diagonal matrix H(s) represents the weight of each dimension action, representing the slack variable and K corresponding to its upper and lower boundsIs a corresponding coefficient, c_σThe notation L represents the Li derivative for the coefficient corresponding to the square term, and k is a function of class k.

7. The method for controlling adaptive cruise control for an automobile under model uncertainty according to claim 6, wherein in step S4, the specific formula of the quadratic programming problem with inequality constraints is:

s.t.-L_gh(s)u-L_fh(s)-L_μh(s)+c_σ|L_σh(s)|-κh(s)≤η,

u_min≤u≤u_max,

where h(s) is a safety control barrier function, u is vehicle motion, u is_maxAnd u_minCorresponding to its upper and lower bounds, the diagonal matrix P(s) represents the weight of each dimension of motion, η represents the relaxation variable and K_ηIs a corresponding coefficient, c_σThe symbol L represents lie for the coefficient corresponding to the square termThe derivative of the equation, κ, is a function of the k class.

8. The model-uncertain automobile adaptive cruise control method according to claim 7, characterized in that a radial basis function kernel and a linear kernel are selected as gaussian process kernel functions, and a corresponding gaussian process model is initialized.

9. The model-uncertain vehicle adaptive cruise control method according to claim 8, wherein there are 4 vehicle-initialized gaussian models corresponding to the position and speed of the vehicle and the position and speed of the preceding vehicle, respectively.

10. Method for model-uncertain adaptive cruise control of a vehicle according to claim 9, characterized in that the state dimensions of the current vehicle and the vehicle in front are chosen as model inputs and the actual interaction error of the vehicle and the environment as output, by collecting the data set(s) on-line(s)_t,a_t,s_t+1) And updating the Gaussian process model.

Images