A rapid MPC automatic driving control method based on road type and working condition recognition

Abstract

The application discloses a kind of fast MPC automatic driving control methods based on road type and working condition identification, method includes: according to the future road curvature of reference trajectory given in current time, identify road type;According to the speed and acceleration information given in current time reference trajectory, identify working condition type;According to the above identification result, adaptive selection actual prediction and control step;Through the lateral and longitudinal error dynamics model of vehicle, establish prediction and control model and build MPC solver framework;Under the MPC framework built, set different prediction and control step, obtain the benchmark of prediction and control step by simulation.Set according to road identification result, working condition identification result, benchmark prediction step and benchmark control step, generate actual application matrix;According to the input actual application matrix, update solver matrix calculation scale, and output control sequence.The application can quickly select the prediction step and control step suitable for the current driving state, significantly improve the calculation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and more specifically, relates to a fast MPC autonomous driving control method and system based on adaptive road type and driving condition recognition. Background Technology

[0002] Model Predictive Control (MPC) has gained widespread attention in autonomous driving technology due to its strong ability to solve multi-constraint optimization problems. The MPC framework includes a model prediction module, a rolling optimization module for solving the optimal control sequence, and a correction feedback module. Its control principle is based on a vehicle dynamics model, establishing a model prediction matrix. Under the constraints of system state variables and control variables, it continuously solves for the optimal control variable at each step of the prediction matrix to obtain the optimal control sequence. The value of the optimal control variable obtained in the first step is output to the actuator, while simultaneously updating future state variable constraints based on the vehicle's current state. The MPC algorithm has excellent capabilities in handling optimization problems in complex systems with multiple state variables, multiple control variables, and multiple constraints, exhibiting strong robustness. Although its control result is not the global optimal solution, it can correct errors caused by environmental disturbances in real time, achieving optimal control performance within the prediction domain.

[0003] The control performance of an MPC solver is directly proportional to the prediction step size and the control step size; longer prediction and control step sizes result in better control performance, but also significantly increase computation time. Achieving better control performance places higher demands on the actuator hardware. In traditional MPC applications for autonomous driving control, the prediction and control step sizes cannot be changed. Therefore, to meet various constraints and achieve good optimization results, the prediction and control step sizes are set relatively large. This results in a large number of matrix operations being required even when the vehicle is driving under simple road conditions, greatly increasing the computation time of the MPC and making its online application more difficult. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a fast MPC autonomous driving control method and system based on adaptive road type and working condition identification, which solves the problems of low computational efficiency and time consumption caused by the inability of traditional MPC solvers to dynamically change the prediction step size or dynamically control the step size.

[0005] This invention provides a fast MPC (Multi-Level Communication) automated driving control method based on adaptive road type and driving condition recognition, comprising the following steps:

[0006] (1) Identify the road type based on the future road curvature given by the reference trajectory at the current moment;

[0007] (2) Identify the future operating condition type based on the future velocity and acceleration information given by the reference trajectory at the current moment;

[0008] (3) Adaptively select the actual prediction step size and control step size based on the above identification results;

[0009] (4) By using the vehicle's lateral and longitudinal error dynamics model, establish prediction and control models and obtain the MPC solver framework;

[0010] (5) Using the built MPC framework, different prediction step size and control step size are set, and the prediction step size and control step size that meet the constraints are selected as the benchmark based on the simulation test results.

[0011] (6) Generate a practical application matrix based on the road identification results, working condition identification results, baseline prediction step size, and baseline control step size;

[0012] (7) Update the solver matrix calculation scale according to the actual prediction and control matrix scale input, and output the control sequence.

[0013] Furthermore, step (1) specifically involves:

[0014] Step 1.1: Obtain the coordinate sequence on the reference trajectory , The difference sequence of coordinates is obtained by numerical solution. , ;

[0015] Step 1.2: Calculate the heading sequence on the reference trajectory point based on the difference sequence of the horizontal and vertical coordinates. The heading sequence is then differentially analyzed to extract the heading increment sequence. and trajectory length increment sequence ;

[0016] Step 1.3: Based on the heading increment sequence and trajectory length increment sequence Calculate the road curvature sequence of the reference trajectory ;

[0017] Step 1.4: Calculate the road curvature sequence The average of absolute values, and based on the average To identify future reference road types.

[0018] Furthermore, step (2) specifically involves: extracting the velocity sequence from the reference trajectory. and acceleration sequence And calculate the difference between the velocity sequence and the current reference velocity value or determine the difference between the acceleration sequence and 0;

[0019] When the difference between the reference speed sequence and the current reference speed remains near 0, the future reference driving condition is constant speed; otherwise, it is variable speed.

[0020] When the value in the reference acceleration sequence remains near 0, the future reference driving condition is constant speed; otherwise, it is variable speed driving.

[0021] Furthermore, in step (3), the selection of the actual prediction step size and control step size is as follows:

[0022] Set evaluation criteria for selecting prediction step size and control step size;

[0023] A simulation test scenario is constructed; the simulation test scenario includes different road types and working conditions, and the simulation test scenario includes constant speed on straight roads, variable acceleration on straight roads, constant speed on large curvature curves, variable acceleration on large curvature curves, constant speed on small curvature curves, and variable acceleration on small curvature curves.

[0024] Based on the simulation test scenario, different prediction step sizes and control step sizes are set, and according to... Select the most suitable prediction step size and control step size as the baseline step size for MPC solution;

[0025] Where n represents the simulation step size, , , These represent the weighting coefficients for single-step time, single-step lateral error, and single-step longitudinal error, respectively.

[0026] Furthermore, step (4) specifically involves:

[0027] A dynamic model of lateral and longitudinal errors is established based on a two-degree-of-freedom vehicle model and considering tire side slip performance.

[0028] The lateral and longitudinal error dynamics model includes six state variables and two control variables. The state variables include lateral state variables and longitudinal state variables. The lateral state variables include lateral error. lateral speed Heading error and heading error angular velocity Longitudinal state variables include position error. and speed error The control variables include lateral control variables and longitudinal control variables. The lateral control variable is the front wheel steering angle. The longitudinal control variable is longitudinal error acceleration compensation. ,in

[0029] The state equation of error dynamics in the longitudinal and transverse error dynamics model is: ;in , , , , , , , , , , These represent the total vehicle mass, longitudinal velocity, front tire stiffness, rear tire stiffness, moment of inertia about the z-axis, and inherent road properties, respectively.

[0030] Discretize the state equation of the error dynamics to obtain ;

[0031] in , , , Indicates the single-step sampling time;

[0032] Discrete incremental equations are obtained by merging state variables and control variables. ;

[0033] The prediction step size is created using discrete incremental equations. And the control step size is The prediction equation: ;

[0034] Set the cost function of the control model and obtain the quadratic programming function. ;

[0035] The control quantity of the quadratic programming is constrained by the upper and lower limits of the actual control quantity increment, the upper and lower limits of the actual control quantity, and the upper and lower limits of the output state quantity. The control sequence is obtained by rolling the calculation based on the actual prediction step size and control step size.

[0036] Furthermore, step (5) specifically involves:

[0037] Step 5.1: Obtain the actual road conditions. When the vehicle is traveling on a straight road and the current heading error is within the allowable error range, vehicle path tracking does not involve lateral control; only longitudinal MPC needs to be considered. When traveling on a curve with small curvature, the smaller the curvature, the closer it is to a straight road, and the lateral control output changes relatively little, so a smaller prediction step size and control step size can be set. When traveling on a curve with large curvature, in order to keep up with the reference trajectory, the lateral control output changes relatively much, so a larger prediction step size and control step size need to be set.

[0038] Step 5.2: Obtain the reference driving condition, and maintain the longitudinal accelerator pedal and braking pressure under constant speed driving conditions. Only the lateral control amount needs to be controlled, and a small control and prediction step size is used. Under variable speed driving conditions, the appropriate prediction step size and control step size need to be set according to the actual road type.

[0039] Step 5.3: Based on the actual road conditions, reference conditions, and simulation results, select the actual prediction step size and control step size of MPC according to the information output by the reference trajectory at the current moment.

[0040] Furthermore, step (6) specifically involves:

[0041] Step 6.1: Based on the prediction step size and control step size Benchmark and currently selected actual prediction step size and control step size Set the control transformation matrix;

[0042] Step 6.2: Substitute the prediction transformation matrix and the control transformation matrix into the MPC solver framework to obtain the constraint expression form as follows: ;

[0043] in, To control the transformation matrix, To predict the transformation matrix.

[0044] More preferably, the currently selected actual prediction step size Less than or equal to prediction step size The currently selected actual control step size is less than or equal to the control step size. .

[0045] Specifically, step (7) involves solving the actual control sequence based on the introduced control transformation matrix and prediction transformation matrix, and outputting the first result of the control sequence to the actuator.

[0046] This invention also provides a fast MPC (Multi-Process Control) autonomous driving control system based on adaptive road type and working condition recognition, comprising: a road type recognition module, a working condition recognition module, a step size control module, a step size reference selection module, a transformation matrix generation module, an MPC module, and an MPC update module; the road type recognition module identifies the road type based on the future road curvature given by the reference trajectory at the current moment; the working condition recognition module identifies the future working condition type based on the future speed and acceleration information given by the reference trajectory at the current moment; the step size control module adaptively selects the actual prediction step size and control step size based on the above recognition results; the MPC module establishes prediction and control models and designs the MPC solver framework through the vehicle's lateral and longitudinal error dynamics model; the step size reference selection module uses the constructed MPC framework to set different prediction and control step sizes, and selects appropriate prediction and control step sizes that meet the constraints as the reference based on simulation test results; the transformation matrix generation module generates an actual application matrix based on the road recognition results, working condition recognition results, reference prediction step size, and reference control step size; the MPC update module updates the solver matrix calculation scale based on the input actual prediction and control matrix scale and outputs the control sequence.

[0047] Compared with the prior art, the above technical solutions conceived in this invention introduce a dynamic input interface for variable prediction step size and control step size within the framework of the traditional MPC solver. The algorithm proposed in this invention can quickly select the prediction step size and control step size suitable for the current driving state based on real-time reference trajectory information, solving the problem that the traditional MPC solver cannot dynamically change the prediction step size and control step size, significantly improving computational efficiency, and providing ideas for the online application of MPC algorithms. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the implementation module structure of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition provided in an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram of the framework of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition provided in an embodiment of the present invention.

[0050] Figure 3 This is a functional module diagram of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition provided in an embodiment of the present invention.

[0051] Figure 4(a) is a schematic diagram of the relationship between vehicle speed and time in the simulation results of the fast MPC autonomous driving control method based on adaptive road type and working condition identification provided in the embodiment of the present invention. Figure 4(b) is a schematic diagram of the relationship between the lateral distance error between the current position of the vehicle and the reference trajectory point and time. Figure 4(c) is a schematic diagram of the relationship between the accelerator pedal opening and time. Figure 4(d) is a schematic diagram of the relationship between the longitudinal distance error between the current position of the vehicle and the reference trajectory point and time.

[0052] Figure 5 The path tracking results are for the fast MPC autonomous driving control method based on adaptive road type and driving condition recognition provided in the embodiments of the present invention.

[0053] Figure 6(a) is a schematic diagram of the relationship between vehicle speed and time in the simulation results based on the traditional MPC autonomous driving control method. Figure 6(b) is a schematic diagram of the relationship between the lateral distance error between the current vehicle position and the reference trajectory point and time. Figure 6(c) is a schematic diagram of the relationship between the accelerator pedal opening and time. Figure 6(d) is a schematic diagram of the relationship between the longitudinal distance error between the current vehicle position and the reference trajectory point and time.

[0054] Figure 7 The results are path tracking results based on the traditional MPC autonomous driving control method.

[0055] Figure 8 This paper compares the single-step computation time of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition with that of the traditional MPC autonomous driving control method.

[0056] Figure 9 This improves the efficiency of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition compared to the traditional MPC autonomous driving control method, which has a longer single-step computation time. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0058] This invention overcomes the shortcomings of traditional MPC in autonomous driving control, such as high hardware requirements, long computation time, and inability to be applied online. It proposes a fast MPC autonomous driving control method based on adaptive road type and working condition recognition, which improves the computational efficiency of the MPC control algorithm while ensuring the constraints of lateral and longitudinal tracking errors.

[0059] Figure 1The implementation module structure of the fast MPC autonomous driving control method based on adaptive road type and working condition recognition provided in the embodiment of the present invention is shown; it includes 7 functional modules, namely: road type recognition module 1, working condition recognition module 2, step size control module 3, MPC module 6, step size reference selection module 4, transformation matrix generation module 5, and MPC update module 7;

[0060] The system comprises the following modules: Road Type Identification Module 1 identifies the road type based on the future road curvature given by the current reference trajectory; Working Condition Identification Module 2 identifies the future working condition type based on the future speed and acceleration information given by the current reference trajectory; Step Size Control Module 3 adaptively selects the actual prediction step size and control step size based on the identification results; MPC Module 6 establishes prediction and control models and designs the MPC solver framework through the vehicle's lateral and longitudinal error dynamics model; Step Size Reference Selection Module 4 uses the constructed MPC framework to select appropriate prediction and control step sizes that meet the constraints as references based on simulation test results by setting different prediction and control step sizes; Transformation Matrix Generation Module 5 generates the actual application matrix based on the road identification results, working condition identification results, reference prediction step size, and reference control step size; and MPC Update Module 7 updates the solver matrix calculation scale based on the input actual prediction and control matrix scale and outputs the control sequence.

[0061] Road type recognition module 1 mainly implements the road type recognition function according to the following steps:

[0062] Step 1.1: Obtain the coordinate sequence on the reference trajectory , The difference sequence of coordinates is obtained by numerical solution. , ; , ;in , represents the horizontal and vertical coordinate sequences, respectively, and n represents the length of the coordinate sequence given by the reference trajectory.

[0063] Step 1.2: Calculate the heading sequence on the reference trajectory points based on the calculated difference sequence of the x and y coordinates. Then, the heading sequence is differentially extracted to obtain the heading increment sequence. and trajectory length increment sequence ;in , .

[0064] Step 1.3: Based on the calculations in the previous step and Calculate the road curvature sequence of the reference trajectory ,in .

[0065] Step 1.4: Calculate the average of the absolute values ​​of the road curvature sequence to identify future reference road types; As mentioned above, road types are divided into straight roads, small-curvature curves, and large-curvature curves.

[0066] The operating condition identification module 2 mainly implements the operating condition identification function according to the following steps:

[0067] Step 2.1: Extract the velocity sequence from the reference trajectory and acceleration sequence It calculates the difference between the velocity sequence and the current reference velocity value, or determines the difference between the acceleration sequence and 0. If the difference between the reference velocity sequence and the current reference velocity remains near 0, the future reference driving condition is considered to be constant speed; otherwise, it is considered to be variable speed. If the value in the reference acceleration sequence remains near 0, the future reference driving condition is considered to be constant speed; otherwise, it is considered to be variable speed.

[0068] , ;in This represents the sum of the absolute values ​​of the differences between each point in the speed reference sequence at the current moment and the current vehicle speed. Represents each reference velocity in the velocity sequence. Indicates the current vehicle speed. This represents the sum of the absolute values ​​of the reference acceleration sequence. This represents each acceleration reference value in the reference acceleration sequence.

[0069] MPC Module 6 establishes prediction and control models based on the vehicle's lateral and longitudinal error dynamics models, and designs the MPC solver framework. The specific implementation includes the following steps:

[0070] Step 3.1: Based on the vehicle's two-degree-of-freedom model, establish a dynamic model for lateral and longitudinal errors, considering tire side slip performance. There are a total of 6 state variables for the integrated lateral and longitudinal control, of which 4 are lateral state variables: lateral error... lateral speed Heading error Angular velocity of heading error There are two longitudinal state variables: position error. Speed ​​error There are two control variables in total for the combined lateral and longitudinal control, with the lateral control variable being the front wheel steering angle. The longitudinal control variable is the longitudinal error acceleration compensation. ,in .

[0071] The error dynamics equation for continuous states is:

[0072]

[0073] in , , , , , These represent the vehicle mass, longitudinal velocity, front tire stiffness, rear tire stiffness, moment of inertia about the z-axis, and inherent road properties, respectively.

[0074] The state equation of error dynamics can be written in the following form: ;in , , , , .

[0075] Step 3.2: Discretize the continuous error dynamics model:

[0076] The continuity equations were discretized using the midpoint Euler method and the backward Euler method. The discretization results are as follows: ;in , , , This represents the single-step sampling time. The discrete equation is written in incremental form: .

[0077] Step 3.3: Merge state variables and control variables:

[0078] The equations in step 3.3 are written in matrix form for easy reference when solving the prediction model later:

[0079]

[0080] To facilitate observation, let , , , , The discrete incremental equation can be written in the following form: .

[0081] Step 3.4: Create a predictive model:

[0082] Based on the incremental equation in step 3.3, a prediction step size is created. Control step size is The prediction equation:

[0083] Output the prediction results for the state variables, by left-multiplying the prediction equation at each step by the block matrix. ;in , , These represent the number of state variables and control variables, respectively.

[0084] make , , , .

[0085] The model prediction equation can be further written in the following form: ;

[0086] in , , , ,

[0087] , .

[0088] Step 3.5: Set the cost function for the control model:

[0089] Using the state variable error and control variable increment as parameters of the cost function, the cost function expression is as follows: ;in , Let Y represent the coefficients of the state variable error and the control increment, respectively. Substitute the value of Y into the above equation, remove the terms unrelated to the control variable, and then introduce the relaxation factor. To speed up the cost function calculation, we obtain the updated cost function expression: Then, the above equation can be written in standard quadratic programming function form: .

[0090] Step 3.6: Solve the quadratic programming function. Constrain the control quantity of the quadratic programming function based on the upper and lower limits of the actual control quantity increment, the upper and lower limits of the actual control quantity, and the upper and lower limits of the output state quantity; The control sequence is solved by rolling based on the actual predicted step size and control step size, and the first item of the control sequence is output to the actuator.

[0091] The step size reference selection module 4 utilizes the built MPC framework to set different prediction step sizes and control step sizes, specifically including:

[0092] Step 4.1: Set the evaluation criteria for selecting the prediction step size and control step size. Under the same simulation conditions, considering the actual calculation time of the simulation model and driving safety, the cumulative simulation time, cumulative lateral error, and cumulative longitudinal error are used as the criteria for evaluating the selection of the prediction step size and control step size; among which... , where n represents the simulation step size. , , These represent the weighting coefficients for single-step time, single-step lateral error, and single-step longitudinal error, respectively.

[0093] Step 4.2: Design and build the simulation test scenario. The simulation scenario includes different road types and working conditions. The simulation test scenario includes straight sections with constant / variable acceleration, high-curvature curves with constant / variable acceleration, and low-curvature curves with constant / variable acceleration.

[0094] Step 4.3: Based on the simulation test scenario built in Step 4.2, design different prediction step size and control step size, and select the most suitable prediction step size and control step size as the baseline step size for MPC solution using the cost function given in Step 4.1.

[0095] The step size control module 3 adaptively selects the actual predicted step size and control step size based on the above identification results, specifically including:

[0096] Step 5.1: Analyze the actual driving road environment. When the vehicle is traveling on a straight road and the current heading error is within a very small tolerance range, vehicle path tracking does not involve lateral control; only longitudinal MPC is considered, and lateral MPC control is ignored. When traveling on curves with small curvature (meeting highway construction standards), the smaller the curvature, the closer it is to a straight road, and the smaller the change in lateral control output. Therefore, a smaller prediction step size and control step size can be set. However, when traveling on curves with large curvature (not meeting highway construction standards), in order to keep up with the reference trajectory, the change in lateral control output will be relatively large, so a larger prediction step size and control step size need to be set.

[0097] Step 5.2: Analyze the reference driving conditions. Under constant speed driving conditions, it is required to maintain longitudinal accelerator pedal and braking pressure. Essentially, only lateral control is needed, and the change in the overall control quantity will be within a relatively small range. Therefore, to reduce calculations, a smaller control and prediction step size is adopted. Under variable speed driving conditions, the change in the overall control quantity is larger, and it is necessary to set appropriate prediction and control step sizes based on the actual road type.

[0098] Step 5.3: Combining Step 5.1, Step 5.2, and the simulation results, select the actual prediction step size and control step size of MPC based on the information output by the reference trajectory at the current moment.

[0099] The transformation matrix generation module 5 generates a practical application matrix based on the road identification results, working condition identification results, baseline prediction step size, and baseline control step size; specifically, it includes:

[0100] Step 6.1: Based on the prediction step size and control step size Benchmark and currently selected actual prediction step size and control step size A control transformation matrix is ​​set. Because the MPC solver outputs the first term of the calculated optimal control sequence to the actuator, the control result calculated at a prediction time further away from the current time has a smaller impact on the vehicle. Therefore, in this invention, the transformation matrix follows a denser-to-sparser arrangement; where... , .

[0101] Step 6.2: Substitute the predicted transformation matrix and the control transformation matrix into the MPC solver framework, and denote the control transformation matrix as... The predicted transformation matrix is In the MPC solver, the control transformation matrix and the prediction transformation matrix are only related to the constraint influence of the state output on the control quantity. Therefore, substituting them into the constraint equations yields the following constraint expression:

[0102]

[0103] MPC update module 7 updates the solver matrix calculation scale based on the input actual prediction and control matrix scale and outputs the control sequence; specifically, it includes: solving the actual control sequence based on the introduced control transformation matrix and prediction transformation matrix, and outputting the first result of the control sequence to the actuator.

[0104] By introducing a dynamic input interface for varying prediction and control step sizes into the traditional MPC solver framework, the algorithm proposed in this invention can quickly select prediction and control step sizes suitable for the current driving state. This solves the problem that traditional MPC solvers cannot dynamically change prediction and control step sizes, significantly improving computational efficiency and providing ideas for online applications of MPC algorithms.

[0105] To further illustrate the autonomous driving control method and system provided in the embodiments of the present invention, a fast MPC autonomous driving control method based on adaptive road type and working condition recognition is used to simulate and test the autonomous driving capability of pure electric vehicles in a simulation scenario.

[0106] The test scenario was a two-way, slope-free road, with each lane 3.75m wide. Apart from the test vehicle, there were no other vehicles, pedestrians, or other obstacles on either side of the road or within the lane lines. Reference road types and operating conditions included straight-line constant speed, straight-line variable acceleration, constant speed on a high-curvature curve, variable acceleration on a high-curvature curve, constant speed on a low-curvature curve, and variable acceleration on a low-curvature curve. The test vehicle's initial position deviated from the reference trajectory, with a lateral deviation of 3.5m from the nearest point on the reference trajectory, a longitudinal deviation of 0.5m from the nearest point on the reference trajectory, and an initial heading deviation of 10 degrees from the reference heading.

[0107] Algorithm model initialization: Set the vehicle's initial position in the global coordinate system, including the x-coordinate ini_Host_x, y-coordinate ini_Host_y, and heading ini_Host_heading. Set the simulation sampling single-step sampling time to 0.001s. Set the MPC call cycle to 0.01s. Import the vehicle's global reference path coordinate sequence, the x-coordinate sequence Global_path_x_list, the y-coordinate sequence Global_path_y_list, and the path point reference velocity and acceleration sequences Global_path_Spd_list and Global_path_acc_list.

[0108] By combining global reference information and vehicle status information, find the matching point information of the closest point of the vehicle to the global path, and output the reference trajectory sequence [reference x-coordinate sequence ref_x_list, reference y-coordinate sequence ref_y_list, reference velocity sequence ref_Spd_list, reference acceleration sequence ref_acc_list] based on the matching point information;

[0109] Processing reference trajectory information to identify future road types: Extract the path sequence from the reference trajectory and use the difference to obtain the reference heading increment sequence. and path increment sequence The road curvature sequence of the future reference trajectory is obtained based on the calculated heading increment and path increment sequences. Based on the value of road curvature, the future reference road type is determined. In this invention, it is considered... At that time, the future reference road type was a straight road. It is a curve with a small curvature. It is a curve with a large curvature.

[0110] Processing reference trajectory information to identify future reference conditions: This step can use two methods to identify future reference conditions. In this invention, the method is to judge the sum of the absolute values ​​of the difference between the reference speed sequence and the current vehicle speed. When it is less than a preset threshold, the future reference condition is considered to be uniform speed driving; otherwise, it is variable speed driving.

[0111] The actual prediction step size and control step size in the current state are determined and the corresponding transformation matrix is ​​generated: Based on the results of steps 3 and 4, the corresponding prediction step size and control step size in the MPC solver are automatically matched, and the prediction transformation matrix and control transformation matrix are generated based on the prediction step size and control step size.

[0112] The MPC solver solves for the optimal control sequence in the prediction domain and outputs the first term of the control sequence to the vehicle actuator for execution.

[0113] The simulation results of the fast MPC autonomous driving control method based on adaptive road type and working condition identification proposed in this invention, compared with traditional MPC solution schemes, and using the reference prediction step size and control step size selected in this invention, are shown in Table 1. The simulation comparison results of the proposed method and traditional MPC in autonomous driving control are shown in Figures 4(a)~(d). Figure 9 As shown.

[0114] As can be seen from the results in Table 1, the fast MPC autonomous driving control method based on adaptive road type and working condition recognition proposed in this invention only took 158.27 seconds in actual operation when the simulation was set to a preset duration of 180 seconds. The computational efficiency is 46.56% higher than that of the traditional MPC solver. In addition, when the vehicle enters the globally planned lane, its maximum lateral deviation is 0.3056m (deviation from the lane centerline) and its maximum longitudinal deviation is 0.5248m. The deviations from the reference point in both the lateral and longitudinal directions meet the preset maximum values, indicating that this invention can effectively improve the MPC computational efficiency.

[0115] Table 1 Simulation Results

[0116]

[0117] Figure 4(a) shows the relationship curve of vehicle speed versus time in the vehicle simulation results of the fast MPC autonomous driving control method based on adaptive road type and working condition identification proposed in this invention. The vehicle speed is in m / s. Figure 4(b) shows the relationship curve of lateral distance error between the current position of the vehicle and the reference trajectory point versus time. The lateral distance error is in m. Figure 4(c) shows the relationship of accelerator pedal opening versus time. The accelerator pedal opening value ranges from [0, 1]. Figure 4(d) shows the relationship of longitudinal distance error between the current position of the vehicle and the reference trajectory point versus time. The longitudinal distance error is in m. Figure 5 The path tracking results of the optimization method proposed in this invention are shown in Figures 6(a) to 6(d). These figures represent the simulation results of speed, lateral and longitudinal errors, and accelerator pedal opening obtained by the traditional MPC control method. Figures 6(a), 6(b), 6(c), and 6(d) respectively show the relationship curves of vehicle speed over time, lateral distance error between the current vehicle position and the reference trajectory point over time, accelerator pedal opening over time, and longitudinal distance error between the current vehicle position and the reference trajectory point over time. Figure 7 The path tracking performance of the traditional MPC control method is shown. Combining the results in Figures 4(a)~(d) and 6(a)~(d), it can be seen that the simulation results of the proposed method are similar to those of the traditional MPC simulation results in terms of velocity curves, lateral and longitudinal errors, and the trend of accelerator pedal opening over time. This indicates that the solution accuracy of the proposed method is comparable to that of the traditional MPC. (See Figures 4(a)~(d) for further details.) Figure 5 As can be seen from 13s to 52s, as shown in Figure 4(c), the vehicle's accelerator pedal is fully open, and the vehicle accelerates at 5m / s to 35m / s on the high-curvature curve with maximum acceleration capability. The speed change over time is shown in Figure 4(a). During this stage of driving, the maximum distance the vehicle deviates from the road centerline is 0.3056m, as shown in Figure 4(b); the maximum longitudinal tracking error is less than 0.55m, as shown in Figure 4(d), and the vehicle always stays within its lane. From 55s to 80s, the vehicle decelerates on the high-curvature curve, and the maximum distance the vehicle deviates from the centerline is about 0.1m. Good lateral and longitudinal tracking performance is observed under other road types and conditions; Figures 4(a) to (d). Figure 5 The results show that the method proposed in this invention has stable path tracking capability even when driving in relatively complex road conditions.

[0118] Figure 8 To compare the single-step calculation time of the optimization method proposed in this invention with that of the traditional MPC solver under this comprehensive simulation test condition, Figure 9 The optimization method proposed in this invention improves efficiency compared to the single-step computation time of traditional MPC solvers. Figure 8 , Figure 9 It is evident that the method proposed in this invention consumes less time for single-step calculations than the traditional MPC solver in most cases.

[0119] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fast MPC autonomous driving control method based on adaptive road type and working condition recognition, characterized in that, Comprise the following steps: (1) According to the future road curvature given by the reference trajectory at the current time to identify the road type; (2) According to the future speed and acceleration information given by the reference trajectory at the current time to identify the future working condition type; (3) According to the above identification results, the actual prediction step and control step are adaptively selected; (4) Through the lateral and longitudinal error dynamics model of the vehicle, the prediction and control model is established and the MPC solver framework is obtained; (5) By setting different prediction steps and control steps, the prediction step and control step that meet the constraint conditions are selected as the benchmark according to the simulation test results; (6) According to the road identification results, the working condition identification results, the benchmark prediction step and the benchmark control step, the actual application matrix is generated; (7) According to the input actual prediction and control matrix scale, the solver matrix calculation scale is updated, and the control sequence is output; Among them, the selection of actual prediction step and control step in step (3) is: Set the evaluation criteria for selecting the prediction step and control step; Build a simulation test scene; The simulation test scene includes different road types and working condition types, and in the simulation test scene, straight road uniform speed, straight road variable acceleration, large curvature bend uniform speed, large curvature bend variable acceleration, small curvature bend uniform speed and small curvature bend variable acceleration are set; Based on the simulation test scene, different prediction steps and control steps are set, and the most suitable prediction step and control step are selected as the MPC solution reference step according to selecting the most suitable prediction step and control step as the MPC solution reference step; wherein n represents a simulation step length, , , respectively represent a single-step time consumption, a single-step horizontal error, and a single-step vertical error weight coefficient. Step (4) is specifically: Based on the two-degree-of-freedom model of the vehicle and considering the tire side slip performance, a lateral and longitudinal error dynamics model is established; The lateral and longitudinal error dynamics model includes 6 state variables and 2 control variables, the state variables include lateral state variables and longitudinal state variables, the lateral state variables include lateral error , lateral velocity , heading error and heading error angular velocity ; the longitudinal state variables include position error and velocity error ; the control variables include lateral control variables and longitudinal control variables, the lateral control variable is front wheel steering angle , the longitudinal control variable is longitudinal error acceleration compensation , wherein The state equation of error dynamics in the lateral and longitudinal error dynamics model is ; wherein , , , , , , , , , , respectively represent the whole vehicle mass, the vehicle longitudinal speed, the front wheel tire stiffness, the rear wheel tire stiffness, the moment of inertia around the z axis, the road inherent property; discretizing the state equation of the error dynamics to obtain ; wherein , , , denotes the single-step sampling time; Discrete delta equations are obtained by combining state and control quantities ; A prediction equation is created by discrete delta equations with a prediction step size of and a control step size of : ; Setting the control model cost function and obtaining a quadratic programming function ; The upper and lower limit constraints of the actual control quantity are obtained by limiting the upper and lower limits of the actual control quantity increment, the upper and lower limits of the actual control quantity, and the upper and lower limits of the output state quantity, and the control quantity of the quadratic programming is constrained; ; and the control sequence is solved by rolling according to the actual prediction step and the control step.

2. The fast MPC autonomous driving control method of claim 1, wherein, Step (1) is specifically: Step 1.1: Obtain the sequence of coordinates on the reference trajectory , , Numerically solve the difference sequence of coordinates , ; Step 1.2: Calculate the heading sequence on the reference trajectory point according to the difference sequence of the horizontal and vertical coordinates , and difference the heading sequence to obtain the heading increment sequence and the trajectory length increment sequence ; Step 1.3: Based on the heading increment sequence and trajectory length increment sequence Calculate the road curvature sequence of the reference trajectory ; Step 1.4: Calculate the sequence of road curvatures The average of the absolute values, and from the average to identify the future reference road type. 3.The fast MPC autonomous driving control method of claim 1, wherein, Step (2) is specifically: extracting the speed sequence in the reference trajectory and the acceleration sequence , and calculating the difference between the speed sequence and the current reference speed value or judging the difference between the acceleration sequence and 0; When the difference between the reference speed sequence and the current reference speed is always kept near 0, the future reference driving working condition is uniform speed, otherwise it is variable speed; When the value in the reference acceleration sequence is always kept near 0, the future reference driving working condition is uniform speed, otherwise it is variable speed driving. 4.The fast MPC autonomous driving control method of claim 1, wherein, Step (5) is specifically: Step 5.1: Obtain the actual driving road condition, and when the vehicle is driving on a straight road and the current heading and reference error are within the allowable error, the vehicle path tracking does not involve lateral control, only the longitudinal MPC needs to be considered; When driving on a small curvature bend, the smaller the curvature, the closer to the straight road, the relative change of the lateral control output is relatively small, and a smaller prediction step and control step can be set; When driving on a large curvature bend, in order to follow the reference trajectory, the change of the lateral control output will be relatively large, and a larger prediction step and control step needs to be set; Step 5.2: Obtain the reference working condition, and keep the longitudinal acceleration pedal and brake pressure in the uniform speed driving working condition, only the lateral control amount needs to be controlled, and a smaller control and prediction step is adopted; In the variable speed driving working condition, the appropriate prediction step and control step need to be set according to the actual road type; Step 5.3: According to the actual driving road condition, the reference working condition and the simulation experiment result, the MPC actual prediction step and control step of the information output by the reference trajectory at the current time are selected. 5.The fast MPC autonomous driving control method of claim 1, wherein, Step (6) is specifically: Step 6.1: Determine the predicted step size and control step size reference and current selected actual prediction step size and control step size , set the control transformation matrix; Step 6.2: The prediction and control transformation matrices are brought into the MPC solver framework to obtain the constraint expression in the form of: ; wherein, is a control transformation matrix, is a prediction transformation matrix.

6. The fast MPC autonomous driving control method of claim 5, wherein, current selected actual prediction step less than or equal to the prediction step the current selected actual control step is less than or equal to the control step .

7. The fast MPC autonomous driving control method of claim 5, wherein, Step (7) is specifically: According to the introduced control conversion matrix and prediction conversion matrix, the actual control sequence is solved, and the first item result of the control sequence is output to the actuator.

8. A fast MPC autonomous driving control system based on adaptive road type and working condition recognition, for implementing the fast MPC autonomous driving control method based on adaptive road type and working condition recognition according to any one of claims 1-7, characterized in that, Comprise: Road type identification module (1), working condition identification module (2), step length control module (3), step length reference selection module (4), conversion matrix generation module (5), MPC module (6) and MPC update module (7); The road type identification module (1) identifies the road type according to the future road curvature given by the reference trajectory at the current time; The working condition identification module (2) identifies the future working condition type according to the future speed and acceleration information given by the reference trajectory at the current time; The step length control module (3) adaptively selects the actual prediction step length and control step length according to the above identification results; The MPC module (6) establishes a prediction and control model through the lateral and longitudinal error dynamics model of the vehicle, and designs an MPC solver framework; The step length reference selection module (4) uses the MPC framework built by setting different prediction step lengths and control step lengths to select appropriate prediction step lengths and control step lengths that meet the constraint conditions as the reference according to the simulation test results; The conversion matrix generation module (5) generates the actual application matrix according to the road identification result, the working condition identification result, the reference prediction step length and the reference control step length; The MPC update module (7) updates the solver matrix calculation scale according to the input actual prediction and control matrix scale and outputs the control sequence.

Images