Autonomous parking control method for autonomous vehicle in irregular scene

Abstract

The invention discloses an autonomous parking control method for an autonomous vehicle in an irregular scene. The method comprises the following steps: 1, establishing an autonomous parking kinematic model of the autonomous vehicle; 2, designing a control input constraint of an autonomous parking task; 3, approximately expressing the autonomous vehicle and the obstacle as a convex polygon design obstacle avoidance constraint; 4, designing a terminal-free model predictive control algorithm based on manual reference; based on the kinematics model of the automatic driving vehicle under the low-speed condition, manual reference and obstacle avoidance constraint are introduced on the premise of ensuring the feasibility of the optimization problem by designing a model prediction control algorithm without terminal cost and terminal constraint, the calculation demand is effectively reduced, the solvability of practical application is improved, and the method is suitable for large-scale popularization and application. The method is very suitable for motion control of autonomous parking of the autonomous driving vehicle in an irregular scene, and finally safe and accurate control is achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of automatic driving vehicle control, and particularly relates to an autonomous parking control method for an automatic driving vehicle in an irregular scene. BACKGROUND

[0002] The autonomous parking task needs to accurately control the throttle, brake and steering to accurately drive the automatic driving vehicle from the initial position to the designated target parking space in the parking lot. During the entire parking process, the automatic driving vehicle must ensure a safe distance from the surrounding obstacles. Due to the generally limited parking space and the complex vehicle dynamics, it is still a core problem for the autonomous parking technology of the automatic driving vehicle to achieve precise and safe motion control in a small space. The nonlinear controller with prediction function is particularly suitable for dealing with such complex motion control requirements.

[0003] The model predictive control algorithm is an optimal control algorithm that can directly handle constraints, and has been widely researched and developed in the past few decades, and is particularly suitable for complex control systems subject to multi-variable and input, state and output constraints. The core idea of the model predictive control algorithm is to predict the future behavior of the system using a model, and to optimize the control input based on these predictions, while considering the constraint conditions. Therefore, model predictive control exhibits significant advantages in solving the complex dynamic problems of automatic driving vehicles. However, the closed-loop stability of a system using the model predictive control algorithm usually needs to be achieved by introducing terminal constraint conditions and terminal cost terms. In practical applications of complex scenarios, such constraints often need complex offline design and increase the real-time computing burden. In order to improve the control performance and robustness, the terminal-free model predictive control strategy has become a promising research direction in the field of autonomous parking control. The stability of a system using this strategy can be guaranteed by setting a sufficiently large prediction horizon and cost controllability conditions. On this basis, the terminal-free model predictive control strategy based on artificial reference can effectively reduce the dependence on long prediction horizons, thereby reducing the overall computing burden. In the irregular autonomous parking scene with obstacles, the reference trajectory generated by the trajectory planning module is usually presented in the form of a sequence of limited discrete points. However, due to safety constraints and dynamic limitations, the automatic driving vehicle may not be able to accurately follow the reference trajectory, i.e., the reference trajectory is not reachable. The model predictive control algorithm based on artificial reference can integrate offline path planning into the online control phase, solving the problem of reference trajectory unreachability. Therefore, designing a terminal-free model predictive control algorithm based on artificial reference has important significance for solving the autonomous parking control of automatic driving vehicles in irregular scenes. SUMMARY

[0004] The purpose of the present application is to solve the problem of obstacle avoidance constraint and reference trajectory unreachability of existing autonomous parking control of autonomous vehicle in irregular scene, and propose an artificial reference based terminal-free model predictive control algorithm, so as to realize the integration of local planning into online control process and reduce the calculation burden.

[0005] The technical solution of the present application is: An autonomous parking control method of autonomous vehicle in irregular scene, comprising the following steps: Step 1: establishing an autonomous parking kinematic model of autonomous vehicle; Step 2: designing control input constraint of autonomous parking task; Step 3: designing obstacle avoidance constraint by approximating autonomous vehicle and obstacle as convex polygon; Step 4: designing artificial reference based terminal-free model predictive control algorithm.

[0006] Further, in step 1, the establishment process of the autonomous parking kinematic model of the autonomous vehicle is as follows: When the vehicle speed is low, the autonomous vehicle is assumed to be a rigid body, and the front and rear wheels of the vehicle only move forward along the wheel rolling direction, and the kinematic equation satisfied by the center point of the rear axle of the vehicle is:

[0007] wherein, and are the axis component and axis component of the rear axle center of the vehicle in the geodetic coordinate system, is the yaw angle of the vehicle, is the yaw rate of the vehicle, is the front wheel steering angle of the vehicle, is the speed of the vehicle, is the wheelbase of the vehicle; the control input of the autonomous vehicle kinematic control system is set as and which remain unchanged within each sampling period , and the kinematic equation of the rear axle center of the vehicle is discretized as:

[0008] wherein, represents the current discrete time, , the vehicle state and the control input are represented as and .

[0009] Further, the design process of the control input constraint of the autonomous parking task is as follows: The design input constraint is that the absolute value of the maximum speed of the vehicle in the autonomous parking process is set as Since the front wheel angle of the vehicle is constrained by the mechanical structure of the vehicle, the absolute values of the maximum left front wheel deflection angle and the maximum right front wheel deflection angle of the vehicle are set as Therefore, the control input constraint of the autonomous parking task of the autonomous vehicle is represented as: .

[0010] Further, the obstacle avoidance constraint of approximating the autonomous vehicle and the obstacles as convex polygons is as follows: At the initial time of the autonomous parking process, the planar space occupied by the autonomous vehicle is represented as the following rectangular region:

[0011] wherein and are constants related to the length and width of the autonomous vehicle, respectively; at time, the rectangular region is changed by rotation and translation to:

[0012] wherein is a rotation transformation matrix, is a translation transformation vector; the planar space occupied by the autonomous vehicle is described as a convex set in the autonomous parking process, and each obstacle is modeled as a polygon of the following form:

[0013] wherein represents the number of obstacles, and are related to the size of the obstacle, respectively, represents the number of polygon edges; is a compact convex set with a non-empty relative interior; the obstacle avoidance constraint of approximating the autonomous vehicle and the obstacles as convex polygons is represented as:

[0014] wherein represents the minimum safety distance required to complete the autonomous parking, and for any , there exist and such that the obstacle avoidance constraint:

[0015] is established.

[0016] Further, the artificial reference based terminal-free model predictive control algorithm design process is as follows: The cost function of the artificial reference based terminal-free model predictive control is defined as:

[0017] Wherein, represents the total step of the prediction horizon starting at the moment, represents the step in the prediction horizon, and are positive definite symmetric weight matrices, and respectively represent the artificial reference state and the artificial reference input, and respectively represent the parking trajectory reference state and the parking trajectory reference input output by the trajectory planning module; at the moment, the optimization problem of the artificial reference based terminal-free model predictive control algorithm is designed as:

[0018] Wherein represents the step in the prediction horizon, represents the obstacle.

[0019] Compared with the prior art, the present application has the following advantages: 1. The artificial reference variable is introduced into the model predictive control framework as an additional decision variable, realizing the integration of the local planning of the autonomous parking process of the autonomous vehicle into the online control process, and avoiding the loss of feasibility of the optimization problem caused by the infeasible reference trajectory.

[0020] 2. The present application accurately converts the non-differentiable obstacle avoidance constraint into a smooth differentiable expression form by using convex optimization, improving the calculation efficiency of the optimization problem solver while ensuring the safe movement of the autonomous vehicle in the irregular scene.

[0021] 3. The present application designs a model predictive control algorithm without terminal cost and terminal constraint, introduces the artificial reference and obstacle avoidance constraint under the premise of ensuring the feasibility of the optimization problem, effectively reduces the calculation demand and improves the solvability of the practical application, and is very suitable for the motion control of the autonomous parking of the autonomous vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a schematic diagram of the autonomous vehicle model of the present application; Figure 2 is a schematic diagram of the autonomous parking in the irregular scene of the present application;​ Figure 3 Structure block diagram of autonomous parking control algorithm of automatic driving vehicle in irregular scene of the application; Figure 4 Schematic diagram of reference trajectory of autonomous parking experiment of automatic driving vehicle of the application; Figure 5 Schematic diagram of motion process of autonomous parking experiment of automatic driving vehicle of the application; Figure 6 State error diagram of autonomous parking experiment of automatic driving vehicle of the application. DETAILED DESCRIPTION

[0023] The application will be described in detail below in combination with the drawings and specific embodiments.

[0024] Step 1, establish the kinematic model of autonomous parking of automatic driving vehicle: Under the condition of low-speed motion, the vehicle is assumed to be a rigid body, and the kinematic model is considered as a kinematic model without load transmission as shown in Figure 1 The kinematic equation satisfied by the center point of the rear axle of the vehicle is:

[0025] wherein, and are the axis component and axis component of the center of the rear axle of the vehicle in the geodetic coordinate system, is the yaw angle of the vehicle, is the yaw angular velocity of the vehicle, is the front wheel steering angle of the vehicle, is the speed of the vehicle, is the wheelbase of the vehicle; the control inputs of the kinematic control system of the automatic driving vehicle are set as and which remain unchanged within each sampling period The kinematic equation of the center of the rear axle of the vehicle is discretized as:

[0026] wherein, represents the current discrete time, the vehicle state and the control input are represented as and respectively.

[0027] Step 2, design the control input constraint of the autonomous parking task: The absolute value of the maximum speed of the vehicle in the autonomous parking process is set as Since the front wheel angle of the vehicle is constrained by the mechanical structure of the vehicle, the absolute values of the maximum left front wheel deflection and the maximum right front wheel deflection of the vehicle are set as Therefore, the control input constraint of the autonomous parking task of the autonomous vehicle is represented as: .

[0028] Step 3: Approximate the autonomous vehicle and obstacles as convex polygons to design obstacle avoidance constraints: At the initial moment of the autonomous parking process, the planar space occupied by the autonomous vehicle is represented as the following rectangular region:

[0029] wherein and are constants related to the length and width of the autonomous vehicle; at the moment , the rectangular region is changed by rotation and translation to:

[0030] wherein is the rotation transformation matrix, is the translation transformation vector; the planar space occupied by the autonomous vehicle is described as a convex set during the autonomous parking process, and each obstacle is modeled as a polygon of the following form:

[0031] wherein denotes the number of obstacles, and are related to the size of the obstacle, denotes the number of polygon edges; is a compact convex set with a non-empty relative interior; the obstacle avoidance constraint condition of approximating the autonomous vehicle and obstacles as convex polygons is represented as:

[0032] wherein denotes the minimum safety distance required to complete autonomous parking, and for any , there exist and such that the obstacle avoidance constraint:

[0033] holds; this obstacle avoidance constraint expression has the characteristics of smoothness and differentiability, reducing the computational burden of the optimization solver.

[0034] Step 4: Design an artificial reference-based terminal-free model predictive control algorithm: Autonomous parking scenarios with other vehicles irregularly parked as shown in Fig. 1 Figure 2 The traditional model predictive control method cannot fully consider the dynamic characteristics of the autonomous vehicle, and due to the inaccuracy of the vehicle model, unknown external disturbance factors and insufficient consideration of collision risks, the autonomous parking reference trajectory output by the planning module may not be accurately tracked by the autonomous vehicle. When the reference trajectory is unreachable, the model predictive algorithm based on artificial reference overcomes this defect by guiding the autonomous vehicle to track the optimal reachable trajectory; the cost function of the artificial reference based on terminal-free model predictive control is defined as follows: wherein, denotes the total step length of the prediction horizon starting at time denotes the step in the prediction horizon, and are positive definite symmetric weight matrices, and denote the artificial reference state and the artificial reference input, respectively, and denote the parking trajectory reference state and the parking trajectory reference input output by the trajectory planning module, respectively; the artificial reference is used as an additional decision variable to avoid the loss of feasibility of the optimization problem caused by the unreachable parking reference trajectory; the construction of this cost function aims to treat the artificial reference as a reachable trajectory planned using the model predictive control algorithm, and guide the actual state and the artificial reference state to gradually converge to the parking trajectory reference state; at time , the artificial reference based on terminal-free model predictive control algorithm is designed, and the structure of the autonomous parking control algorithm is shown in Fig. 2; the optimization problem to be solved is: Figure 3

[0035] wherein, denotes the step in the prediction horizon, denotes the th obstacle; the constraint condition defines the state transition equation of the state prediction trajectory of the autonomous vehicle kinematic control system and the artificial reference trajectory, limits the actual control input and the artificial reference control input, and ensures that the shortest distance between the autonomous vehicle and the obstacle is greater than the specified minimum safety distance; by solving the optimization problem, the state and control input sequence of the optimal prediction trajectory at time are obtained, and the first value of the optimal control input sequence is used for autonomous parking of the autonomous vehicle ​​​​The control signal of the moment; most model predictive control methods need to set appropriate terminal cost and terminal constraint conditions to ensure the stability of the control system and the recursive feasibility of the optimization problem, but in practical applications, terminal constraints will bring additional computational burden, this artificial reference based terminal-free model predictive control algorithm removes the terminal cost or terminal constraint condition when constructing, by selecting a suitable prediction horizon to construct the forward invariant set in the kinematic control system of the autonomous vehicle, the stability of the autonomous vehicle kinematic system and the recursive feasibility of the optimization problem are guaranteed.

[0036] Embodiments: This embodiment uses the autonomous driving platform of Tianjin University as the experimental platform, which is mainly composed of four subsystems: sensing module, navigation module, computing unit and driving system. The sensing module integrates high-resolution front camera, laser radar and millimeter wave radar. The global positioning system and inertial navigation system jointly constitute the hybrid navigation module to realize accurate positioning. NVIDIA Jetson AGX Orin serves as the computing core, responsible for real-time processing of sensor data, decision making and control instructions. The computing unit interacts with the driving system through the communication interface, and the driving system includes the steer-by-wire steering module, the steer-by-wire driving and braking module and the basic system adaptation unit. The experimental scene is set as an irregularly parked obstacle vehicle in a restricted space, and the reference trajectory output by the planning module is as shown in Figure 4 The reference trajectory is obtained by the hybrid A-star algorithm, and the dark filled rectangle represents the irregularly parked vehicle, and the unfilled rectangle frame represents the autonomous vehicle. The reference trajectory curve connection exists curvature mutation, and the distance to the obstacle is too close, making the autonomous parking control difficult. The autonomous parking control method of the autonomous vehicle in the irregular scene proposed in this embodiment is used, and the parameters of the weight matrix , , and are obtained after debugging, and the experimental process is as shown in Figure 5 The black curve line represents the motion trajectory of the rear axle center point, and the autonomous vehicle always maintains a safe distance from the obstacle during the entire autonomous parking process and accurately parks into the end point of the parking space. The error change between the state of the kinematic control system of the autonomous vehicle and the state of the parking reference trajectory during the experiment is as shown in Figure 6 The error is maintained within a controllable range, and the absolute error with the end point state is as shown in the following table:

Claims

1. A method for controlling autonomous parking of an autonomous driving vehicle in an irregular scene, characterized in that: The following steps are involved: Step 1: Establish a kinematic model for autonomous parking of an autonomous vehicle; Step 2: Design control input constraints for the autonomous parking task; Step 3: Approximate the autonomous driving vehicle and obstacles as convex polygons to design obstacle avoidance constraints; Step 4: Design a terminal-free model predictive control algorithm based on artificial reference.

2. The autonomous parking control method for an autonomous driving vehicle in an irregular scene according to claim 1, characterized in that: In step 1, the autonomous parking kinematics of the autonomous driving vehicle are established as follows: When the vehicle speed is low, assuming that the autonomous vehicle is a rigid body, the front and rear wheels of the vehicle only move in the direction of wheel rolling. The kinematic equation satisfied by the center point of the rear axle of the vehicle is: ， in, and is the center of the vehicle's rear axle in the geodetic coordinate system Axis components and Axis component, is the vehicle's yaw angle, is the vehicle's yaw rate, is the front wheel turning angle of the vehicle, is the speed of the vehicle, is the wheelbase of the vehicle; the control input of the kinematic control system of the autonomous driving vehicle is and In each sampling period The interior remains unchanged, and the kinematic equation of the vehicle rear axle center is discretized as: ， in, represents the current discrete moment, , the vehicle state and control input are expressed as and .

3. The autonomous parking control method for an autonomous driving vehicle in an irregular scene according to claim 2, characterized in that: In step 2, the design process of the control input constraints of the autonomous parking task is as follows: The absolute value of the vehicle's maximum speed during autonomous parking is set to Since the front wheel angle of the vehicle is constrained by the vehicle's mechanical structure, the absolute values ​​of the vehicle's maximum left front wheel deflection angle and the maximum right front wheel deflection angle are set to , so the control input constraints of the autonomous parking task of the autonomous driving vehicle are expressed as: 。 4. The autonomous parking control method for an autonomous driving vehicle in an irregular scene according to claim 3, characterized in that: In step 3, the autonomous driving vehicle and the obstacle are approximately expressed as convex polygons to design the obstacle avoidance constraints. The specific process is as follows: At the initial moment of the autonomous parking process, the planar space occupied by the autonomous vehicle is represented as the following rectangular area: ， in, and are constants related to the length and width of the autonomous vehicle; At this moment, the rectangular area changes through rotation and translation to: ， in, is the rotation transformation matrix, is the translation transformation vector; each obstacle is modeled as a polygon of the following form: ， in Indicates the number of the obstacle. and are related to the size of the obstacle, represents the number of polygon edges; the obstacle avoidance constraint condition for approximating the autonomous driving vehicle and obstacles as convex polygons is expressed as: ， in, Indicates the minimum safety distance required to complete autonomous parking. ,exist and Make the obstacle avoidance constraint: ， Established.

5. The autonomous parking control method for an autonomous driving vehicle in an irregular scene according to claim 4, characterized in that: In step 4, the design process of the terminal-free model predictive control algorithm based on manual reference is as follows: The cost function of the human reference-based terminalless model predictive control is defined as: ， in, express The total step length of the prediction horizon starting at time t, Indicates the first step, and are all positive definite symmetric weight matrices, and represent the artificial reference state and artificial reference input respectively, and They represent the parking trajectory reference state and parking trajectory reference input output by the trajectory planning module respectively; At this moment, the optimization problem of the terminal-free model predictive control algorithm based on artificial reference is formulated as: ， in Indicates the first step, Indicates the An obstacle.

Images