Intelligent automatic driving control method for bicycle

Abstract

The invention discloses a bicycle intelligent automatic driving control method, which comprises the steps of performing path planning according to bicycle positioning information, a bicycle current state and surrounding environment perception information, performing action decision control, controlling each execution mechanism to complete action and realizing bicycle automatic driving, and the action decision control comprises the following steps: acquiring an initial map center line; performing smooth fitting on the central line of the initial map through a segmented quintic polynomial to generate a local planning track; constructing and utilizing a transverse predictive control model, taking the local planning track and the current state of the single vehicle as input, and outputting a transverse control signal; outputting a speed predicted value by using a speed model established based on the acceleration response of the bicycle under a preset condition and taking the current state of the bicycle as input; based on the deviation between the speed predicted value and the current actual state of the bicycle, estimating the current disturbing force of the bicycle; and calculating an expected accelerator force and an expected brake force for speed control according to the speed prediction value and the estimated current disturbance force.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and particularly to an autonomous driving control method for single-vehicle intelligence. Background Art

[0002] With the rapid development of artificial intelligence and autonomous driving technologies, intelligent autonomous driving technology has become an important research direction in the automotive industry. Currently, autonomous driving control technology generally includes key links such as environmental perception, positioning, path planning, and control decision-making. In the prior art, high-precision positioning is the basis of autonomous driving. Although GPS / GNSS (Global Navigation Satellite System) combined with IMU (Inertial Measurement Unit) based on RTK (Real-Time Kinematic) technology can provide centimeter-level positioning, when satellite signals are blocked (such as in urban canyons, tunnels, indoor parking lots) or interfered with, the positioning accuracy will seriously decline or even fail. Although SLAM (Simultaneous Localization and Mapping) technology based on lidar or vision can, to a certain extent, make up for the deficiencies of GNSS, its robustness and long-term consistency still face challenges in the case of lack of obvious features or drastic changes in the dynamic environment. Relying solely on a certain positioning method is difficult to meet the high-precision and high-reliability requirements of L4-level autonomous driving in all scenarios.

[0003] An autonomous vehicle needs to plan a safe, smooth, and traffic-rule-compliant driving trajectory in a complex and dynamic environment. Traditional local path planning methods, such as simple geometric curve fitting (e.g., cubic polynomials or spline curves), although having a relatively small computational load, are insufficient in ensuring the smoothness of the trajectory, especially the continuity of curvature and curvature change rate. This may lead to problems such as steering wheel jitter, poor ride experience, or even unstable control when the vehicle is driving at high speed or making maneuvers such as lane changes, especially in scenarios where precise satisfaction of the starting and ending pose and curvature constraints is required (e.g., requiring the starting and ending points to be in a straight-line driving state, i.e., the curvature is zero).

[0004] Achieving high-precision tracking of the planned trajectory is the key to vehicle control. Traditional control methods (such as PID controllers, pure tracking algorithms, Stanley algorithms, etc.) have limited capabilities in dealing with the non-linear and time-varying characteristics of vehicles as well as multiple constraints (such as steering angle limits, tire side slip limits, ride comfort requirements), especially in high-speed or large-curvature working conditions, and may be difficult to simultaneously ensure tracking accuracy and driving stability. Although model predictive control (MPC) is theoretically more suitable for dealing with such problems, the accuracy of its model, computational efficiency, and how to effectively combine vehicle stability control are still the focus of research.

[0005] Traditional PID-based speed control methods are sensitive to changes in vehicle model parameters (such as load changes) and external disturbances (such as slopes and wind resistance), making it difficult to achieve precise and robust speed tracking. This may lead to speed fluctuations, overshoots, or sluggish responses, affecting ride comfort and traffic efficiency. At the same time, how to achieve intelligent adaptive cruise speed adjustment based on traffic rules, speed limit signs, and map information is also the key to enhancing the autonomous driving experience. Summary of the Invention

[0006] To address the deficiencies in the prior art, the present invention provides the following technical solutions: A single-vehicle intelligent autonomous driving control method includes path planning based on single-vehicle positioning information, the current state of the single vehicle, and surrounding environment perception information, and performing action decision control to control each actuator to complete actions and achieve single-vehicle autonomous driving. The action decision control includes: Map centerline processing: Obtain the initial map centerline and perform smooth fitting on the initial map centerline through piecewise quintic polynomials to generate a local planning trajectory; Lateral control: Construct and utilize a lateral predictive control model, with the local planning trajectory and the current state of the single vehicle as inputs, and output a lateral control signal; Longitudinal control: Utilize a speed model established based on the acceleration response of the single vehicle under preset conditions, with the current state of the single vehicle as the input, and output a speed prediction value; Based on the deviation between the speed prediction value and the current actual state of the single vehicle, estimate the interference force currently acting on the single vehicle; According to the speed prediction value and the estimated current interference force, calculate the desired throttle and desired braking force for speed control.

[0007] Preferably, the constraint conditions set when performing smooth fitting on the initial map centerline through piecewise quintic polynomials include: The curvature of the local planning trajectory at the initial position and the target position of each trajectory segment formed by the fitting is zero.

[0008] Preferably, the step of performing smooth fitting on the initial map centerline through piecewise quintic polynomials to generate a local planning trajectory further includes the following steps: S101. For the currently to-be-generated trajectory segment, set its position and heading angle at the initial position and the target position of this segment as further constraint conditions; S102. Based on the position, the heading angle, and the constraint condition that the curvature at the initial position and the target position of this segment is zero set in step S101, solve the coefficients of the piecewise quintic polynomial for generating this current trajectory segment; S103. Generate this current trajectory segment according to the coefficients solved in step S102 as a part of the local planning trajectory.

[0009] Preferably, the lateral prediction control model takes the local planned trajectory and the current state of the bicycle as inputs and outputs a lateral control signal, specifically including: S201. Establish a prediction model based on the kinematics of the vehicle model; S202. Define a cost function within the prediction time domain, which aims to minimize the deviation between the predicted state and the ideal state, the magnitude of the control input or the deviation from the ideal control input, and the deviation between the predicted terminal state and the ideal terminal state; S203. Set the constraint conditions for the optimization solution, and the constraint conditions include: the kinematic constraints of the vehicle model, the constraints with the current state of the bicycle as the initial state, and the control input constraints; S204. By solving the optimization problem in each control cycle, minimize the cost function on the premise of satisfying the constraint conditions, obtain the optimal control input sequence, and generate the lateral control signal according to the optimal control input sequence.

[0010] Preferably, the lateral control further includes an adjustment step for enhancing the yaw stability of the vehicle, specifically including: S211. Estimate the current sideslip angle of the center of mass of the bicycle based on a preset two-degree-of-freedom vehicle model and a state observer; S212. Conduct a yaw stability analysis of the vehicle to determine the yaw stability conditions in the current state; S213. Calculate the additional yaw moment for satisfying the yaw stability conditions, and adjust the lateral control signal according to the additional yaw moment.

[0011] Preferably, the state variables used in the lateral prediction control model include: lateral error, the change rate of the lateral error, heading error, and the change rate of the heading error.

[0012] Preferably, the map centerline processing further includes the optimization of the local planned trajectory: S111. Generate multiple alternative local trajectories covering a predetermined time range; S112. Set an evaluation function for evaluating the alternative local trajectories, and the evaluation function at least includes a cost item for evaluating the lateral offset of the bicycle from the map centerline and a cost item for evaluating the comfort of the trajectory; S113. Calculate the evaluation function value corresponding to each alternative local trajectory; S114. Select the alternative local trajectory with the optimal evaluation function value as the planned trajectory.

[0013] Preferably, the method for obtaining the current disturbance force includes: S301. Calculate the speed residual between the actual measured speed of the bicycle and the predicted speed value; S302. Generate an estimated value of the interference force by performing proportional-integral processing on the speed residual.

[0014] Beneficial effects In the present invention, by using a piecewise quintic polynomial to smoothly fit the center line of the map, and taking the curvatures of the starting point and the ending point (e.g., set to zero) as constraint conditions, a local planning trajectory with more continuous and smoother curvature and curvature change rate can be generated. This effectively avoids the problem of steering wheel jitter caused by uneven trajectories, especially in scenarios such as high-speed driving or when precise control of the starting and ending states (such as straight-line merging or leaving) is required, significantly improving the driving stability and ride comfort of the vehicle.

[0015] The present invention uses model predictive control for lateral control, which can prospectively consider the vehicle dynamics and trajectory deviations within a future period of time, and systematically optimize the control input (such as the steering angle), so as to achieve a higher-precision tracking of the planned trajectory. At the same time, MPC can explicitly handle the kinematic / dynamic constraints of the vehicle and the control quantity constraints, ensuring the safety and feasibility of the control. Combined with the optional yaw stability correction based on the two-degree-of-freedom model and the state observer, it can actively estimate and compensate for potential instability risks, further improving the driving stability of the vehicle under dynamic limits or low-adhesion road surfaces.

[0016] The present invention adopts a speed model established based on the vehicle's acceleration response, and combines the analysis of the real-time speed deviation to estimate the total interference force currently received (including unmodeled factors such as slope, load, and wind resistance), enabling the longitudinal controller to prospectively calculate the desired driving / braking force according to the predicted speed and the estimated interference force. Compared with traditional PID control, this method can more effectively suppress the influence of external disturbances on the speed, improve the accuracy and robustness of speed tracking, reduce unnecessary acceleration and deceleration, enhance ride comfort and potential fuel / energy economy, and better adapt to changes in the vehicle's own state (such as load). Brief description of the drawings

[0017] Figure 1 It is a schematic flow chart of a single-vehicle intelligent autonomous driving control method provided in a preferred embodiment of the present invention; Figure 2 It is a schematic diagram comparing the trajectory effects generated by the cubic polynomial adopted in the prior art in the field and the quintic polynomial adopted in the present invention. Detailed implementation manners

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings. In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.

[0019] Embodiment 1 As Figure 1 shown, the present invention provides a method for intelligent autonomous driving control of a bicycle, including performing path planning based on bicycle positioning information, the current state of the bicycle, and surrounding environment perception information, and performing motion decision-making control to control each actuator to complete actions and achieve autonomous driving of the bicycle.

[0020] Among them, the motion decision-making control refers to the entire calculation and processing process of converting the planned driving intention or reference path (such as the center line of the map) into specific and executable vehicle low-level control commands after the path planning at a higher level (such as global path planning or behavior decision-making, which has determined the general driving route or intention) is completed; it is the key bridge connecting "macro path planning" and "micro vehicle execution", and it is responsible for converting the abstract planning intention of "where to go and how to go" into precise and real-time control commands (Commands) sent to the vehicle low-level actuators (steering, drive, and braking systems) through specific trajectory generation, lateral control, and longitudinal control algorithms, so as to directly drive the vehicle to complete specific driving actions (such as keeping in the lane, turning, accelerating, and decelerating).

[0021] The motion decision-making control provided by the present invention includes: Map center line processing: Obtain the initial map center line, and perform smooth fitting on the initial map center line through a piecewise fifth-order polynomial to generate a local planning trajectory.

[0022] It should be understood that the initial map centerline represents the geometric center of the lane or path that the vehicle expects to travel on. However, it may be directly sourced from a high-precision map database or generated as a sequence of discrete points or a rough curve by a higher-level path planning module (such as a task- or behavior-based planner) based on navigation tasks and real-time environmental perception. However, such an initial centerline often has inherent problems. For example, it may be composed of discrete points, have noise, or be geometrically non-smooth, especially with discontinuous or sudden changes in curvature. If the vehicle directly attempts to track such a centerline, especially at high speeds, it will cause frequent and drastic adjustments to the control commands (especially steering commands), resulting in steering wheel jitter, reducing ride comfort, and even affecting driving stability. Therefore, it must be subjected to smooth fitting processing.

[0023] In the prior art, autonomous driving trajectory planning typically uses cubic polynomials or cubic B-spline functions to generate a cluster of candidate trajectories. The current position and heading angle of the vehicle are used as initial state constraints, and the positions and heading angles of the upsampled target points on the road are used as target state constraints. Based on this, a curved trajectory connecting the initial state and the target state is generated. However, a cubic polynomial can only guarantee the first-order continuity of the steering angle and cannot guarantee the continuity of the steering angular velocity (i.e., second-order continuity). This technical limitation has relatively little impact under low-speed driving conditions, and the discontinuity can be ignored. However, under high-speed driving conditions, such discontinuities will be significantly amplified, leading to unstable oscillations in the steering system. In addition, cubic polynomials have inherent limitations in dealing with specific boundary conditions. As Figure 2 shown, when the curvatures of the vehicle's initial state and target state are both zero, the cubic polynomial cannot simultaneously satisfy the zero-curvature constraint conditions at these two points (shown by the dashed lines in the figure), resulting in a sudden change in curvature during the local trajectory update process and causing unstable steering control.

[0024] In view of this, the present invention innovatively uses a fifth-degree polynomial method to generate a local planning trajectory. The fifth-degree polynomial has sufficient mathematical degrees of freedom to simultaneously satisfy the continuity constraints of position, heading angle, and curvature. In the actual fitting process, the algorithm processes the initial map centerline in segments and introduces the curvature values of the initial state and the target state as additional constraint conditions into the polynomial solution process. The fifth-degree polynomial function is applied to each path segment for precise fitting, while ensuring a high-order smooth transition at the segment joints.

[0025] Specifically, the fifth-degree polynomial is a function in the following form: ; Among them, \(x(t)\) is the horizontal coordinate at time \(t\); \(y(t)\) is the vertical coordinate at time \(t\). They are respectively determined by 6 coefficients (\(a_0\) to \(a_5\), \(b_0\) to \(b_5\)), so there is enough freedom (6 coefficients) to simultaneously satisfy the three key geometric constraints of the position, tangent direction (corresponding to the heading angle), and curvature at the starting and ending points of each path segment, and even the rate of change of curvature can be constrained. By solving these constraint conditions to determine the polynomial coefficients, a globally smooth (at least reaching curvature continuity, i.e., \(C^2\) continuity) local planning trajectory is finally pieced together. The generated trajectory not only closely adheres to the intention of the original centerline, but more importantly, ensures a high degree of smoothness in its inherent geometric properties, providing a high-quality tracking target for subsequent lateral and longitudinal controllers, thereby ensuring that the vehicle can execute the autonomous driving task smoothly, precisely, and comfortably.

[0026] In the present invention, the entire path is not represented by a single quintic polynomial function, but is cut into multiple continuous segments. Each path segment is described by its own independent set of quintic polynomials. These segments are smoothly connected at the connection points (referred to as nodes or knots). The advantage of using segmented processing is that relatively low-order polynomials (quintic in this scenario to meet sufficient constraints) can be used to fit complex paths, avoiding the Runge phenomenon that may occur with a single high-order polynomial (i.e., severe oscillations at the endpoints). At the same time, the calculation of each segment is relatively independent, facilitating processing.

[0027] The so-called Smooth Fitting means that given a set of discrete data points or a possibly non-smooth initial curve, a new, smooth curve is sought. This new curve is as close as possible to the original data or curve under a certain metric (such as distance, shape), and it itself meets certain smoothness requirements. The purpose is to obtain a curve that is easy to analyze, calculate, and (in autonomous driving) easy for the vehicle to accurately and comfortably track. In the prior art, it can be achieved by means of spline fitting, etc. In many actual driving scenarios, the vehicle needs to transition between straight driving and curved driving. For example, when entering or exiting a curve, returning to the center of the lane to drive straight after completing a lane change, starting from a standstill or preparing to stop, it is usually expected that the trajectory of the vehicle is smooth and natural at the transition point. If the planned curve segment still has non-zero curvature at the starting point or the ending point, then a sudden change in curvature will occur when connecting the straight line segments. This sudden change will require the steering system of the vehicle to instantaneously change the steering angle, which may cause jerk or vibration of the steering wheel, affecting ride comfort and control stability. Therefore, in some preferred embodiments, the constraint conditions set when smoothing and fitting the initial map center line by means of piecewise quintic polynomials include: the curvature of the local planned trajectory is zero at the initial position and the target position of each trajectory segment formed by the fitting. Through the above constraints, smooth and continuous transition between the trajectory and the straight line segment can be achieved, improving the smoothness, comfort, and control stability of vehicle driving, better simulating natural driving behavior, and meeting the requirements of specific driving tasks. This is an effective means for the present invention to utilize the high-degree-of-freedom characteristics of the quintic polynomial to optimize the trajectory quality. As Figure 2 shown by the solid black line in, this technical solution not only ensures the continuity of the trajectory in the position and heading angle dimensions, but more importantly, achieves the continuity of the curvature and its derivative (curvature change rate), enabling the vehicle to achieve high-precision smooth steering when performing steering control, effectively eliminating the sudden change and oscillation of the steering system, and significantly improving the control stability and driving safety of the vehicle, especially under high-speed working conditions.

[0028] In some preferred embodiments, a method for accurately constructing each component of the local planned trajectory by means of piecewise quintic polynomials is provided, including: S101. For the currently to-be-generated trajectory segment, set its position and heading angle at the initial position and the target position of this segment as further constraint conditions. That is, set the exact position coordinates (x, y) and heading angle of this trajectory segment at the initial moment (t = 0) and the target moment (t = T). These values are obtained from the sampling and analysis of the initial map center line, aiming to ensure seamless docking with the end state (position and heading) of the previously generated trajectory segment and point to the expected starting state of the next trajectory segment. Specifically, the above constraint conditions can be represented by the following formulas in the x and y directions: ; ; ; ; ; wherein, is the horizontal coordinate at time t; is the vertical coordinate at time t; is the speed of the bicycle in the x - direction at time t, is the speed of the bicycle in the y - direction at time t; is the acceleration of the bicycle in the x - direction at time t; theta is the heading angle, k is the curvature, and v is the speed.

[0029] S102. Based on the positions at the initial position and the target position of this section set in step S101, the heading angle, and the constraint condition that the curvature is zero at the initial position and the target position of this section, solve for the coefficients of the piecewise quintic polynomial used to generate the current trajectory section. The above six constraint conditions together describe the six undetermined coefficients contained in each of the two quintic polynomials required for the x - coordinate and y - coordinate of this trajectory section. Based on this, a system of linear equations can be constructed for solution. The system of linear equations is expressed in matrix form as follows: ; wherein, , , respectively represent the horizontal coordinate, speed, and acceleration of the initial position; , , respectively represent the horizontal coordinate, speed, and acceleration of the position at time t1; , , respectively represent the vertical coordinate, speed, and acceleration of the initial position; , , respectively represent the vertical coordinate, speed, and acceleration of the position at time t1; It should be understood that the times t0 and t1 are known and represent the times of the initial position and the end position respectively. Therefore, the X, Y, and T matrices are all known, and the coefficient matrix A in the x - direction and the coefficient matrix B in the y - direction can also be obtained.

[0030] S103. Generate the current trajectory segment based on the coefficients obtained in step S102 as a part of the local planned trajectory. By continuously varying the time parameter t from 0 to T and substituting it into the two fifth-degree polynomial equations x(t) and y(t), the exact vehicle position coordinates corresponding to any moment within this time period are calculated. Connecting these continuous position points generates a smooth trajectory curve for the current segment. This curve segment is then regarded as a component of the entire local planned trajectory, which is stored and used as the starting point for the calculation of the next segment (providing the initial position, heading angle, and curvature constraints for the next segment). By repeatedly executing the three steps S101 to S103, segment by segment generation and splicing are performed, and finally a complete local planned trajectory that meets all constraint conditions and is highly smooth is formed for the vehicle control system to track.

[0031] Furthermore, if the time interval t cannot be directly given at the moments of t0 and t1, assume a constant vehicle speed of v at this time to estimate the time from point A to point B: . When the independent variable of the polynomial curve is time t, and the coefficient matrices A and B have been obtained previously, the position, speed, etc. of each point on the curve are then determined, and the trajectory can be obtained. Specifically, the calculation methods for the heading angle theta, curvature kappa, and curvature change rate dkappa at any point on the local planned trajectory are as follows: ; ; ; Among them, is the lateral coordinate at time t; is the longitudinal coordinate at time t; is the vehicle speed in the x direction at time t, is the vehicle speed in the y direction at time t; is the vehicle acceleration in the x direction at time t; is the vehicle acceleration in the y direction at time t.

[0032] In some other preferred embodiments, the single smooth trajectory generated by the foregoing process may only be the one that is geometrically the smoothest or closest to the center line. However, in order to find the driving path that optimally performs in multiple evaluation dimensions and realizes a decision-making process that pursues higher comfort, efficiency, and intelligence on the basis of meeting basic safety and driving requirements, an optimization method for the local planned trajectory is also provided, including: S111. Generate multiple alternative local trajectories that cover a predetermined time range. It should be understood that there are many methods to generate multiple slightly different but all potentially safe and feasible path options as alternative local trajectories. For example, the method of slightly perturbing the basic trajectory can be adopted, or different parameters (such as different target speeds, different avoidance strategy parameters) can be used to drive the solution of the quintic polynomial to produce different results. The present invention does not make further limitations on this.

[0033] S112. Set an evaluation function for evaluating the alternative local trajectories. This evaluation function at least includes a cost term for evaluating the lateral offset of the single vehicle from the center line of the map, and a cost term for evaluating the comfort of the trajectory.

[0034] The cost term for evaluating the lateral offset of the single vehicle from the center line of the map is used to penalize those trajectories that deviate too far from the expected driving path, so as to ensure that the vehicle stays within the lane as much as possible or follows a predetermined route.

[0035] The cost term for evaluating the comfort of the trajectory is used to penalize those trajectories that include severe acceleration and deceleration, sharp turns or high-frequency oscillations. Usually, it is achieved by calculating the jerk or the magnitude of the curvature and the curvature change rate of the trajectory, with the aim of ensuring a smooth and comfortable riding experience for passengers.

[0036] In some other preferred embodiments, the evaluation function may further include other important factors, such as the collision risk with perceived static or dynamic obstacles (the closer the distance, the higher the cost), the total length or travel time of the trajectory (evaluating efficiency), energy consumption, the deviation from the target speed, etc.

[0037] S113. Calculate the evaluation function value corresponding to each alternative local trajectory. This means that along the time or space process of each alternative trajectory, all relevant cost terms (multiplied by their respective weights) are accumulated (or integrated), and finally a total evaluation value or total cost score of this trajectory is obtained.

[0038] S114. Select the alternative local trajectory with the optimal evaluation function value as the planned trajectory. Obviously, the planned trajectory is the best solution obtained by comprehensively considering path keeping, riding comfort and possible other factors (such as safety, efficiency) under the current evaluation criteria and weight system. This optimization selection process enables autonomous driving to not only mechanically track a line, but also make intelligent decisions in the possibility space to achieve better overall driving performance.

[0039] Lateral control: Construct and utilize a lateral predictive control model, which takes the local planned trajectory and the current state of the bicycle as inputs and outputs a lateral control signal. It should be understood that the lateral control undertakes the core vehicle tracking function, and its fundamental goal is to accurately and smoothly control the steering system of the vehicle so that the vehicle can closely follow the local planned trajectory.

[0040] The lateral model predictive control (MPC) can be realized by establishing a mathematical model that can describe the lateral motion behavior of the vehicle. Specifically, the discrete state space form can be used to capture how key state variables such as the lateral position, orientation angle (yaw angle), lateral velocity, and yaw angular velocity of the vehicle respond to changes in the steering input (control quantity, usually the front wheel steering angle or the steering wheel steering angle). It should be understood that the establishment of the model is based on simplified vehicle kinematics or more complex dynamics principles. Since this part of the content is not the focus of the present invention, it will not be elaborated here, and those skilled in the art can make specific designs according to the existing technology. In some other preferred embodiments, an example method of optimizing and solving through a cost function is given, specifically including: S201. Establish a prediction model based on vehicle model kinematics.

[0041] S202. Define the cost function within the prediction horizon. The cost function aims to minimize the deviation between the predicted state and the ideal state, the magnitude of the control input or its deviation from the ideal control input, and the deviation between the predicted terminal state and the ideal terminal state.

[0042] The design of the cost function aims to minimize the costs in multiple aspects simultaneously: one is the deviation between the predicted state and the ideal state (i.e., the target state on the planned trajectory), which ensures that the vehicle can closely track the planned path and reduce the lateral error and orientation error; the second is the magnitude of the control input or its deviation from the ideal control input, which encourages the controller to use smoother and more economical steering actions, avoid overly aggressive operations, improve ride comfort and reduce actuator wear; the third is the deviation between the predicted terminal state and the ideal terminal state, which helps to ensure that the vehicle is in a good and stable state at the end of the prediction horizon, prepares for the subsequent control cycle, and enhances the long-term stability of the system. By assigning reasonable weights to these different cost terms, trade-offs and adjustments can be made among multiple objectives such as tracking accuracy, control smoothness, and stability to meet the requirements of different driving scenarios.

[0043] S203. Set the constraint conditions for the optimization solution. The constraint conditions include: the vehicle model kinematic constraints, the constraints with the current state of the bicycle as the initial state, and the control input quantity constraints. The constraint conditions clarify the boundaries and rules that must be observed during the optimization solution process, specifically including: 1. Kinematic / dynamic constraints of the vehicle model, i.e., any predicted vehicle state evolution must strictly follow the physical laws described by the mathematical model established in S201; 2. Constraints with the current state of a single vehicle as the initial state, which means that all predictions and optimizations must be based on the actual measured or estimated state of the vehicle at the beginning of the current control cycle (such as the current position error, heading error, etc.) to ensure that the control decision is based on the real situation; 3. Third, control input quantity constraints, which represent the limitations of the vehicle's physical actuators, such as the maximum steering angle limit of the steering wheel or front wheels, the maximum steering angular velocity limit, etc., to ensure that the calculated steering command does not exceed the capabilities of the vehicle hardware.

[0044] It should be understood that there may also be other constraints, such as state constraints like the maximum allowable tire side slip angle, the maximum allowable lateral acceleration (for comfort or safety). Optimizing within the feasible region formed by these constraints is the key to ensuring the effectiveness and safety of the MPC output result.

[0045] S204. By solving the optimization problem in each control cycle to minimize the cost function under the premise of satisfying the constraint conditions, an optimal control input sequence is obtained, and the lateral control signal is generated according to the optimal control input sequence. It should be understood that the optimal control input sequence covers the entire prediction time domain, and the lateral prediction control model will extract one control input (which can be the first one or other control inputs selected according to other preset rules) from the sequence, and the remaining control inputs in the sequence are discarded. In the next control cycle, the above steps are repeated. This way of continuously re - planning and correcting the control strategy based on the latest information makes the MPC have strong robustness and can effectively cope with model uncertainties, external disturbances, and environmental changes, thus achieving continuous, accurate, and stable lateral control.

[0046] Those skilled in the art can know that when the vehicle is approaching or in a potentially unstable state (such as high - speed cornering, emergency obstacle avoidance, or driving on a low - adhesion road surface), active intervention and adjustment of control are required to maintain or restore the vehicle's driving stability and prevent situations such as sideslip, fishtailing, or even loss of control. In some preferred embodiments, the lateral control further includes adjustment steps for enhancing the vehicle's yaw stability, specifically including: S211. Estimate the current centroidal side slip angle of the vehicle based on a preset two-degree-of-freedom vehicle model and a state observer. Those skilled in the art can know that the centroidal side slip angle refers to the angle between the velocity direction at the vehicle's centroid and the vehicle's longitudinal axis, and it is the most direct indicator to measure the degree of vehicle sideslip. Since there is no sensor that can directly measure this angle, it is usually difficult to directly measure. This embodiment provides an indirect estimation based on a preset two-degree-of-freedom (2-DOF) vehicle model. Specifically, it can be a simplified vehicle dynamics model that mainly considers the lateral motion of the vehicle and the yaw motion around the vertical axis, and it captures how the lateral velocity and yaw angular velocity of the vehicle respond to the steering input and tire lateral force. To use this model for estimation, it is also necessary to combine the actual measurement information of the vehicle (such as yaw angular velocity, lateral acceleration, wheel speed, steering angle, etc.), and estimate the current centroidal side slip angle of the vehicle in real time by designing and running a state observer (such as a Kalman filter or its variants, such as the extended Kalman filter EKF, unscented Kalman filter UKF, or sliding mode observer, etc.). This state observer can fuse the model prediction and sensor measurement values, overcome measurement noise and model uncertainty.

[0047] S212. Conduct a vehicle yaw stability analysis to determine the yaw stability condition in the current state. It should be understood that the yaw stability of the vehicle is usually closely related to two state variables, the centroidal side slip angle and the yaw angular velocity. The stability analysis can be based on various theories or methods, such as phase plane analysis, linear or nonlinear stability theories (such as Lyapunov stability theory), etc., and the present invention does not make further limitations.

[0048] S213. Calculate the additional yaw moment for satisfying the yaw stability condition, and adjust the lateral control signal according to the additional yaw moment. The additional yaw moment is a virtual or actually required additional rotational moment applied to the vehicle around the vehicle's vertical axis, and its function is to help the vehicle resist excessive steering or insufficient steering tendency and pull the vehicle state back into the stable region. For example, if the vehicle exhibits excessive steering (fishtailing), a reverse (rotation-inhibiting) yaw moment needs to be applied; if there is insufficient steering (understeering), a yaw moment that helps the vehicle turn needs to be applied. Its magnitude and direction can be calculated according to the stability condition determined in step S212 and the distance between the current vehicle state and the stability boundary.

[0049] Longitudinal control: Use a speed model established based on the acceleration response of the vehicle under preset conditions, take the current state of the vehicle as the input, and output a speed prediction value; estimate the current disturbing force on the vehicle based on the deviation between the speed prediction value and the current actual state of the vehicle; calculate the desired throttle and desired braking force for speed control according to the speed prediction value and the estimated current disturbing force.

[0050] The longitudinal control is responsible for managing the forward and backward movement of the vehicle. The core task is to precisely control the speed of the vehicle so that it can safely, smoothly and efficiently follow the planned speed instruction or achieve adaptive cruise control.

[0051] The speed model can be a parameterized mathematical model (such as transfer function, state space equation), or a data-driven model (such as look-up table, neural network). This model takes the current state of the vehicle (including at least the currently measured speed, and possibly also the gear, current control input, etc.) as input, and then outputs a speed prediction value. This prediction value represents how much the vehicle speed should become in the next extremely short time step under the current state, and it predicts the speed evolution without unmodeled disturbances.

[0052] Those skilled in the art can understand that the actual speed response of the vehicle often deviates from the prediction of the above ideal model because there are many factors that are not accurately included in the model or change in real time, such as changes in road slope, changes in real-time wind resistance (headwind / tailwind), changes in vehicle load, changes in rolling resistance (such as changes in road surface material), etc. These unmodeled factors together constitute a disturbing force, which affects the actual acceleration of the vehicle. In some preferred embodiments, the method for obtaining the current disturbing force includes: S301. Calculate the speed residual between the actually measured speed of the vehicle and the speed prediction value. This residual signal directly quantifies the mismatch between the model prediction and the actual situation. If the residual is zero or close to zero, it indicates that the model can well explain the current speed behavior; if the residual persists in being non-zero, it strongly implies that there is an external force (i.e., disturbing force) not considered by the model affecting the movement of the vehicle - a positive residual may mean there is assistance (such as going downhill), and a negative residual may mean there is resistance (such as going uphill or headwind). Therefore, the speed residual is the original input signal for detecting and quantifying the disturbing effect.

[0053] S302. Generate an estimated value of the disturbing force by performing proportional-integral processing on the speed residual. Those skilled in the art can understand that simply regarding the speed residual directly as the disturbing force is not accurate because it is only the result of the disturbing effect (speed change), rather than the disturbing force itself. This step converts the measurable speed deviation into a quantitative estimate of the non-directly measurable disturbing force, providing a key basis for subsequent longitudinal control compensation. Among them, proportional-integral (PI) processing is a conventional technique in the art, and the present invention will not be further described.

[0054] The calculation of the desired throttle opening (for acceleration or speed maintenance) and the desired braking force (for deceleration) for final speed control is simultaneously based on (or, in a feed-forward manner, takes into account) the aforementioned speed prediction value (associated with the basic driving / braking force required to reach the target speed) and the currently estimated current disturbance force (as a compensation term for the basic driving / braking force). For example, if a large uphill resistance is estimated (the disturbance force is negative), the controller will increase the additional throttle opening on top of the basic throttle opening calculated by feed-forward to compensate for this resistance, so as to ensure that the vehicle can still achieve the target acceleration or maintain the target speed. On the contrary, if a downhill assist is estimated (the disturbance force is positive), the throttle may be reduced or even a slight brake may be required. This longitudinal control strategy that combines model prediction and real-time estimation and compensation of disturbance forces makes the system more robust and adaptive to external environmental changes and vehicle own parameter changes (such as load), enables more accurate and smoother speed tracking, and improves the comfort and reliability of autonomous driving.

[0055] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A bicycle intelligent automatic driving control method, which includes performing path planning based on bicycle positioning information, the current state of the bicycle, and surrounding environment perception information, and making motion decision control to control each actuator to complete motions, so as to achieve automatic driving of the bicycle. It is characterized in that, The action decision-making control includes: Map centerline processing: Obtain the initial map centerline, and perform smooth fitting on the initial map centerline through piecewise quintic polynomials to generate a local planning trajectory; Lateral control: Construct and utilize a lateral predictive control model, with the local planning trajectory and the current state of the bicycle as inputs, and output a lateral control signal; Longitudinal control: Utilize a speed model established based on the acceleration response of the bicycle under preset conditions, with the current state of the bicycle as the input, and output a speed prediction value; Based on the deviation between the speed prediction value and the current actual state of the bicycle, estimate the interference force currently received by the bicycle; According to the speed prediction value and the estimated current interference force, calculate the desired throttle and desired braking force for speed control.

2. The bicycle intelligent automatic driving control method according to claim 1, characterized in that: The constraint conditions set when performing smooth fitting on the initial map centerline through piecewise quintic polynomials include: The curvature of the local planning trajectory at the initial position and the target position of each trajectory segment formed by the fitting is zero.

3. The bicycle intelligent automatic driving control method according to claim 2, characterized in that, The step of performing smooth fitting on the initial map centerline through piecewise quintic polynomials to generate a local planning trajectory further includes the following steps: S101. For the currently to-be-generated trajectory segment, set the position and heading angle at the initial position and the target position of this segment as further constraint conditions; S102. Based on the position at the initial position and the target position of this segment set in step S101, the heading angle, and the constraint condition that the curvature at the initial position and the target position of this segment is zero, solve the coefficients of the piecewise quintic polynomial for generating this current trajectory segment; S103. Generate this current trajectory segment according to the coefficients solved in step S102 as a part of the local planning trajectory.

4. The bicycle intelligent automatic driving control method according to claim 1, characterized in that, The construction of the lateral predictive control model, with the local planning trajectory and the current state of the bicycle as inputs, and output a lateral control signal, specifically includes: S201. Establish a prediction model based on the kinematics of the vehicle model; S202. Define the cost function within the prediction time domain, and the cost function aims to minimize the deviation between the predicted state and the ideal state, the magnitude of the control input or the deviation from the ideal control input, and the deviation between the predicted end state and the ideal end state; S203. Set the constraint conditions for optimization solution, and the constraint conditions include: the kinematic constraints of the vehicle model, the constraint with the current state of the bicycle as the initial state, and the control input quantity constraint; S204. By solving the optimization problem in each control cycle, minimize the cost function on the premise of satisfying the constraint conditions, obtain the optimal control input sequence, and generate the lateral control signal according to this optimal control input sequence.

5. The bicycle intelligent automatic driving control method according to claim 4, characterized in that, The lateral control further includes an adjustment step for enhancing the yaw stability of the vehicle, specifically including: S211. Based on a preset two-degree-of-freedom vehicle model and a state observer, estimate the current sideslip angle of the center of mass of the bicycle; S212. Conduct vehicle yaw stability analysis to determine the yaw stability condition under the current state; S213. Calculate the additional yaw moment for satisfying the yaw stability condition, and adjust the lateral control signal according to this additional yaw moment.

6. The bicycle intelligent automatic driving control method according to claim 4, characterized in that, The state variables used in the lateral prediction control model include: lateral error, the change rate of the lateral error, heading error, and the change rate of the heading error.

7. The bicycle intelligent automatic driving control method according to claim 1, characterized in that, The map centerline processing further includes the optimization of the local planned trajectory: S111. Generate multiple alternative local trajectories covering a predetermined time range; S112. Set an evaluation function for evaluating the alternative local trajectories, and the evaluation function at least includes a cost term for evaluating the lateral offset of the single vehicle from the map centerline and a cost term for evaluating the comfort of the trajectory; S113. Calculate the evaluation function value corresponding to each alternative local trajectory; S114. Select the alternative local trajectory with the optimal evaluation function value as the planned trajectory.

8. The bicycle intelligent automatic driving control method according to claim 1, characterized in that, The method for obtaining the current interference force includes: S301. Calculate the speed residual between the actually measured speed of the single vehicle and the speed prediction value; S302. Generate an estimated value of the interference force by performing proportional-integral processing on the speed residual.

Images