A cooperative braking method and device for automatic driving vehicle lane changing

Abstract

The present application relates to the field of braking, in particular to a cooperative braking method and device for automatic driving vehicle lane changing. The method comprises: obtaining ego state, environment vehicle state and target lane average speed information; based on a trajectory prediction model and a preset safety threshold, calculating a space-time occupation space containing static and dynamic occupation; taking driving into the target lane as the termination condition, constructing a parameterized candidate lateral and longitudinal trajectory set and performing coupling feasibility verification, and screening feasible trajectory clusters; determining the optimal lateral trajectory through a comprehensive evaluation function containing lane changing time, comfort and longitudinal space accommodation capacity, and then solving the longitudinal optimization trajectory with speed matching as the core; solving the expected steering angle and speed sequence, and generating braking instructions according to the speed difference. The present application solves the problems of lack of space-time dynamic constraints in existing lane changing braking, safety hazards caused by separation of lateral and longitudinal control, low efficiency and poor comfort.

Description

Technical Field

[0001] This invention relates to the field of braking, and more particularly to a cooperative braking method and apparatus for lane changing in autonomous vehicles. Background Technology

[0002] With the deep integration of autonomous driving and vehicle-to-everything (V2X) technologies, heterogeneous speed lane-changing scenarios on structured roads have become one of the core scenarios in the operation of autonomous connected vehicles. In this scenario, autonomous vehicles need to change lanes from high-speed lanes to low-speed lanes. This not only presents a complex environment where both human-driven and autonomous vehicles coexist in the current and target lanes, but also requires precise deceleration and braking to match the speed of vehicles in the target lane, thereby ensuring the safety, efficiency, and smoothness of the lane-changing process. In existing technologies, vehicles typically rely on multi-source devices such as cameras, LiDAR, millimeter-wave radar, and roadside sensing units to acquire environmental and self-state information. After data fusion, an obstacle distribution pattern is formed and a safe distance is estimated. Subsequently, a collision-free path and speed profile curve are generated, which are then converted into steering, throttle, and braking control signals through trajectory tracking to complete the lane change. However, existing technologies still have many shortcomings that urgently need to be addressed: First, the perceived information is mostly used only for static spatial division, lacking dynamic calculation and evolution analysis of the spatiotemporal occupancy information of environmental obstacles. This results in the subsequent control unit being unable to fully utilize the vehicle's motion trend in the environment, leading to a lack of optimal braking decisions. Second, the control logic mostly adopts a path and speed decomposition approach, with steering control and throttle and braking control being decided independently in sequence, ignoring the overall nature of vehicle motion. This leads to insufficient coordination between braking and steering during lane changes, making it difficult to guarantee overall control quality. Third, existing braking methods, such as the "Braking Method Based on the Comfort of Autonomous Vehicles" disclosed in CN119568090A, although dynamically adjusting the Q matrix of the model predictive control through the TD3 deep reinforcement learning algorithm to optimize parameters such as pitch angle during braking to improve comfort, focus on simple braking conditions (emergency or normal braking). They do not address the coupling characteristics of lateral motion and longitudinal braking in lane change scenarios, nor do they consider the core requirement of matching the vehicle speed with the target lane during lane changes. They cannot solve the comprehensive problem of balancing lane change efficiency, safety distance, comfort, and speed matching when braking during lane changes at heterogeneous vehicle speeds. Furthermore, existing lane-change braking methods often fail to accurately depict the dynamic evolution of vehicles in the environment, and the definition of dynamic danger zones lacks timeliness and accuracy. This leads to delayed or inaccurate braking command generation, easily resulting in problems such as excessive braking, insufficient speed matching, or loss of safe distance control, seriously affecting passenger experience and driving safety. Therefore, how to integrate multi-dimensional state information, accurately depict spatiotemporal dynamic constraints, and achieve coordinated optimization of lateral and longitudinal trajectories in heterogeneous lane-change scenarios on structured roads, thereby generating coordinated braking commands that balance safety, efficiency, comfort, and speed matching, has become a key issue that urgently needs to be addressed in the field of lane-change braking for autonomous vehicles. Summary of the Invention

[0003] This invention provides a cooperative braking method and device for lane changing in autonomous vehicles, aiming to solve the problems of lack of spatiotemporal dynamic constraints in lane changing braking, safety hazards caused by separation of lateral and longitudinal control, low efficiency and poor comfort in the prior art.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] A cooperative braking method for lane changing in an autonomous vehicle includes the following steps: Obtain information including the vehicle's status, the status of vehicles in the environment, and the average speed of vehicles in the target lane; Based on the vehicle's environmental status, the spatiotemporal occupancy space, including static and dynamic occupancy, is calculated using a trajectory prediction model and preset safety thresholds. Based on the vehicle's status, a parameterized set of candidate lateral and longitudinal trajectories is constructed with entering the target lane as the termination condition. Using the aforementioned spatiotemporal occupancy space, the feasibility of coupling the candidate lateral and longitudinal trajectory sets is verified, and feasible trajectory clusters that simultaneously satisfy lateral and longitudinal constraints are selected. The optimal lateral trajectory is determined from the feasible trajectory family based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity. Using the optimal lateral trajectory as a constraint, the longitudinal trajectory optimization problem with velocity matching as the core objective and constrained by the spatiotemporal space is solved to obtain the optimal longitudinal trajectory; Calculate the desired steering angle sequence and desired velocity sequence based on the optimal lateral trajectory and the optimal longitudinal trajectory; Based on the difference between the desired speed sequence and the current actual vehicle speed, a vehicle braking command is generated to brake the vehicle.

[0006] Optionally, the steps for calculating the spatiotemporal space occupancy, including both static and dynamic occupancy, specifically include: Based on the physical length L of the vehicle in the environment and the preset static safety redundancy distance Calculate the static occupancy radius And based on this, define the static occupancy space. , where x i Let be the longitudinal position of the i-th environmental vehicle; calculate the vehicle collision time TTC based on the relative motion relationship between the vehicle and the environmental vehicles, when the distance between the two vehicles decreases to... hour ,in This represents the difference between the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicles. The difference between the longitudinal velocity of the vehicle and the longitudinal velocity of the surrounding vehicles is calculated using the same subtraction order for dx and dv, combined with a preset TTC safety threshold. Define dynamic space usage ; The spatiotemporal occupied space is the static occupied space. With the aforementioned dynamic occupied space The union of the represented danger zones.

[0007] Optionally, the trajectory prediction model is a hybrid deep learning model combining a convolutional neural network and a long short-term memory network; the convolutional neural network is used to extract spatial correlation features between the states of multiple environmental vehicles at the same time, and the long short-term memory network is used to extract temporal features of the state changes of a single environmental vehicle over a historical period; the hybrid deep learning model, by fusing the spatial correlation features and the temporal features, processes the historical and current state sequences of environmental vehicles to predict the trajectory of environmental vehicles over a future period, and the predicted future trajectory is used to calculate the dynamically occupied space. The spatiotemporal evolution.

[0008] Optionally, the step of constructing a parameterized set of candidate lateral and longitudinal trajectories specifically includes: In the Frenet coordinate system, the origin is the vehicle's current position, the longitudinal axis is parallel to the centerline of the current lane, and the lateral axis is perpendicular to the longitudinal axis and points to the side of the target lane; a fifth-order polynomial parametric curve is used to represent the lateral movement. With longitudinal movement Modeling is performed, where: ; For time variables, These are the parameters of the curve in the horizontal and vertical directions, respectively; the initial horizontal coordinates of the vehicle state are... Initial lateral velocity Initial lateral acceleration Initial ordinate Initial longitudinal velocity Initial longitudinal acceleration As the fifth-degree polynomial at the initial time Boundary conditions; termination time The lateral position is set as the centerline coordinate of the target lane. The lateral velocity and lateral acceleration are set to zero as the lateral termination boundary condition, and the termination time is set to... vertical position Longitudinal velocity Longitudinal acceleration As variable sampling parameters to form longitudinal termination boundary conditions; by systematically sampling different , , and The system combines and solves for the polynomial coefficients that satisfy all boundary conditions, generating in batches the smooth sets of candidate lateral trajectories and the candidate longitudinal trajectories.

[0009] Optionally, a feasibility verification of coupling the candidate lateral and longitudinal trajectory sets can be performed, specifically including: For each candidate lateral trajectory, based on the lateral position of the vehicle it describes... As time changes, the vehicle's position relative to the current lane line and the target lane line is determined in real time, and the set of environmental vehicles that constitute spatial constraints on the vehicle at each moment is dynamically determined based on the positional relationship. The rules for determining the set are as follows: When a vehicle has not crossed the lane lines, only the vehicle in front in the current lane is considered; Once the vehicle has fully entered the target lane, only the vehicles in front of and behind in the target lane are considered. When a vehicle is crossing a lane line, both the vehicle in front in the current lane and the vehicles in front and behind in the target lane should be considered simultaneously. For each candidate longitudinal trajectory, examine the vehicle motion state point (x(t), v) jointly determined by it and a candidate lateral trajectory at the same time t. x (t)) whether it falls into the union of the spatiotemporal occupied space corresponding to the set of environmental vehicles at that time; if for a combination of a candidate lateral trajectory and a candidate longitudinal trajectory, the motion state point corresponding to all sampling times t does not fall into any occupied space, then the combination is determined to be safe and feasible, and it is added to the feasible trajectory cluster.

[0010] Optionally, the comprehensive evaluation function J is specifically expressed as: ; Where ω1, ω2, and ω3 are positive weighting coefficients, and ω4 is a negative weighting coefficient; the first term This represents lane-changing efficiency and encourages shorter lane-changing times; the second item... The third term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration. The fourth term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration, which are used together to evaluate the smoothness and comfort of lateral motion. It is an indicator function whose function value is equal to the number of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the currently evaluated candidate lateral trajectory. The fourth term is used to evaluate the size of the feasible space reserved by the lateral trajectory for subsequent longitudinal velocity planning. By calculating and comparing the J values ​​corresponding to all candidate lateral trajectories in the feasible trajectory cluster, the candidate lateral trajectory that minimizes the J value is selected as the optimal lateral trajectory. Specifically: t is the end time of the lane change; t0 is the start time of the lane change; a y This refers to lateral acceleration; a x For longitudinal acceleration; jerk y jerk is the lateral jerk, specifically the first derivative of lateral acceleration with respect to time. x C(t) represents the longitudinal jerk, i.e., the first derivative of the longitudinal acceleration with respect to time; tw is the termination time of the longitudinal trajectory optimization, and the length of the longitudinal optimization time window corresponding to tw is fixed in seconds; I is the indicator function, whose function value is equal to the number of elements in the set; C(t) end ) is the feasible longitudinal trajectory pairing set, representing the set of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the current candidate lateral trajectory.

[0011] Optionally, the mathematical model for the longitudinal trajectory optimization problem with velocity matching as the core objective is as follows: Objective function: ; Constraints: ; Among them, the first two terms of the objective function and These represent the square integrals of longitudinal acceleration and jerk, respectively, used to ensure comfort during longitudinal braking or acceleration; the third term of the objective function For speed matching items, where for The speed of the main vehicle is constantly monitored, with entering the target lane as the termination condition. The termination time tw for longitudinal trajectory optimization is set. Given the average speed in the target lane, the third term of the direct drive control algorithm calculates a longitudinal trajectory that makes the vehicle's speed approach the target lane speed at the end of the lane change; R sta The static occupied space; x is the longitudinal position of the vehicle; v x R is the longitudinal speed of the vehicle. dyn It dynamically occupies space.

[0012] A cooperative braking device for lane changing in an autonomous vehicle includes: The information acquisition module is configured to acquire information including the vehicle's status, the status of vehicles in the environment, and the average speed of vehicles in the target lane. The spatiotemporal occupancy calculation module is configured to calculate the spatiotemporal occupancy space, including static occupancy space and dynamic occupancy space, based on the vehicle state in the environment, through a trajectory prediction model and a preset safety threshold. The trajectory planning module is configured to construct a parameterized set of candidate lateral and longitudinal trajectories based on the vehicle's state and with entering the target lane as the termination condition. It uses the spatiotemporal occupancy space output by the spatiotemporal occupancy calculation module to perform coupled feasibility verification on the set of candidate lateral and longitudinal trajectories, and selects feasible trajectory clusters. Based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity, it determines the optimal lateral trajectory from the feasible trajectory clusters. Using the optimal lateral trajectory as a constraint, it solves the longitudinal trajectory optimization problem with speed matching as the core objective and constrained by the spatiotemporal occupancy space, thereby obtaining the optimal longitudinal trajectory for deceleration and braking. The control quantity calculation module is configured to calculate the desired steering angle sequence and desired speed sequence of the vehicle during the lane changing process based on the optimal lateral trajectory and the optimal longitudinal trajectory output by the trajectory planning module. The braking command generation module is configured to generate a vehicle braking command for the vehicle braking control unit based on the difference between the desired speed sequence output by the control quantity calculation module and the current actual vehicle speed.

[0013] Optionally, the spatiotemporal occupancy calculation module includes a static occupancy calculation unit and a dynamic occupancy calculation unit; The static occupancy calculation unit is configured to calculate the static occupancy space based on the physical length L of the environmental vehicle and the static safety redundancy distance. The dynamic occupancy calculation unit integrates the trajectory prediction model and is configured to use the environmental vehicle future trajectory information output by the trajectory prediction model, combined with a preset TTC safety threshold, to calculate the dynamically changing dynamic occupancy space. The trajectory planning module includes a trajectory generation unit, a coupling verification unit, a lateral decision-making unit, and a longitudinal optimization unit. The trajectory generation unit is configured to perform the operation of constructing a parameterized set of candidate lateral and longitudinal trajectories. The coupling verification unit is configured to perform the operation of verifying the coupling feasibility. The lateral decision-making unit is configured to perform the operation of determining the optimal lateral trajectory based on a comprehensive evaluation function. The longitudinal optimization unit is configured to perform the operation of solving the longitudinal trajectory optimization problem with velocity matching as the core objective.

[0014] Optionally, the cooperative braking device for lane changing in an autonomous vehicle is integrated into the vehicle's onboard control unit or braking system domain controller; the information acquisition module is connected to the onboard environmental perception sensor and the vehicle network communication unit; the output of the braking command generation module is connected to the vehicle braking control unit for directly transmitting the control commands required for deceleration and braking.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention acquires multi-dimensional information such as the vehicle's status, the status of surrounding vehicles, and the average speed of the target lane through a system. Based on a trajectory prediction model and a preset safety threshold, it calculates the spatiotemporal occupancy space, including static and dynamic occupancy, constructs a parameterized set of candidate lateral and longitudinal trajectories, and verifies the coupling feasibility. Then, it selects the optimal lateral trajectory through a comprehensive evaluation function, and solves the longitudinal trajectory optimization problem with speed matching as the core. Finally, it generates a braking command based on the difference between the expected speed and the actual vehicle speed. This invention effectively solves the technical problems of existing lane change braking, such as the lack of spatiotemporal dynamic constraints, safety hazards caused by the separation of lateral and longitudinal control, low efficiency, and poor comfort. It achieves deep coordination between braking and lateral lane change during lane change, ensuring that the vehicle can safely, smoothly, and efficiently enter the target lane in complex environments, and significantly improves the overall performance of lane change braking. Furthermore, the specific calculation logic for static and dynamic occupancy space was clarified. Static danger zones were defined using the physical length L of the environmental vehicle and the static safety redundancy distance, while dynamic danger zones were characterized by combining the time-of-collision (TTC) safety threshold. This accurately defined the safety boundaries during lane changes, providing a reliable basis for trajectory feasibility verification and further enhancing the safety and accuracy of braking. A hybrid deep learning model combining convolutional neural networks (CNN) and long short-term memory networks (LSTM) was employed. This model can extract spatial correlation features between multiple environmental vehicles and capture the temporal state changes of a single environmental vehicle, accurately predicting the future trajectory of environmental vehicles. This significantly improved the timeliness and accuracy of dynamic occupancy space calculation, enhancing the real-time adaptability of braking to complex environments. Finally, lateral and longitudinal motion were modeled using fifth-order polynomial parameter curves. Smooth candidate trajectories were generated in batches based on the vehicle's initial state and lane change termination conditions, providing a high-quality trajectory foundation for subsequent optimization and ensuring the smoothness of the braking process from the outset. Based on the dynamic adjustment of the vehicle's lateral position and environmental constraints, precise coupling verification of lateral and longitudinal trajectories was achieved, ensuring that braking at different lane-change stages could adapt to the corresponding environmental constraints, further improving the targeting and safety of braking. The comprehensive evaluation function simultaneously considers lane-change time, comfort, and longitudinal space capacity, ensuring that the selected optimal lateral trajectory can balance multiple requirements, laying a good foundation for longitudinal braking optimization. The longitudinal trajectory optimization model takes speed matching as its core objective, combining acceleration and jerk constraints, enabling the braking process to quickly match the target lane speed while avoiding sudden braking changes, reducing subsequent speed adjustment amplitude, and improving driving stability and comfort. The control device, through modular design, achieves efficient coordination of information acquisition, spatiotemporal occupancy calculation, trajectory planning, control quantity calculation, and braking command generation, engineering the above methods to ensure stable output and reliable execution of braking commands, significantly improving the practicality and operability of the method, and enhancing the feasibility of the solution in real vehicle systems. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the control flow of an embodiment of the cooperative braking method for lane changing of an autonomous vehicle according to the present invention.

[0017] Figure 2 This is a schematic diagram of a module of an embodiment of a cooperative braking device for lane changing in an autonomous vehicle according to the present invention.

[0018] Figure 3 This is a schematic diagram illustrating the application mode of a cooperative braking device for lane changing in an autonomous vehicle according to an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram illustrating a scenario of an embodiment of the cooperative braking method for lane changing of an autonomous vehicle according to the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0021] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0022] Example 1

[0023] like Figure 1 As shown, this embodiment is used in a structured road scenario where an autonomous connected vehicle changes lanes at different speeds from a high-speed lane to a low-speed lane. Through coordinated lateral lane changing and longitudinal braking, a safe, smooth, and efficient lane-changing process is achieved. The following provides a detailed description of each step of the method: First, information including the vehicle's status, the status of surrounding vehicles, and the average speed of vehicles in the target lane is acquired. Acquiring the vehicle's status relies on real-time interaction between the onboard sensor suite and the vehicle control system. The sensor suite includes a lidar mounted on the roof, millimeter-wave radars on the front and rear bumpers, a monocular or binocular camera inside the windshield, and an inertial measurement unit integrated into the vehicle chassis. The lidar scans the surrounding environment by emitting laser beams, acquiring three-dimensional information such as the relative distance and azimuth angle between the vehicle and surrounding objects. Combined with angular velocity and acceleration data output by the inertial measurement unit, it calculates the vehicle's real-time position coordinates, longitudinal velocity, longitudinal acceleration, lateral velocity, lateral acceleration, and steering angle. Acquiring the status of surrounding vehicles is achieved through multi-source sensor data fusion. The camera identifies lane lines, traffic signs, and vehicle outlines. The millimeter-wave radar works in conjunction with the lidar to accurately capture the motion status of vehicles in the current lane, the target lane, and behind the target lane, including the longitudinal displacement, longitudinal velocity, and longitudinal acceleration of each vehicle. This data is transmitted to the computing unit via the vehicle's Ethernet to form an environmental vehicle status dataset. The average speed of the target lane is calculated based on the speed data of multiple vehicles in the target lane within a certain time window, and is obtained through a statistical averaging algorithm. The length of the time window can be dynamically adjusted according to the road traffic density. When the traffic flow is dense, the window length is shortened to ensure real-time performance, and when the traffic flow is sparse, it is appropriately extended to improve the stability of the average speed.

[0024] Based on the environmental vehicle status, the spatiotemporal occupancy space, including static and dynamic occupancy, is calculated using a trajectory prediction model and preset safety thresholds. The static occupancy space is calculated based on the physical length L of the environmental vehicle, which is obtained through image algorithms after the vehicle's outline is identified by a camera, or directly obtained from the environmental vehicle's factory parameters via vehicle-to-everything (V2X) communication. The static safety redundancy distance setting must comprehensively consider the road adhesion coefficient, vehicle braking performance, and driving safety regulations to ensure that the vehicle still has sufficient avoidance space even in emergency situations. The static occupancy radius is the sum of the environmental vehicle's physical length L and the static safety redundancy distance. The static occupancy space is defined as the longitudinal range covered by the static occupancy radius, centered on the environmental vehicle's longitudinal position. Any longitudinal position of the vehicle falling within this range is considered to pose a static collision risk.

[0025] The core of dynamic space occupancy calculation lies in the accurate prediction of future vehicle trajectories and the real-time assessment of vehicle collision time (TTC). The trajectory prediction model employs a hybrid deep learning model combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). The model's input includes two parts: first, a state data matrix of multiple vehicles at the same moment, encompassing parameters such as position, velocity, and acceleration, used to extract spatial correlation features; second, a state sequence of a single vehicle over a historical time period, used to extract temporal variation features. The CNN part extracts features from the simultaneous state matrices of multiple vehicles through convolutional layers, capturing spatial dependencies such as relative positional relationships and velocity differences between vehicles. After sliding calculations of multiple convolutional kernels and dimensionality compression through pooling layers, a high-dimensional spatial feature vector is obtained. The LSTM part processes the historical state sequence of a single vehicle through a recursive structure, memorizing the temporal patterns of vehicle motion, such as the changing trends of acceleration, deceleration, and constant speed. Gating units control the forgetting and updating of information, outputting a temporal feature vector. After concatenating the spatial feature vector and the temporal feature vector, the concatenation is input into the fully connected layer for feature fusion and nonlinear mapping. The final output is the predicted trajectory of the vehicle in the environment over multiple time steps in the future. The predicted time step matches the estimated lane change time to ensure that the predicted trajectory can cover the entire lane change process.

[0026] The Time-of-Collision (TTC) calculation is based on the relative motion between the vehicle and surrounding vehicles. TTC is calculated separately for the vehicle and the vehicle in front in the current lane, the vehicle and the vehicle in front in the target lane, and the vehicle and the vehicle behind in the target lane; when the distance between the two vehicles decreases to... hour , Where dx is the difference between the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicles, and dv is the difference between the longitudinal speed of the vehicle and the longitudinal speed of the surrounding vehicles. dx and dv are subtracted in the same order. When the longitudinal distance increases or remains constant over time, TTC is determined to be infinite, meaning there is no collision risk. Dynamic occupancy space is defined in conjunction with a preset TTC safety threshold. The preset TTC safety threshold is determined based on road type, vehicle speed range, and vehicle braking response time, ensuring the vehicle has sufficient time to make braking or avoidance decisions. Dynamic occupancy space is defined as the set of all (position, speed) state points where TTC is less than or equal to the safety threshold and the longitudinal distance decreases over time. Combined with the future trajectory of the surrounding vehicles output by the trajectory prediction model, dynamic occupancy space is updated in real time with the movement of the surrounding vehicles, accurately depicting the spatiotemporal evolution of dynamic hazard areas. Spatiotemporal occupancy space is the union of static and dynamic occupancy spaces, comprehensively covering areas where collision risk may exist during lane changes.

[0027] Based on the vehicle's state, a set of parameterized candidate lateral and longitudinal trajectories is constructed with entering the target lane as the termination condition. This method uses the Frenet coordinate system for trajectory modeling. The origin of this coordinate system is set at the center position of the vehicle at the start of the lane change. The longitudinal axis is parallel to the centerline of the current lane and points in the direction of vehicle travel, while the lateral axis is perpendicular to the longitudinal axis and points to the side of the target lane. Lateral and longitudinal motions are modeled using fifth-order polynomial parametric curves. The fifth-order polynomial form ensures continuous and smooth trajectory position, velocity, and acceleration, avoiding sharp turns or accelerations during lane changes.

[0028] In the fifth-order polynomial model of the lateral trajectory, the time variable ranges from the start to the end of the lane change. The initial boundary conditions are determined by the vehicle's initial lateral state, including initial lateral coordinates (i.e., the vehicle's offset relative to the current lane centerline at the start of the lane change), initial lateral velocity (obtained through data fusion from the inertial measurement unit and wheel speed sensors), and initial lateral acceleration (directly measured by the inertial measurement unit). The termination boundary conditions are centered on entering the target lane. The termination lateral position is set to the coordinates of the target lane centerline, determined through high-precision map matching or lane line data transmitted by the roadside unit. Both the termination lateral velocity and termination lateral acceleration are set to zero to ensure stable lateral movement after the vehicle smoothly enters the target lane.

[0029] The initial boundary conditions of the fifth-order polynomial model for the longitudinal trajectory include the vehicle's initial longitudinal coordinates, initial longitudinal velocity, and initial longitudinal acceleration, all measured in real time by onboard sensors. The longitudinal position, longitudinal velocity, and longitudinal acceleration in the termination boundary conditions are variable sampling parameters, generated through systematic sampling to produce multiple different combinations. During sampling, the sampling range of the termination time is determined based on road speed limits, lane-changing safety distances, and target lane speeds. For example, the sampling range of the termination time can be 0–7.5 seconds, with a sampling interval of 0.1 seconds. Specific values ​​can be adjusted according to onboard computing power and real-time requirements to ensure both efficiency and safety during lane changes. The sampling range of the longitudinal termination position is set based on the vehicle's expected travel distance during lane changes. The sampling range of the longitudinal termination velocity covers the interval from the current lane speed to the average speed of the target lane. The sampling range of the longitudinal termination acceleration is set based on the vehicle's braking performance, ensuring it remains within physical reach. By traversing all combinations of sampling parameters, the fifth-order polynomial coefficients satisfying the initial and termination boundary conditions are solved, generating a large number of smooth candidate lateral and longitudinal trajectory sets, forming a candidate trajectory set.

[0030] By utilizing spatiotemporal occupancy, a coupled feasibility verification is performed on candidate lateral and longitudinal trajectory sets to filter out feasible trajectory clusters that simultaneously satisfy both lateral and longitudinal constraints. The core of the coupled verification lies in dynamically matching environmental and vehicle constraints at different lane-changing stages and checking whether there is a collision risk in the combination of lateral and longitudinal trajectories.

[0031] For each candidate lateral trajectory, the vehicle's position relative to the current lane line and the target lane line is determined in real time based on the described lateral position change pattern over time. Lane line position information is obtained from road images captured by cameras using algorithms such as edge detection and Hough transform, or determined from lane line coordinate data provided by high-precision maps. The rules for determining the positional relationship are as follows: when the vehicle's lateral position does not exceed the current lane line range, it is determined that it has not crossed the lane line; when the vehicle's lateral position partially exceeds the current lane line but has not completely entered the target lane line range, it is determined that it is crossing the lane line; when the vehicle's lateral position completely falls within the target lane line range, it is determined that it has completely entered the target lane.

[0032] Based on the above positional relationships, the set of environmental vehicles that constitute spatial constraints on the vehicle at each moment is dynamically determined: When not crossing the lane line, only the motion state of the vehicle in front in the current lane will affect the longitudinal braking of the vehicle, so the set of environmental vehicles only includes the vehicle in front in the current lane; when crossing the lane line, the vehicle is simultaneously in the transition area between two lanes, and the vehicle in front in the current lane and the vehicle in front of and behind the target lane may all pose a collision risk to the vehicle, so the set of environmental vehicles includes the vehicle in front in the current lane, the vehicle in front of the target lane, and the vehicle behind the target lane; after fully entering the target lane, the movement of the vehicle is constrained by the vehicles in the target lane, and the set of environmental vehicles includes the vehicle in front of and behind the target lane.

[0033] For each candidate longitudinal trajectory, it is paired with a candidate lateral trajectory, and the combined trajectory is subjected to collision risk verification at all sampling times. At each sampling time, the longitudinal position and longitudinal velocity of the vehicle corresponding to the combined trajectory are extracted to form a motion state point. It is checked whether the state point falls into the spatiotemporal occupancy space union corresponding to the current vehicle set in the environment. If the motion state point of a combined trajectory does not fall into any static or dynamic occupancy space at all sampling times, and satisfies vehicle dynamics constraints (such as acceleration and jerk not exceeding the vehicle's physical limits), the combined trajectory is determined to be safe and feasible, and is added to the feasible trajectory cluster; if the motion state point at any time falls into the occupancy space or violates the dynamics constraints, the combined trajectory is removed.

[0034] The optimal lateral trajectory is determined from a family of feasible trajectories based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity. The comprehensive evaluation function is designed to balance lane change efficiency, ride comfort, and the flexibility of subsequent longitudinal trajectory planning, and its expression covers four core evaluation items.

[0035] The first item is the lane change time, which is the difference between the termination time and the start time of the candidate lateral trajectory. This evaluation item directly reflects the lane change efficiency; the smaller the value, the more efficient the lane change, reducing the vehicle's dwell time in the transition area and lowering the risk of collision. The second item is the sum of the integrals of the squares of lateral acceleration and longitudinal acceleration, with the integration interval being the entire lane change time. Fluctuations in lateral acceleration directly affect ride comfort; the smaller the integral value, the smoother the lateral movement and the less lateral impact felt by passengers. The third item is the sum of the integrals of the squares of lateral acceleration and longitudinal acceleration. Acceleration is the rate of change of acceleration; the smaller the integral value, the smoother the change in lateral acceleration, avoiding body sway caused by sharp turns and further improving comfort. The fourth item is the indicator function, whose value equals the number of candidate longitudinal trajectories in the feasible trajectory family that can successfully pair with the current candidate lateral trajectory. This evaluation item reflects the size of the feasible space reserved by the lateral trajectory for longitudinal speed planning; the larger the value, the wider the range of longitudinal trajectory selection, which is more conducive to subsequent longitudinal braking optimization.

[0036] The calculation process of the comprehensive evaluation function is as follows: For each candidate lateral trajectory in the feasible trajectory cluster, the values ​​of the above four evaluation items are calculated respectively, and then combined according to the summation method to form a comprehensive evaluation index. By sorting the comprehensive evaluation indices of all candidate lateral trajectories, the candidate lateral trajectory with the smallest index value is selected as the optimal lateral trajectory, which can achieve the best balance between lane-changing efficiency, comfort, and longitudinal planning flexibility.

[0037] Using the optimal lateral trajectory as a constraint, this paper solves the longitudinal trajectory optimization problem, which has speed matching as the core objective and is constrained by spatiotemporal space occupancy, to obtain the optimal longitudinal trajectory. The mathematical model of the longitudinal trajectory optimization problem consists of two parts: an objective function and constraints. The objective function aims to ensure the comfort and speed matching during the longitudinal braking process, while the constraints ensure the safety of the trajectory.

[0038] The objective function consists of three parts: the first part is the integral of the square of the longitudinal acceleration, the second part is the integral of the square of the longitudinal jerk. These two parts together ensure the smoothness of the longitudinal braking process. The smaller the integral value, the smoother the braking, avoiding the forward tilting or dizziness of passengers caused by sudden braking. The third part is the speed matching term, which is the sum of the squares of the differences between the vehicle speed and the average speed of the target lane at multiple key time points during the lane change process. The key time points include the start time of crossing the lane, the completion time of crossing the lane, and the end time of the lane change. This part directly drives the optimization algorithm to make the vehicle speed gradually approach the average speed of the target lane during the lane change process, reducing the magnitude of speed adjustment after the lane change is completed and improving driving stability.

[0039] The constraints include spatial and velocity constraints: the spatial constraint requires that the vehicle's longitudinal position does not fall into static occupied space during the entire lane-changing process; the velocity constraint requires that the combination of the vehicle's longitudinal velocity and longitudinal position does not fall into dynamic occupied space. The optimization process employs gradient descent, specifically the Adam optimization algorithm, with an initial learning rate of 0.01 and 100-500 iterations. The convergence condition is that the rate of change of the objective function is less than 0.1% for 10 consecutive iterations. The polynomial coefficients of the candidate longitudinal trajectory are used as optimization variables, and the coefficient values ​​are continuously adjusted through iterative calculations to minimize the objective function value while satisfying all constraints. During the iteration process, after each coefficient adjustment, it is necessary to verify whether the trajectory meets the constraints. If the constraints are violated, a penalty term is used to correct the optimization direction, ensuring that the final solved longitudinal trajectory satisfies both comfort and speed matching requirements while avoiding all collision risk areas.

[0040] The desired steering angle and desired speed sequences are calculated based on the optimal lateral and longitudinal trajectories. The calculation of the desired steering angle sequence is based on the geometric relationship between the optimal lateral and longitudinal trajectories. First, the heading angle of the vehicle at each sampling moment is calculated using the derivatives of the optimal lateral and longitudinal trajectories. The heading angle is the arctangent of the lateral displacement with respect to the longitudinal displacement. Then, according to the vehicle dynamics model, the desired steering angle at each moment is obtained from the heading angle and its rate of change (yaw rate), forming a steering angle time series. This sequence can precisely control the vehicle's steering actions, ensuring that the vehicle smoothly completes lane changes according to the optimal lateral trajectory.

[0041] The desired velocity sequence is obtained by synthesizing the velocity components of the optimal longitudinal trajectory and the optimal lateral trajectory. The desired longitudinal velocity at each sampling time is directly given by the optimal longitudinal trajectory, and the desired lateral velocity is obtained by the derivative of the optimal lateral trajectory. The desired velocity is the square root of the sum of the squares of the desired longitudinal velocity and the desired lateral velocity, forming a velocity time series. This sequence provides an accurate speed reference for longitudinal braking, ensuring smooth speed changes during lane changes.

[0042] Based on the difference between the desired speed sequence and the current actual vehicle speed, a vehicle braking command is generated. The onboard controller compares the desired speed with the actual speed in real time, calculates the speed difference, and generates a braking command when the actual speed is higher than the desired speed; when the actual speed is equal to or lower than the desired speed, it generates a command to maintain or release the brakes, depending on the situation. The strength of the braking command is positively correlated with the speed difference; a larger speed difference results in a stronger braking command to quickly reduce the vehicle speed; a smaller speed difference results in a weaker braking command to slowly adjust the vehicle speed and avoid sudden changes in braking intensity. The braking command is transmitted to the vehicle's brake ECU via the CAN bus. The brake ECU controls the hydraulic or pneumatic pressure of the braking system according to the command signal to achieve precise braking force adjustment, ensuring that the vehicle completes deceleration and braking according to the desired speed sequence and smoothly enters the target lane.

[0043] Example 2

[0044] This invention is applicable to lane-changing scenarios with heterogeneous vehicle speeds on structured roads, such as... Figure 1 and Figure 4 As shown, it includes two key lanes and three key vehicles. The lane before a vehicle changes lanes is... The target lane is The target lane speed is significantly lower than the current lane speed, and the set of environmental vehicle states is as follows: These represent the statuses of the vehicle in front in the current lane, the vehicle in front in the target lane, and the vehicle behind in the target lane, respectively. , , For the first Vehicle information for the vehicle. For the first The displacement of the vehicle For the first The speed of the car For the first The vehicle's acceleration. The main vehicle's status is... , These are the vehicle's position, speed, acceleration, and steering angle, respectively. All of these attributes are based on the Frenet reference frame, with the vehicle's position at the start of the lane change as the origin.

[0045] like Figure 3 As shown, this device comprises three layers. The first layer consists of sensing devices and roadside units. The sensing devices include video sensors and radar sensors; the video sensors can be cameras; the radar sensors can include lidar and millimeter-wave radar; and the roadside units can be roadside sensor devices. These devices work together to acquire environmental status information. The second layer consists of information processing devices, including storage units and computing units. The storage unit stores historical environmental information, and the computing unit stores intermediate calculation results. The computing unit calculates feasible space, trajectory, and actual vehicle control quantities. The third layer consists of vehicle control devices, specifically the vehicle's bogie, engine ECU, and brake ECU. These three components work together to control the vehicle's actual movement, completing vehicle steering control and longitudinal deceleration and braking.

[0046] The core of this control system lies in the information flow process from environmental perception information to the final actual vehicle control quantity. It includes two core links: the first link is the conversion from perception information to the vehicle's feasible space, and the second link is the conversion from the feasible space to the vehicle trajectory and control quantity. The specific execution methods of the system and equipment are described below.

[0047] In the process of converting perceived information into feasible space for the vehicle, this system primarily relies on sensing and information processing equipment. The perceived information has a historical time frame. Complete environmental state within To save storage space, it will be deleted every time it is updated. All information is acquired beforehand. After acquiring the information, the vehicle determines the feasible space. First, the computing unit processes the perceived data to determine if there are any obstacles in the environment that affect the vehicle's movement. If no obstacles exist, the vehicle can freely plan its movement, so the complete space is directly transmitted to the storage unit for recording. If obstacles exist, space occupancy is determined. Second, to better handle the spatiotemporal changes in the feasible space, the computing unit uses a deep learning-based model to predict the vehicle's trajectory. The model structure generally combines CNN and LSTM. The CNN can contain 3 convolutional layers (3×3 kernel size, 16 / 32 / 64 channels), and the LSTM can contain 2 hidden layers (64 neurons per layer). The prediction time step is 5 seconds, and the temporal resolution is 0.1 seconds. The advantage of this approach is that CNN can extract the spatial relationships between vehicles, and LSTM can extract the temporal relationships within the vehicles themselves. Furthermore, the model has a small number of parameters, which can meet the needs of real-time computation. Finally, based on the prediction results, the computing unit calculates the width of the vehicle's spatiotemporal occupancy to ultimately define the feasible space, where static and dynamic occupancy are considered simultaneously.

[0048] The calculation of static occupancy by the computing unit only considers the need for lane-changing vehicles to maintain a safe distance from the target vehicle. Therefore, it can be represented by the vehicle's own structure and static redundancy distance: ; In the formula, For static occupancy radius, To occupy vehicle length, For redundant spatial distance, static safety redundancy distance The value range can be 2 to 5 meters. The static space occupied is: ; Dynamic occupancy takes into account the dynamic safety relationship between vehicles, using Time-of-Collision (TTC) as a metric. The expression for TTC is: ; In the formula, This represents the difference between the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicles. The difference between the vehicle's longitudinal velocity and the longitudinal velocity of surrounding vehicles is represented by dx and dv, which are subtracted in the same order. TTC is calculated separately for the vehicle and the vehicle in front in the current lane, the vehicle and the vehicle in front in the target lane, and the vehicle and the vehicle behind in the target lane. To ensure safety, the vehicle must maintain a TTC that meets a safety threshold during control. , As the key threshold, The value range can be 2 to 5 seconds, so dynamic occupancy can be represented as: ; Once the above occupancy information is calculated, it will be stored in the calculation unit for use in subsequent control quantity calculations.

[0049] After spatial partitioning, the second stage begins. The optimal control parameters for the vehicle are calculated based on the current state and environment, and then control is executed. In actual calculations, the vehicle's trajectory can be modeled as a continuous curve. Generally, this curve is simplified to a quintic function, which simultaneously ensures both the complexity and smoothness of the trajectory curve. Therefore, in the Frenet coordinate system, this curve takes the following form: ; In the formula, , i represents the parameters of the curve in the horizontal and vertical directions, respectively, where i is 1, 2, 3, 4, 5; This is a time variable. For the lane-changing vehicle, the initial state includes... ,in, As the initial ordinate, The initial longitudinal velocity, This is the initial longitudinal acceleration; As the initial x-coordinate, The initial lateral velocity, The initial lateral acceleration can be obtained by estimating its own state; the final state of the main vehicle is... ,in, To terminate the ordinate, To terminate the speed, To terminate acceleration, this does not include The directional state is due to the fact that, under the termination condition, the lateral velocity and acceleration are 0, and the termination x-coordinate is zero. Align with the center line of the target lane. Lane change end time is... Based on the initial and final states, the following initial and final state equations can be obtained: ; According to this equation, it can be found that through sampling... A complete horizontal trajectory can be obtained by sampling. We can obtain the longitudinal trajectory, thus obtaining a complete set of trajectories. However, there are four key sampling variables. To ensure the speed of the final trajectory, the sampling parameters cannot all be sparse. Considering that there are fewer sampling variables for the lateral trajectory, we consider performing fine sampling in the lateral direction first and sparse sampling in the longitudinal direction to select the overall better trajectory. After filtering, we solve for the optimal longitudinal trajectory based on the optimal lateral trajectory.

[0050] In the spatial partitioning of the previous stage, lateral and longitudinal movements can be coupled by the time a vehicle spends in each lane. Specifically, when a vehicle has not yet partially entered the target lane, it is only affected by vehicles in its current lane. Once the vehicle has fully entered the target lane, it will be affected by vehicles in that lane. When crossing lanes, it will be affected by all vehicles. Therefore, based on the spatiotemporal occupancy, we can determine whether each longitudinal trajectory corresponding to each lateral trajectory is feasible. If feasible, it is added to the trajectory cluster; otherwise, it is ignored. Ultimately, a complete feasible trajectory cluster can be obtained. .

[0051] Solving for the optimal lateral trajectory considers several characteristics: firstly, high efficiency in lane changing (completing the lane change in the shortest time); secondly, maximizing overall trajectory comfort; and thirdly, providing greater feasibility for longitudinal movement. In summary, the entire problem can be modeled as follows: ; Where ω1, ω2, and ω3 are positive weighting coefficients, and ω4 is a negative weighting coefficient; the first term This represents lane-changing efficiency and encourages shorter lane-changing times; the second item... The third term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration (where the longitudinal integral is the optimal value among the feasible longitudinal trajectories corresponding to the lateral trajectory). The fourth term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration, which are used together to evaluate the smoothness and comfort of lateral motion. It is an indicator function whose function value is equal to the number of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the currently evaluated candidate lateral trajectory. The fourth term is used to evaluate the size of the feasible space reserved by the lateral trajectory for subsequent longitudinal velocity planning. By calculating and comparing the J values ​​corresponding to all candidate lateral trajectories in the feasible trajectory cluster, the candidate lateral trajectory that minimizes the J value is selected as the optimal lateral trajectory. Specifically: t is the end time of the lane change; t0 is the start time of the lane change; a y This refers to lateral acceleration; a x For longitudinal acceleration; jerk y jerk is the lateral jerk, specifically the first derivative of lateral acceleration with respect to time.x C(t) represents the longitudinal jerk, i.e., the first derivative of the longitudinal acceleration with respect to time; tw is the termination time of the longitudinal trajectory optimization, with a fixed longitudinal optimization time window length in seconds; I is the indicator function, whose value is equal to the number of elements in the set; C(t) end ) is the set of feasible longitudinal trajectory pairs, representing the set of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the current candidate lateral trajectory.

[0052] Based on the optimal lateral trajectory selected above, the longitudinal trajectory can be finely planned. Here, an optimization-based method is used to directly solve the longitudinal trajectory. The problem is modeled as follows: ; ; Among them, the first two terms of the objective function and These represent the square integrals of longitudinal acceleration and jerk, respectively, used to ensure comfort during longitudinal braking or acceleration; the third term of the objective function For speed matching items, where for The speed of the main vehicle is constantly monitored, with entering the target lane as the termination condition. The termination time tw for longitudinal trajectory optimization is set. Given the average speed in the target lane, the third term of the direct drive control algorithm calculates a longitudinal trajectory that makes the vehicle's speed approach the target lane speed at the end of the lane change; R sta The static occupied space; x is the longitudinal position of the vehicle; v x R is the longitudinal speed of the vehicle. dyn This involves dynamic space allocation. Since the current vehicle speed is higher than the average speed of the target lane, it can be assumed that the closer the vehicle's speed is to the average speed of the target lane after entering it, the smaller the subsequent speed adjustment will be, resulting in smoother and safer driving. This encourages the vehicle to decelerate and brake. Solving the above problem yields the complete planned trajectory, which can be expressed as: ; In the formula, For the optimal trajectory, These represent the temporal variations of the optimal trajectory in the longitudinal, lateral, and heading directions, respectively. The rate of change of the heading angle over time (yaw rate) can be expressed by the following formula: ; According to vehicle dynamics models, given a fixed vehicle structural parameter, there is a definite mapping relationship between the rate of change of the steering angle and the vehicle's heading angle (yaw rate). For example, using a bicycle kinematics model, the desired steering angle is: .

[0053] The vehicle's desired speed is: ; So and These are the vehicle's attitude control parameters. The values ​​of two key control quantities can be mapped to the final control units of the vehicle's bogie, throttle ECU, and brake ECU, ultimately achieving steering control and longitudinal deceleration and braking when the vehicle changes lanes.

[0054] Example 3

[0055] like Figure 2 and Figure 3 As shown in this embodiment, a cooperative braking device for lane changing in an autonomous vehicle is integrated into the vehicle's on-board control unit or braking system domain controller. Its hardware architecture works in coordination with the vehicle's sensing devices, information processing devices, vehicle control devices, and roadside units to achieve cooperative braking control during lane changing.

[0056] The information acquisition module in the device is connected to the onboard environmental perception sensors and the vehicle-to-everything (V2X) communication unit. The onboard environmental perception sensors, corresponding to the sensing devices shown in the attached diagram, include video sensors and radar sensors. The video sensors utilize high-definition industrial cameras deployed inside the vehicle's windshield to capture road images and identify lane line contours and the appearance features of surrounding vehicles. The radar sensors include millimeter-wave radar and lidar. The millimeter-wave radar is deployed at the front and rear bumpers of the vehicle, while the lidar is deployed on the roof. Both work together to collect the vehicle's real-time motion status, as well as data such as the relative distance, longitudinal speed, and longitudinal acceleration of vehicles in the current and target lanes. The V2X communication unit establishes a wireless connection with the roadside unit shown in the attached diagram. The roadside unit acquires speed data from multiple vehicles in the target lane through sensing devices deployed along the road. After statistical calculation to obtain the average speed of the target lane, it transmits the data to the information acquisition module. The information acquisition module summarizes and organizes the multi-source data to obtain information on the vehicle's status, the status of surrounding vehicles, and the average speed of the target lane.

[0057] The spatiotemporal occupancy calculation module is deployed within the computing unit of the information processing equipment shown in the attached diagram. This computing unit is an onboard high-performance computing platform with low-latency data processing capabilities. The spatiotemporal occupancy calculation module includes a static occupancy calculation unit and a dynamic occupancy calculation unit. The static occupancy calculation unit calculates the longitudinal range of the static occupancy space by combining the physical length L of the environmental vehicle obtained from the information acquisition module with a preset static safety redundancy distance. The dynamic occupancy calculation unit integrates a trajectory prediction model. This model uses the computing unit to call the historical state data of the environmental vehicle stored in the storage unit, extracts spatial correlation features and temporal change features, and outputs the trajectory information of the environmental vehicle over a future period. Subsequently, the dynamic occupancy calculation unit, combined with a preset TTC safety threshold, calculates the dynamically changing dynamic occupancy space based on the future trajectory of the environmental vehicle and the relative motion relationship between the vehicle and the environmental vehicles. The union of the static and dynamic occupancy spaces constitutes the spatiotemporal occupancy space. The relevant calculation results are stored in the storage unit of the information processing equipment for subsequent modules to access.

[0058] The trajectory planning module is also deployed within the computing unit of the information processing equipment, comprising a trajectory generation unit, a coupling verification unit, a lateral decision-making unit, and a longitudinal optimization unit. The trajectory generation unit uses vehicle state data obtained from the information acquisition module, with entering the target lane as the termination condition, to construct a parameterized set of candidate lateral and longitudinal trajectories. The relevant trajectory data is temporarily stored in the storage unit. The coupling verification unit retrieves spatiotemporal occupancy data and candidate trajectory data from the storage unit, performs pairwise verification of the candidate lateral and longitudinal trajectory sets, dynamically determines the environmental vehicle constraint set based on the vehicle's lateral position at different times, checks whether the motion state points corresponding to the combined trajectories fall within the spatiotemporal occupancy space, and filters out feasible trajectory clusters that meet the constraints. The lateral decision-making unit calculates and evaluates the candidate lateral trajectories in the feasible trajectory cluster based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity, selecting the trajectory with the best comprehensive evaluation result as the optimal lateral trajectory. The longitudinal optimization unit uses the optimal lateral trajectory as a constraint, calls spatiotemporal occupancy data and the average vehicle speed information of the target lane, solves the longitudinal trajectory optimization problem with speed matching as the core objective, and obtains the optimal longitudinal trajectory for deceleration and braking. The information of the optimal lateral trajectory and the optimal longitudinal trajectory is stored in the storage unit.

[0059] The control variable calculation module is deployed within the computing unit of the information processing equipment. It retrieves the time-series data of the optimal lateral and longitudinal trajectories from the storage unit. Based on the lateral displacement changes of the optimal lateral trajectory, the control variable calculation module calculates the heading angle of the vehicle at each moment during the lane change process. Then, it performs a time-dimension derivative operation on the heading angle to obtain the desired steering angle sequence. Simultaneously, the control variable calculation module combines the longitudinal velocity data of the optimal longitudinal trajectory with the lateral velocity data of the optimal lateral trajectory to synthesize the desired velocity sequence. The information of the desired steering angle sequence and the desired velocity sequence is transmitted to the vehicle control equipment.

[0060] The output of the braking command generation module is connected to the brake ECU of the vehicle control equipment shown in the attached diagram. The brake ECU is the vehicle braking control unit. The braking command generation module acquires the vehicle's current actual speed data in real time and calculates the difference between it and the desired speed sequence output by the control quantity calculation module. Based on the magnitude and rate of change of the difference, the braking command generation module generates the corresponding vehicle braking command and transmits it directly to the brake ECU. The brake ECU adjusts the hydraulic pressure of the braking system according to the command, controlling the movement of the braking actuator. Simultaneously, the desired steering angle sequence output by the control quantity calculation module is transmitted to the bogie of the vehicle control equipment. The bogie is the actuator of the electric power steering system, adjusting the vehicle's steering angle according to the desired steering angle sequence. The engine ECU and brake ECU of the vehicle control equipment work together to adjust the power output according to the desired speed sequence, assisting in achieving vehicle speed control.

[0061] The workflow of this device is as follows: the sensing device collects multi-source data and transmits it to the information processing device. The information processing device completes data calculation and trajectory planning through various modules, generates control commands, and transmits them to the vehicle control device. Finally, through the coordinated action of the bogie engine ECU and brake ECU, the vehicle decelerates, brakes, and smoothly changes lanes, ensuring the safety, efficiency, and ride comfort of the lane-changing process.

[0062] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A cooperative braking method for lane changing in an autonomous vehicle, characterized in that, Includes the following steps: Obtain information including the vehicle's status, the status of vehicles in the environment, and the average speed of vehicles in the target lane; Based on the vehicle's environmental status, the spatiotemporal occupancy space, including static and dynamic occupancy, is calculated using a trajectory prediction model and preset safety thresholds. Based on the vehicle's status, a parameterized set of candidate lateral and longitudinal trajectories is constructed with entering the target lane as the termination condition. Using the aforementioned spatiotemporal occupancy space, the feasibility of coupling the candidate lateral and longitudinal trajectory sets is verified, and feasible trajectory clusters that simultaneously satisfy lateral and longitudinal constraints are selected. The optimal lateral trajectory is determined from the feasible trajectory family based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity. Using the optimal lateral trajectory as a constraint, the longitudinal trajectory optimization problem with velocity matching as the core objective and constrained by the spatiotemporal space is solved to obtain the optimal longitudinal trajectory; Calculate the desired steering angle sequence and desired velocity sequence based on the optimal lateral trajectory and the optimal longitudinal trajectory; Based on the difference between the desired speed sequence and the current actual vehicle speed, a vehicle braking command is generated to brake the vehicle.

2. The cooperative braking method for lane changing of an automated vehicle according to claim 1, characterized in that, The steps for calculating the spatiotemporal space occupancy, including both static and dynamic occupancy, specifically include: Based on the physical length L of the vehicle in the environment and the preset static safety redundancy distance Calculate the static occupancy radius And based on this, define the static occupancy space. , where x i Let be the longitudinal position of the i-th environmental vehicle; calculate the vehicle collision time TTC based on the relative motion relationship between the vehicle and the environmental vehicles, when the distance between the two vehicles decreases to... hour ,in This represents the difference between the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicles. The difference between the longitudinal velocity of the vehicle and the longitudinal velocity of the surrounding vehicles is calculated using the same subtraction order for dx and dv, combined with a preset TTC safety threshold. Define dynamic space usage ； The spatiotemporal occupied space is the static occupied space. With the aforementioned dynamic occupied space The union of the represented danger zones.

3. The cooperative braking method for lane changing of an automated vehicle according to claim 2, characterized in that, The trajectory prediction model is a hybrid deep learning model combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network. The CNN extracts spatial correlation features between the states of multiple vehicles in the environment at the same time, while the LSTM extracts temporal features of state changes for a single vehicle over a historical period. The hybrid deep learning model integrates these spatial correlation features and temporal features to process the historical and current state sequences of the vehicles, predicting their trajectories over a future period. The predicted future trajectories are used to calculate the dynamically occupied space. The spatiotemporal evolution.

4. The cooperative braking method for lane changing of an automated vehicle according to claim 1, characterized in that, The steps for constructing a parameterized set of candidate lateral and longitudinal trajectories specifically include: In the Frenet coordinate system, the origin is the vehicle's current position, the longitudinal axis is parallel to the centerline of the current lane, and the lateral axis is perpendicular to the longitudinal axis and points to the side of the target lane; a fifth-order polynomial parametric curve is used to represent the lateral movement. With longitudinal movement Modeling is performed, where: ； For time variables, These are the parameters of the curve in the horizontal and vertical directions, respectively, where i is 1, 2, 3, 4, and 5; the initial horizontal coordinates of the vehicle state are... Initial lateral velocity Initial lateral acceleration Initial ordinate Initial longitudinal velocity Initial longitudinal acceleration As the fifth-degree polynomial at the initial time Boundary conditions; termination time The lateral position is set as the centerline coordinate of the target lane. The lateral velocity and lateral acceleration are set to zero as the lateral termination boundary condition, and the termination time is set to... vertical position Longitudinal velocity Longitudinal acceleration As variable sampling parameters to form longitudinal termination boundary conditions; by systematically sampling different , , and Combine and solve the polynomial coefficients that satisfy all boundary conditions to generate a batch of smooth candidate lateral trajectory sets and candidate longitudinal trajectory sets.

5. The cooperative braking method for lane changing of an automated vehicle according to claim 4, characterized in that, The feasibility of coupling the candidate lateral and longitudinal trajectory sets is verified, specifically including: For each candidate lateral trajectory, based on the lateral position of the vehicle it describes... As time changes, the vehicle's position relative to the current lane line and the target lane line is determined in real time, and the set of environmental vehicles that constitute spatial constraints on the vehicle at each moment is dynamically determined based on the positional relationship. The rules for determining the set are as follows: When a vehicle has not crossed the lane lines, only the vehicle in front in the current lane is considered; Once the vehicle has fully entered the target lane, only the vehicles in front of and behind in the target lane are considered. When a vehicle is crossing a lane line, both the vehicle in front in the current lane and the vehicles in front and behind in the target lane should be considered simultaneously. For each candidate longitudinal trajectory, examine the vehicle motion state point (x(t), v) jointly determined by it and a candidate lateral trajectory at the same time t. x (t)) whether it falls into the union of the spatiotemporal occupied space corresponding to the set of environmental vehicles at that time; if for a combination of a candidate lateral trajectory and a candidate longitudinal trajectory, the motion state point corresponding to all sampling times t does not fall into any occupied space, then the combination is determined to be safe and feasible, and it is added to the feasible trajectory cluster.

6. The cooperative braking method for lane changing of an automated vehicle according to claim 5, characterized in that, The comprehensive evaluation function is specifically expressed as follows: ； Where ω1, ω2, and ω3 are positive weighting coefficients, and ω4 is a negative weighting coefficient; the first term Represents lane change efficiency, and is the time when the lane change ends; the second term The third term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration. The fourth term is the sum of the integrals of the squares of the lateral acceleration and the longitudinal acceleration, which are used together to evaluate the smoothness and comfort of lateral motion. It is an indicator function whose function value is equal to the number of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the currently evaluated candidate lateral trajectory. The fourth term is used to evaluate the size of the feasible space reserved by the lateral trajectory for subsequent longitudinal velocity planning. By calculating and comparing the J values ​​corresponding to all candidate lateral trajectories in the feasible trajectory cluster, the candidate lateral trajectory that minimizes the J value is selected as the optimal lateral trajectory. Specifically: t0 is the start time of the lane change; a y This refers to lateral acceleration; a x For longitudinal acceleration; jerk y jerk is the lateral jerk, specifically the first derivative of lateral acceleration with respect to time. x C(t) represents the longitudinal jerk, i.e., the first derivative of the longitudinal acceleration with respect to time; tw is the termination time of the longitudinal trajectory optimization, and the length of the longitudinal optimization time window corresponding to tw is fixed in seconds; I is the indicator function, whose function value is equal to the number of elements in the set; C(t) end ) is the feasible longitudinal trajectory pairing set, representing the set of all distinct candidate longitudinal trajectories in the feasible trajectory cluster that can be successfully paired with the current candidate lateral trajectory.

7. The cooperative braking method for lane changing of an automated vehicle according to claim 6, characterized in that, The mathematical model for the longitudinal trajectory optimization problem with velocity matching as the core objective is as follows: Objective function: ; Constraints: ; Among them, the first two terms of the objective function and These represent the square integrals of longitudinal acceleration and jerk, respectively, used to ensure comfort during longitudinal braking or acceleration; the third term of the objective function For speed matching items, where for The speed of the main vehicle is constantly monitored, with entering the target lane as the termination condition. The termination time tw for longitudinal trajectory optimization is set. Given the average speed in the target lane, the third term of the direct drive control algorithm calculates a longitudinal trajectory that makes the vehicle's speed approach the target lane speed at the end of the lane change; R sta The static occupied space; x is the longitudinal position of the vehicle; v x R is the longitudinal speed of the vehicle. dyn It dynamically occupies space.

8. A cooperative braking device for lane changing in an autonomous vehicle, based on the cooperative braking method for lane changing in an autonomous vehicle according to any one of claims 1-7, characterized in that, include: The information acquisition module is configured to acquire information including the vehicle's status, the status of vehicles in the environment, and the average speed of vehicles in the target lane. The spatiotemporal occupancy calculation module is configured to calculate the spatiotemporal occupancy space, including static occupancy space and dynamic occupancy space, based on the vehicle state in the environment, through a trajectory prediction model and a preset safety threshold. The trajectory planning module is configured to construct a parameterized set of candidate lateral and longitudinal trajectories based on the vehicle's state and with entering the target lane as the termination condition. It uses the spatiotemporal occupancy space output by the spatiotemporal occupancy calculation module to perform coupled feasibility verification on the set of candidate lateral and longitudinal trajectories, and selects feasible trajectory clusters. Based on a comprehensive evaluation function that includes lane change time, comfort, and longitudinal space capacity, it determines the optimal lateral trajectory from the feasible trajectory clusters. Using the optimal lateral trajectory as a constraint, it solves the longitudinal trajectory optimization problem with speed matching as the core objective and constrained by the spatiotemporal occupancy space, thereby obtaining the optimal longitudinal trajectory for deceleration and braking. The control quantity calculation module is configured to calculate the desired steering angle sequence and desired speed sequence of the vehicle during the lane changing process based on the optimal lateral trajectory and the optimal longitudinal trajectory output by the trajectory planning module. The braking command generation module is configured to generate a vehicle braking command for the vehicle braking control unit based on the difference between the desired speed sequence output by the control quantity calculation module and the current actual vehicle speed.

9. A cooperative braking device for lane changing in an automated vehicle according to claim 8, characterized in that, The spatiotemporal occupancy calculation module includes a static occupancy calculation unit and a dynamic occupancy calculation unit; The static occupancy calculation unit is configured to calculate the static occupancy space based on the physical length L of the environmental vehicle and the static safety redundancy distance. The dynamic occupancy calculation unit integrates the trajectory prediction model and is configured to use the environmental vehicle future trajectory information output by the trajectory prediction model, combined with a preset TTC safety threshold, to calculate the dynamically changing dynamic occupancy space. The trajectory planning module includes a trajectory generation unit, a coupling verification unit, a lateral decision-making unit, and a longitudinal optimization unit; the trajectory generation unit is configured to perform the operation of constructing a parameterized set of candidate lateral and longitudinal trajectories; the coupling verification unit is configured to perform the operation of verifying the coupling feasibility. The lateral decision-making unit is configured to perform the operation of determining the optimal lateral trajectory based on a comprehensive evaluation function; the longitudinal optimization unit is configured to perform the operation of solving the longitudinal trajectory optimization problem with velocity matching as the core objective.

10. A cooperative braking device for lane changing in an automated vehicle according to claim 9, characterized in that, The aforementioned lane-changing cooperative braking device for autonomous vehicles is integrated into the vehicle's onboard control unit or braking system domain controller; the information acquisition module is connected to the onboard environmental perception sensor and the vehicle network communication unit; the output of the braking command generation module is connected to the vehicle braking control unit for directly transmitting the control commands required for deceleration and braking.

Images