Trajectory planning and decision making method, device, vehicle, medium and program product

Abstract

This application relates to the field of automotive driving technology, and discloses a trajectory planning and decision-making method, device, vehicle, medium, and program product to ensure more reasonable avoidance decisions while reducing computational power consumption. Specifically, the method includes: detecting an avoidance target during vehicle operation; determining a target reference line for the vehicle to bypass the target, wherein the target reference line indicates the bypass direction and endpoint of the vehicle to bypass the target, and the target reference line is one of at least two reference lines, and at least two of the at least two reference lines have different bypass directions; determining a target avoidance trajectory that meets set constraints based on the target reference line and obstacle and road information in the surrounding environment of the vehicle; and controlling the vehicle to steer around the target based on the target avoidance trajectory.

Description

Technical Field

[0001] This application relates to the field of vehicle driving technology, and in particular to a trajectory planning and decision-making method, device, vehicle, medium, and program product. Background Technology

[0002] With the development of Advanced Driver Assistance Systems (ADAS), Automatic Emergency Braking (AEB) has been widely applied in mass-produced vehicles. However, in certain high-risk scenarios (such as stationary vehicles ahead, pedestrians suddenly crossing the road, or oncoming vehicles intruding), longitudinal braking alone cannot prevent collisions. Therefore, Automatic Emergency Steering (AES) has been proposed as a supplementary function to AEB, aiming to achieve obstacle avoidance through lateral intervention.

[0003] The existing AES solutions are mainly divided into two categories: (1) Emergency Steering Assist (ESA): provides auxiliary torque when the driver is understeering, without actively taking over the steering; (2) Fully Automatic Emergency Steering: autonomously performs steering to avoid obstacles after confirming that there is room to avoid obstacles in the environment.

[0004] However, in actual implementation, fully automated AES currently faces problems such as ambiguous avoidance direction decision, insufficient trajectory safety (e.g., the generated avoidance trajectory may cross the line, approach the curb, or conflict with obstacles in adjacent lanes), and poor curve adaptability, which result in poor emergency avoidance performance of vehicles. Summary of the Invention

[0005] This application provides a trajectory planning and decision-making method, device, vehicle, medium, and program product to ensure more reasonable avoidance decisions while reducing computing power consumption.

[0006] In a first aspect, embodiments of this application provide a trajectory planning and decision-making method, applied to a trajectory planning and decision-making device, comprising: detecting a target to be avoided by the vehicle during vehicle operation; determining a target reference line for the vehicle to bypass the target, the target reference line being used to indicate the bypass direction and endpoint of the vehicle to bypass the target, the target reference line being one of at least two reference lines, and at least two of the at least two reference lines having different bypass directions; determining a target avoidance trajectory that satisfies set constraints based on the target reference line and obstacle and road information of the surrounding environment of the vehicle; and controlling the vehicle to steer based on the target avoidance trajectory to bypass the target.

[0007] Thus, this application can use the target reference line to perform prior calculations on information such as the avoidable space (e.g., the space between the starting point and the ending point) and the detour direction, and then determine the avoidance trajectory based on the target reference line, ensuring a more reasonable avoidance decision. Meanwhile, since fine-grained trajectory solving requires high computational power, this scheme effectively avoids the unnecessary computational consumption caused by performing full a posteriori calculations for multiple avoidance directions by using coarse-grained pre-solution of the reference line, based on the pre-determined optimal avoidance direction. Furthermore, combining structured road information such as obstacles and curbs for trajectory planning can cover avoidance decisions in emergency avoidance scenarios in curves, improving the emergency avoidance effect in curve scenarios. Therefore, it improves the emergency avoidance effect in various scenarios.

[0008] In one possible implementation of the first aspect above, determining the target reference line for the vehicle to detour around the target includes: acquiring at least two reference lines for the vehicle, the at least two reference lines including at least one or more of the following: a reference line for detouring to the left within the current lane, a reference line for detouring to the left across a lane, a reference line for detouring to the right within the current lane, and a reference line for detouring to the right across a lane; wherein the endpoint of each of the at least two reference lines is spaced a first distance from the starting point of the vehicle in a first direction, and the endpoint of each reference line is spaced a second distance from the lane line associated with the detouring direction in a second direction; and determining the target reference line from the at least two reference lines.

[0009] In this way, by pre-setting multiple reference lines for left / right, within the lane / across lanes, and uniformly constraining their endpoint positions with offsets of a first longitudinal distance and a second lateral distance, high-quality candidate paths covering typical obstacle avoidance scenarios are generated in a structured manner. This balances the adequacy of detours with driving safety, improves planning efficiency and scenario adaptability, and provides clear and reasonable boundary conditions for subsequent trajectory optimization, effectively supporting reliable obstacle avoidance by the autonomous driving system in complex environments.

[0010] In one possible implementation of the first aspect above, determining the target reference line from at least two reference lines includes: generating M verification scores for each of the at least two reference lines; determining the target reference line from the at least two reference lines based on the M verification scores for each reference line; wherein the M verification scores include at least one of the following: obstacle verification score, urgency verification score, smoothness verification score, lane keeping verification score, historical decision consistency verification score, roadside safety verification score, solid line crossing verification score, and vulnerable road user traversing direction verification score.

[0011] Thus, this application comprehensively evaluates the feasibility and merits of each reference line based on multi-dimensional verification scores, thereby accurately determining the target reference line. This solution effectively integrates multi-source information such as environmental perception, high-precision maps, traffic rules, and behavior prediction, ensuring not only the geometric feasibility of the selected reference line but also achieving an optimized balance in terms of safety, compliance, comfort, and behavioral rationality. Especially in complex urban scenarios, it can prioritize avoiding high-risk behaviors such as crossing solid lines, approaching curbs, and encroaching on pedestrian paths, while also considering vehicle dynamics constraints and driving experience, significantly improving the robustness and safety of the autonomous driving system's obstacle avoidance decisions.

[0012] In one possible implementation of the first aspect above, determining the target reference line from at least two reference lines based on the M verification scores of each reference line includes: determining a valid reference line from at least two reference lines, wherein the first verification score of the valid reference line is greater than or equal to a corresponding preset threshold, and the first verification score includes at least one of the following: obstacle verification score, urgency level verification score, or curbside safety verification score; determining the corresponding total verification score based on the M verification scores of each valid reference line; and determining the reference line with the highest total verification score as the target reference line.

[0013] In this way, by first screening effective reference lines that meet key safety thresholds (such as obstacle, urgency level, or curb safety), and then selecting the target reference line by comprehensively considering multi-dimensional verification scores, the basic safety of the avoidance path is ensured, while also taking into account comfort and rationality, effectively improving the reliability of autonomous driving system decision-making and environmental adaptability.

[0014] In one possible implementation of the first aspect above, the obstacle verification score of the reference line is determined by: determining the intersection time of each obstacle in the surrounding environment of the vehicle with the reference line according to the type of obstacle; and determining the obstacle verification score of the reference line according to the minimum intersection time and the maximum default intersection time.

[0015] In one possible implementation of the first aspect described above, determining the intersection time of each obstacle in the vehicle's surrounding environment with the reference line based on the type of obstacle includes: when the obstacle is a dynamic obstacle, determining whether the obstacle intersects with the reference line; if the obstacle intersects with the reference line, determining a first time based on the predicted trajectory of the obstacle and a second time based on the reference line; and when the difference between the first time and the second time is less than a set time difference, determining the minimum of the first time and the second time as the corresponding intersection time; when the difference between the first time and the second time is greater than or equal to the set time difference, determining the default maximum time as the corresponding intersection time; if the obstacle does not intersect with the reference line, determining the default maximum time as the corresponding intersection time.

[0016] In one possible implementation of the first aspect above, the intersection time of each obstacle in the surrounding environment of the vehicle with the reference line is determined according to the type of obstacle, including: when the obstacle is a static obstacle, determining whether the obstacle intersects with the reference line; if the obstacle intersects with the reference line, the default minimum time is determined as the corresponding intersection time; if the obstacle does not intersect with the reference line, the default maximum time is determined as the corresponding intersection time.

[0017] In one possible implementation of the first aspect above, the urgency score of the reference line is determined by: determining the maximum curvature among the curvatures of each sampling point in the reference line; determining the maximum lateral acceleration of the reference line based on the maximum curvature and the set vehicle speed; determining the acceleration difference between the square of the set lateral acceleration and the square of the maximum lateral acceleration; determining the ratio of the acceleration difference to the square of the set lateral acceleration; and determining the urgency score of the reference line.

[0018] In one possible implementation of the first aspect above, the smoothness check score of the reference line is determined by: determining the root mean square of the lateral acceleration rate of change of the reference line; determining the root mean square of the default lateral acceleration rate of change of the reference line; determining the difference between the root mean square of the default lateral acceleration rate of change and the root mean square of the lateral acceleration rate of change, and determining the smoothness check score of the reference line based on the ratio of the difference to the root mean square of the default lateral acceleration rate of change.

[0019] In one possible implementation of the first aspect above, the lane keeping check score of the reference line is determined as follows: if the reference line is a reference line that detours to the left within the current lane or a reference line that detours to the right within the current lane, then the upper limit of the lane score is determined as the lane keeping check score of the reference line; or, if the reference line is a reference line that detours to the left across a lane or a reference line that detours to the right across a lane, then the lower limit of the lane score is determined as the lane keeping check score of the reference line.

[0020] In one possible implementation of the first aspect above, the historical decision information verification score of the reference line is determined as follows: if the endpoint of the detour indicated by the reference line is the same as the endpoint of the detour corresponding to the historical detour decision information, then the upper limit of the historical decision score is determined as the historical decision information verification score of the reference line; if the endpoint of the detour indicated by the reference line is different from the endpoint of the detour corresponding to the historical detour decision information, then the lower limit of the historical decision score is determined as the historical decision information verification score of the reference line. The historical detour decision information includes the historical reference line generated for the avoidance target in the previous control cycle, and the endpoint of the detour indicated by the historical detour decision information is the endpoint of the detour indicated by the historical reference line.

[0021] In one possible implementation of the first aspect described above, the safety verification score of the reference line's curb side is determined by: determining the minimum distance between the reference line and discrete points of the curb in the vehicle's surrounding environment; if the minimum distance is greater than or equal to a set safety distance, then the upper limit of the curb score is determined as the safety verification score of the reference line's curb side; if the minimum distance is less than the set safety distance or greater than a set danger distance, then the square of the difference between the minimum distance and the set danger distance is determined as a first result, the square of the difference between the set safety distance and the minimum distance is determined as a second result, and the ratio of the first result and the second result is determined as the safety verification score of the reference line's curb side; if the minimum distance is less than or equal to the set danger distance, then the lower limit of the curb score is determined as the safety verification score of the reference line's curb side.

[0022] In one possible implementation of the first aspect above, the cross-solid line verification score of the reference line is determined as follows: if the reference line is a reference line that crosses the solid lane line, the lower limit of the cross-solid line score is determined as the roadside safety verification score of the reference line; if the reference line is not a reference line that crosses the solid lane line, the upper limit of the cross-solid line score is determined as the roadside safety verification score of the reference line.

[0023] In one possible implementation of the first aspect above, the cross-traffic direction verification score of the reference line is determined by: determining the lateral movement direction of the vulnerable road user in the surrounding environment of the vehicle; determining the set of positions of the vulnerable road user according to the lateral movement direction and the collision time window at a set time interval; if the reference line includes the positions in the set, then the lower limit of the cross-traffic score is determined as the cross-traffic direction verification score of the reference line; if the reference line does not include the positions in the set, then the upper limit of the cross-traffic score is determined as the cross-traffic direction verification score of the reference line.

[0024] In one possible implementation of the first aspect above, the reference line is fitted with a set of polynomials, wherein the polynomials include at least one of the following: a polynomial constraining the position of the vehicle's starting point, a polynomial constraining the position of the vehicle's avoidance point, a polynomial constraining the vehicle's heading at the starting point, a polynomial constraining the vehicle's heading at the ending point, or a polynomial constraining the curvature of the vehicle at the starting point; wherein the position of the vehicle's avoidance point is associated with the position of the avoidance target and the detour direction, and the heading of the vehicle's ending point is associated with the lane line associated with the detour direction.

[0025] Thus, this embodiment utilizes polynomials, such as fifth-order polynomials, to accurately generate smooth reference lines that conform to road geometry and vehicle kinematics. This method ensures that the reference lines accurately fit the desired detour path in space, while simultaneously meeting continuous position, orientation, and curvature requirements, effectively avoiding control jitter or ride discomfort caused by sudden path changes or directional jumps. Furthermore, constraining the heading based on structured road information (such as the left lane line) gives the generated reference lines a natural lane-keeping or safe deviation capability, significantly improving the rationality and environmental consistency of the avoidance trajectory. In addition, the analytical form of the fifth-order polynomial facilitates efficient solution of the subsequent avoidance trajectory, balancing computational efficiency and trajectory quality, providing a reliable foundation for autonomous driving systems to achieve safe, comfortable, and compliant avoidance behavior in complex scenarios.

[0026] In one possible implementation of the first aspect above, determining a target avoidance trajectory that satisfies set constraints based on the target reference line and the obstacle and road information of the vehicle's surrounding environment includes: constructing a trajectory solution boundary based on the target reference line and the obstacle and road information of the vehicle's surrounding environment; and using a preset algorithm to determine a target avoidance trajectory that satisfies set constraints within the trajectory solution boundary.

[0027] Thus, by dynamically constructing the trajectory solution boundary by combining the target reference line, surrounding obstacles, and road information, and using preset algorithms such as QP or CILQR within this boundary to generate a target avoidance trajectory that meets safety, dynamics, and traffic rule constraints, this method effectively ensures the feasibility and environmental adaptability of the trajectory, avoids collisions or control instability caused by ignoring local constraints, and simultaneously considers driving safety, comfort, and compliance, thereby improving the reliability and real-time performance of the autonomous driving system in complex scenarios.

[0028] In one possible implementation of the first aspect above, the trajectory solution boundary includes at least one of the following: the left driving boundary and the right driving boundary of the vehicle; the distance between the vehicle's avoidance trajectory and the avoidance point of the avoidance target is less than or equal to the minimum lateral distance; the vehicle's avoidance trajectory does not include the positions of vulnerable road users in the set of positions along the lateral movement direction; the vehicle's avoidance trajectory does not include the positions of solid lane lines.

[0029] Thus, by introducing constraints such as driving boundaries, minimum lateral avoidance distance, vulnerable road user movement areas, and solid lane lines into the trajectory solution boundary, collision risks and traffic violations are eliminated from the source, effectively improving the safety, compliance, and environmental adaptability of the avoidance trajectory, and enhancing the decision-making reliability of the autonomous driving system in complex scenarios.

[0030] In one possible implementation of the first aspect mentioned above, the constraints include convergence conditions and feasibility constraints, and a preset algorithm is used to determine the target avoidance trajectory that satisfies the set constraints within the trajectory solution boundary. This includes: constructing a state-space model and a cost function using the preset algorithm; performing reverse solving on the avoidance trajectory in the current iteration process based on the cost function to obtain the control gain of each sampling point, where the control gain is related to the control quantity of the state-space model; performing forward updating on the avoidance trajectory in the current iteration process based on the control gain of each sampling point to obtain the state quantity and control quantity corresponding to each sampling point in the updated state-space model; determining whether the convergence conditions are met based on the change in the penalty value corresponding to the cost function during the current iteration process, where the convergence conditions include: the number of iterations reaching the maximum number of iterations, the cost difference decreasing by a percentage greater than a set percentage, and the absolute value of the cost difference being greater than a set absolute value; if the convergence conditions are met, determining whether the avoidance trajectory in the current iteration process satisfies the feasibility constraints, where the feasibility constraints are related to the vehicle's maximum steering angle and maximum lateral acceleration; and determining the avoidance trajectory in the current iteration process as the target avoidance trajectory if the feasibility constraints are met.

[0031] Thus, traditional methods that directly apply hard constraints are prone to failure, while this scheme approximates hard constraints through soft penalties, which retains the flexibility of optimization and ultimately ensures safety.

[0032] In one possible implementation of the first aspect above, the state variables in the state-space model include: vehicle position coordinates, vehicle heading angle, and vehicle front wheel steering angle; the control variable in the state-space model is the rate of change of the vehicle front wheel steering angle.

[0033] In one possible implementation of the first aspect described above, the cost function includes at least one of the following: reference line tracking cost, lateral acceleration cost, driving boundary cost, steering angle boundary cost, and steering angular velocity cost; wherein, the reference line tracking cost is used to characterize the degree of lateral deviation between each sampling point of the avoidance trajectory and the corresponding position of the target reference line, and the weight corresponding to a predetermined number of key sampling points at the end of the target reference line is greater than a first predetermined weight; the lateral acceleration cost is used to characterize the lateral acceleration corresponding to each sampling point in the avoidance trajectory; the driving boundary cost is used to characterize the lateral distance relationship between each sampling point in the avoidance trajectory and the left and right driving boundaries; the steering angle boundary cost is used to characterize the degree of deviation between the steering angle corresponding to each sampling point in the avoidance trajectory and the steering angle boundary angle; and the steering angular velocity cost is used to characterize the steering angular velocity corresponding to each sampling point in the avoidance trajectory.

[0034] In one possible implementation of the first aspect above, the method further includes: adding a cost augmentation penalty term to the cost function when the penalty value of the cost function does not meet the feasibility constraints, wherein the cost augmentation penalty term is related to the maximum steering angle constraint and / or the maximum lateral acceleration constraint; adjusting the penalty value of the cost function according to the cost augmentation penalty term until the penalty value of the cost function meets the feasibility constraints, and taking the avoidance trajectory in the current iteration as the target avoidance trajectory.

[0035] In one possible implementation of the first aspect above, the method further includes: performing reverse verification on the target avoidance trajectory, wherein the reverse verification includes at least one of the following: dynamic feasibility verification, collision detection verification, lane compliance verification, and curb safety distance verification.

[0036] As an example, the dynamic feasibility check verifies whether the avoidance trajectory meets the vehicle's steering or acceleration requirements. The collision detection check checks whether the trajectory conflicts with dynamic or static obstacles in time and space. The lane compliance check determines whether the trajectory illegally crosses solid lines or enters restricted areas. The curb safety distance check ensures that the trajectory maintains sufficient lateral distance from the curb. In this way, the final output target avoidance trajectory not only avoids obstacles but also meets multiple requirements such as dynamic constraints, road boundaries, and traffic rules.

[0037] In a second aspect, embodiments of this application provide a trajectory planning and decision-making device, comprising: a memory for storing instructions executed by one or more processors of the trajectory planning and decision-making device, and a processor, one of the processors of the trajectory planning and decision-making device, for executing the trajectory planning and decision-making method in the first aspect and any possible implementation thereof.

[0038] Thirdly, embodiments of this application provide a vehicle including the trajectory planning and decision-making device described in the second aspect.

[0039] Fourthly, embodiments of this application provide a readable medium storing instructions that, when executed on a trajectory planning and decision-making device, cause the device to perform the trajectory planning and decision-making method of the first aspect and any possible implementation thereof.

[0040] Fifthly, embodiments of this application provide a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the trajectory planning and decision-making method as described in the first aspect and any possible implementation thereof. Attached Figure Description

[0041] Figure 1 According to some embodiments of this application, a schematic diagram of the structure of a trajectory planning and decision-making device is shown.

[0042] Figure 2 According to some embodiments of this application, a schematic diagram of a vehicle coordinate system is shown.

[0043] Figure 3 According to some embodiments of this application, a flowchart of a trajectory planning and decision-making method is shown.

[0044] Figure 4A According to some embodiments of this application, a schematic diagram of a vehicle reference line is shown.

[0045] Figure 4B According to some embodiments of this application, a schematic diagram of a vehicle reference line is shown.

[0046] Figure 5 According to some embodiments of this application, a schematic diagram of a target reference line screening process is shown.

[0047] Figure 6 According to some embodiments of this application, a schematic diagram of a reference line for verification based on static obstacles is shown.

[0048] Figure 7 According to some embodiments of this application, a schematic diagram of a reference line traversing a dynamic obstacle is shown.

[0049] Figure 8 According to some embodiments of this application, a schematic diagram of a reference line generation process framework is shown.

[0050] Figure 9 According to some embodiments of this application, a schematic diagram of a solution process for an obstacle avoidance trajectory is shown.

[0051] Figure 10 According to some embodiments of this application, a schematic diagram of high-weight setting of sampling points based on the CILQR algorithm is shown.

[0052] Figure 11 According to some embodiments of this application, a flowchart of obstacle avoidance path planning is shown.

[0053] Figure 12 According to some embodiments of this application, a schematic diagram of a trajectory optimization process based on the CILQR algorithm is shown.

[0054] Figure 13 According to some embodiments of this application, a structural schematic diagram of a vehicle is shown. Detailed Implementation

[0055] The illustrative embodiments of this application include, but are not limited to, trajectory planning and decision-making methods, devices, vehicles, media, and program products.

[0056] This application relates to the fields of intelligent driving and active safety technology, such as trajectory planning scenarios for emergency avoidance in automatic emergency steering (AES) systems on structured roads.

[0057] To improve the emergency avoidance performance of vehicles, this application provides a trajectory planning and decision-making method. The method includes: during vehicle operation, detecting an avoidance target, such as a static or dynamic target in front of the vehicle; determining at least two target reference lines for the vehicle to bypass the target, whereby the target reference lines indicate the detour direction and endpoint. Two of the at least two target reference lines have different detour directions, for example, a detour direction that crosses the lane to the left or a detour direction that crosses the current lane to the right. Then, based on the target reference lines and obstacles and curbs in the vehicle's surrounding environment, a target avoidance trajectory that meets the constraints can be determined. Finally, the vehicle is steered based on the target avoidance trajectory to bypass the target.

[0058] Thus, this application can use the target reference line to obtain prior information such as the avoidable space (e.g., the space between the starting point and the ending point) and the detour direction, and then determine the avoidance trajectory based on the target reference line. This ensures more reasonable avoidance decisions while reducing computational power consumption. Furthermore, combining structured road information such as obstacles and curbs for trajectory planning can cover avoidance decisions in emergency avoidance scenarios, including curves, improving emergency avoidance performance in curve scenarios. Therefore, it enhances emergency avoidance performance in various scenarios.

[0059] It is understood that the trajectory planning and decision-making method in this application can realize the evaluation mechanism for cross-lane avoidance and lane-in-lane avoidance, the reference line generation strategy, trajectory planning and other control links, and is applicable to road scenarios with clear lane lines such as highways and urban expressways, especially supporting emergency avoidance under curve conditions.

[0060] In some embodiments, the trajectory planning and decision-making method provided in this application can be applied to emergency turning and avoidance trajectory planning and decision-making scenarios of AES function.

[0061] In some embodiments, the subject executing the trajectory planning and decision-making method provided in this application may be a vehicle or a device or equipment in a vehicle (also known as a trajectory planning and decision-making device), and this application does not specifically limit this.

[0062] Reference Figure 1 The diagram shown is a structural schematic of the trajectory planning and decision-making device provided in an embodiment of this application. Figure 1 The trajectory planning and decision-making device 10 shown includes an environmental information processing module 11, a planning module 12, and a decision-making module 13.

[0063] The environmental information processing module 11 is used to collect, filter, analyze, and preprocess environmental information around the main vehicle to obtain preprocessed data, providing accurate and effective input data for the subsequent planning module 12 and decision-making module 13. As an example, the preprocessed data for the main vehicle includes: the main target, environmental targets, and road information. The main target is the target that the main vehicle needs to avoid under the AES function; the environmental targets are all dynamic and static obstacles (i.e., environmental targets) in the main vehicle's lane and adjacent lanes within a preset future travel time; the road information includes lane line information (including lane line position, direction, type, etc.) and road curb information for the main vehicle's lane and adjacent lanes. For example, the environmental information processing module 11 can collect images around the main vehicle through one or more onboard cameras and obtain environmental information based on these images.

[0064] In some embodiments, the environmental information processing module 11 includes an AES main target filtering submodule 111, an AES environmental target processing submodule 112, and an AES road information processing submodule 113.

[0065] The AES primary target filtering submodule 111 is used to filter out the primary targets that need to be avoided by the AES function based on all targets detected in the environment around the main vehicle. As an example, the primary targets of the AES function are consistent with the primary targets filtered by the AEB function to ensure the consistency and continuity of target recognition in emergency scenarios and avoid false or missed triggering of functions due to differences in targets.

[0066] The AES environmental target processing submodule 112 is used to filter out dynamic and static targets, i.e., obstacles, that will interfere with the driving space of the main vehicle in its own lane and adjacent lanes within a preset time period, based on all targets detected in the environment surrounding the main vehicle. As an example, the preset time can be 4 seconds (s). This preset time can be calibrated and adjusted according to the actual vehicle model, application scenario, and safety requirements to ensure that the filtered targets can fully cover the potential driving risks of the main vehicle.

[0067] The AES road information processing submodule 113 is used to detect the road information of the main vehicle in the environment around the main vehicle, namely the lane line information of the main vehicle's own lane and adjacent lanes (including lane line position, direction, type, etc.) and road edge information, providing road geometric basis data for subsequent reference line generation, avoidance space judgment and trajectory boundary constraints.

[0068] The planning module 12 is used to generate reference lines, define avoidance boundaries, solve for optimal trajectories, and verify the validity of trajectories based on the preprocessed data output by the environmental information processing module 11, so as to realize the scientific planning of avoidance trajectories in emergency scenarios.

[0069] In some embodiments, the planning module 12 includes: a reference line generation and main target detour decision submodule 121, a boundary generation submodule 122, a planning and trajectory solution submodule 123, and a planning result back verification submodule 124.

[0070] The reference line generation and main target detour decision submodule 121 is used to generate detour decision results based on the main target, surrounding environmental targets, and road information output by the environmental information processing module 11. These detour decision results include four reference lines. These reference lines are used to determine the optimal detour direction for the main target, pre-verify whether the avoidance space for each detour direction meets safety requirements, and select a detour lane suitable for the current scenario, providing basic guidance for subsequent trajectory solving.

[0071] The boundary generation submodule 122 is used to generate the detour decision results output by the main target detour decision submodule 121 based on the aforementioned reference line, and combine them with the environmental targets selected by the environmental target processing submodule 11 and the road information extracted by the road information processing submodule 11 to define the boundary constraints for trajectory solving. These boundary constraints are used to limit the travel range of the trajectory and prevent the trajectory from exceeding the road boundary or colliding with obstacles.

[0072] The trajectory planning and solving submodule 123 is used to solve for the avoidance trajectory by employing a preset algorithm with the endpoint position of the reference line as the target reference, combined with the boundary constraints defined by the boundary generation submodule 122, thus obtaining the target avoidance trajectory. In this way, through this optimized solution method, the generated trajectory can be ensured to balance safety, smoothness, and dynamic feasibility. For example, the preset algorithm can be a quadratic programming (QP) algorithm, a constrained iterative linear quadratic regulator (CILQR) algorithm, or other optimization algorithms.

[0073] The planning result reverse verification submodule 124 is used to perform dual validity verification and validation on the target avoidance trajectory obtained through the planning and trajectory solving submodule 123. For example, it verifies whether the target avoidance trajectory meets the dynamic constraints of the main vehicle (including maximum steering angle, steering angular velocity, acceleration, deceleration, etc.) to ensure that the trajectory can be actually executed by the main vehicle. In addition, it re-verifies the safe distance between the trajectory and obstacles in the surrounding environment to confirm that the trajectory will not collide with any obstacles, ensuring the safety and reliability of the trajectory.

[0074] In some embodiments, the decision module 13 is used to make decisions and risk verifications for the AES function based on the preprocessed data output by the environmental information processing module 11 and the target avoidance estimation output by the planning module 12, combined with the driver's operating intention and scenario confidence, so as to avoid the function being triggered erroneously.

[0075] In some embodiments, the decision module 13 includes a driver intent calculation submodule 131, a confidence calculation submodule 132, and an AES decision processing submodule 133.

[0076] The driver intent calculation submodule 131 is used to collect and analyze the driver's real-time operating behavior (including steering, braking, and throttle operations), calculate the driver's takeover intent and the need to suppress the AES function. When a clear takeover operation is detected, the intervention of the AES function can be suppressed to ensure driver control and improve the human-machine collaboration of the function.

[0077] The confidence calculation submodule 132 is used to calculate the scene confidence based on obstacles, road information, driver operation information, and trajectory planning results for different emergency driving scenarios. Then, by judging the scene confidence, it is possible to effectively distinguish between real emergency scenarios and misidentified scenarios, avoid false alarms and false interventions of the AES function due to scene misjudgment, and improve the reliability of the function.

[0078] The AES decision processing submodule 133 is used to perform risk decision verification on the timing and intensity of AES function intervention based on the target avoidance trajectory output by the planning module 12, the driver takeover intention output by the driver intention calculation submodule 131, and the scene confidence level output by the confidence level calculation submodule 132. Only when the trajectory is safe and effective, the scene confidence level meets the preset confidence threshold, and the driver has no clear takeover intention, will the AES function be triggered to perform the avoidance operation, ensuring the safety and rationality of the function intervention.

[0079] Reference Figure 2 The diagram shown is a schematic representation of a vehicle coordinate system provided in an embodiment of this application. For example, taking vehicle 100 as an example, the vehicle coordinate system includes an x-axis and a y-axis. The x-axis represents the vertical direction of the vehicle coordinate system, with the front direction of the vehicle as the positive direction. The y-axis represents the horizontal direction of the vehicle coordinate system, with the left side of vehicle 100 as the positive direction. Furthermore, the origin of the coordinate axes is the center of the rear axle of vehicle 100. In the following embodiments, the vehicle coordinate system is... Figure 2 The vehicle coordinate system shown is used as an example for explanation.

[0080] Reference Figure 3 The diagram shown is a flowchart of a trajectory planning and decision-making method provided in an embodiment of this application. The main body executing this process is... Figure 1The trajectory planning and decision-making device 10 is shown. Specifically, this process can be a process for controlling the steering of vehicle 100 to avoid obstacles, and the process includes the following steps:

[0081] S301: During vehicle operation, trajectory planning and decision-making equipment 10 detects the avoidance target of vehicle 100.

[0082] As an example, taking vehicle 100 as an example, the target that vehicle 100 avoids can be the main target of the AES function, that is, the main target of the ABE function.

[0083] As an example, the trajectory planning and decision-making device 10 obtains the environmental information of the vehicle 100 and obtains preprocessed data through the environmental information processing module 11, and analyzes the preprocessed data through the AES main target filtering submodule 111 to determine the avoidance target of the vehicle 100.

[0084] S302: The trajectory planning and decision-making device 10 acquires at least two reference lines of the vehicle 100.

[0085] As an example, the trajectory planning and decision-making device 10 obtains at least two reference lines through the reference line generation and main target detour decision submodule 121 in the planning module 12.

[0086] It is understandable that at least two of the reference lines have different directions of detour.

[0087] As an example, at least two reference lines include at least one or more of the following: a reference line for detouring to the left within the current lane, a reference line for detouring to the left across a lane, a reference line for detouring to the right within the current lane, and a reference line for detouring to the right across a lane; wherein the endpoint of each reference line is separated from the vehicle's starting point by a first distance in a first direction, and is separated from the lane lines associated with the detouring direction by a second distance in a second direction. For example, the first distance is the distance the vehicle will travel in a preset future time, such as the distance the vehicle will travel at a set speed (such as the current speed) in the next 4 seconds. The second distance is 0.8 meters (m). In this way, by pre-setting multiple reference lines for left / right, within the current lane / across a lane, and uniformly constraining their endpoint positions with the offset of the longitudinal first distance and the lateral second distance, high-quality candidate paths covering typical obstacle avoidance scenarios are generated in a structured manner. Thus, the adequacy of detouring and driving safety are balanced, planning efficiency and scenario adaptability are improved, and clear and reasonable boundary conditions are provided for subsequent trajectory optimization, effectively supporting the reliable obstacle avoidance of the autonomous driving system in complex environments.

[0088] refer to Figure 4A The image shown is a schematic diagram of a vehicle reference line provided in an embodiment of this application. Figure 4A As shown, Figure 4AThe diagram shows the starting point H of the main vehicle, the left avoidance point TL, the right avoidance point TR, and the four endpoints L1, L2, R1, and R2. At this point, Figure 4A The reference lines shown are: the reference line for detouring to the left within the same lane where the endpoint L1 is located (denoted as reference line 1); the reference line for detouring to the left across the lane where the endpoint L2 is located (denoted as reference line 2); the reference line for detouring to the right within the same lane where the endpoint R1 is located (denoted as reference line 3); and the reference line for detouring to the right across the lane where the endpoint R2 is located (denoted as reference line 4). Furthermore, Figure 4A The distances S (i.e., the first distance) between the main vehicle and the endpoints L1, L2, R1, and R2 along the driving direction (i.e., the first direction) are the distances within the next 4 seconds at the main vehicle's current speed. The distances d (i.e., the second distances) between endpoints L2 and L1 and the lane lines for left-turning are both 0.8m, as are the distances d (i.e., the second distances) between endpoints R2 and R1 and the lane lines for right-turning are also 0.8m. The left avoidance point TL and the right avoidance point TR are the left and right boundary points of the main target, respectively. It can be understood that the left avoidance point TL and the right avoidance point TR are positional constraints of the reference lines, while the endpoints L1, L2, R1, and R2 are positional and heading constraints of the reference lines. At this time, at least two of the above reference lines are... Figure 4A Reference lines 1, 2, 3 and 4 are shown.

[0089] In some embodiments, the reference line is fitted with a set of polynomials, wherein the polynomials include at least one of the following: a polynomial constraining the position of the vehicle's starting point, a polynomial constraining the position of the vehicle's avoidance point, a polynomial constraining the vehicle's heading at the starting point, a polynomial constraining the vehicle's heading at the ending point, or a polynomial constraining the curvature of the vehicle at the starting point. The position of the vehicle's avoidance point is associated with the position of the avoidance target and the detour direction, and the heading of the vehicle's ending point is associated with the lane line associated with the detour direction. For example, the heading of the vehicle's ending point coincides with the lane line associated with the detour direction.

[0090] As an example, combined Figure 4A The four endpoints L1, L2, R1 and R2 shown, as well as the starting point H of the main vehicle, the left avoidance point TL and the right avoidance point TR, can be combined into a fifth-degree polynomial based on the following formula (1).

[0091] (1)

[0092] Among them, the coefficients in formula (1) , , , and The value of is determined based on actual needs, and no specific limitation is made here. Representing position coordinates, such as coordinates below the x-axis. This indicates the position coordinates, such as the coordinates on the y-axis.

[0093] by Figure 4A Taking the path (H-TL-L1) from the starting point H of the main vehicle through the left avoidance point TL to the end point L1 as an example, the coordinate points under this path are combined based on formula (1) to obtain the fifth-degree polynomial shown in formulas (2) to (7).

[0094] (2)

[0095] (3)

[0096] (4)

[0097] (5)

[0098] (6)

[0099] (7)

[0100] Formulas (2) to (4) above represent the positional constraints of the starting point H, the left avoidance point TL, and the ending point L1, respectively. Formulas (5) to (6) represent the heading constraints of the starting point H and the ending point L1, wherein the heading is consistent with the lane line associated with the position (such as the left lane line of the main vehicle's lane). Formula (7) represents the curvature constraint of the starting point H. At this time, the fifth-order polynomial shown in formulas (2) to (7) represents reference line 1.

[0101] It can be understood that formula (2) represents a polynomial used to constrain the position of the vehicle's starting point, formula (3) represents a polynomial used to constrain the position of the vehicle's avoidance point, formula (4) represents a polynomial used to constrain the position of the vehicle's avoidance point, formula (5) represents a polynomial used to constrain the vehicle's heading at the starting point, formula (6) represents a polynomial used to constrain the vehicle's heading at the end point, and formula (7) represents a polynomial used to constrain the curvature of the vehicle at the starting point.

[0102] Similarly, the polynomials of at least two reference lines, i.e., the other reference lines among the four reference lines, can be referred to the fifth-degree polynomials corresponding to reference line 1 shown in formulas (2) to (7). This application embodiment will not elaborate on this.

[0103] Thus, this embodiment utilizes polynomials, such as fifth-order polynomials, to accurately generate smooth reference lines that conform to road geometry and vehicle kinematics. This ensures that the reference lines accurately fit the desired detour path in space, while simultaneously meeting continuous position, direction, and curvature requirements, effectively avoiding control jitter or ride discomfort caused by sudden path changes or directional jumps. Furthermore, constraining the heading based on structured road information (such as the left lane line) gives the generated reference lines a natural lane-keeping or safe deviation capability, significantly improving the rationality and environmental consistency of the avoidance trajectory. In addition, the analytical form of the fifth-order polynomial facilitates efficient solution of the subsequent avoidance trajectory, balancing computational efficiency and trajectory quality, providing a reliable foundation for autonomous driving systems to achieve safe, comfortable, and compliant avoidance behavior in complex scenarios.

[0104] In some embodiments, assuming that the speed of the main vehicle remains constant throughout the entire length of each reference line, information such as the position, heading and curvature of the sampling points can be obtained by sampling at fixed longitudinal position intervals (e.g., sampling interval of 1m) to generate each reference line, that is, to generate the fifth-order polynomial of each reference line.

[0105] For ease of description, the reference line will be referred to as a polynomial reference line, a fifth-degree polynomial reference line, or a polynomial reference line in some places in the following embodiments, but this does not affect the nature of the reference line.

[0106] S303: The trajectory planning and decision-making device 10 determines the target reference line for the vehicle 100 to bypass the target. The target reference line is used to indicate the bypass direction and destination of the vehicle to bypass the target. The target reference line is one of at least two reference lines.

[0107] That is, the trajectory planning and decision-making device 10 determines the target reference line from at least two reference lines.

[0108] In some embodiments, the planning and trajectory solving submodule 123 in the planning module 12 can determine a target reference line from at least two reference lines for the vehicle to bypass the target. For example, the target reference line can be... Figure 3 The reference line 2 shown is the left-hand lane detour reference line where the endpoint L2 is located.

[0109] In some embodiments, after acquiring at least two reference lines, each reference line can be scored, and a target reference line can be selected based on the scores of each reference line.

[0110] S304: The trajectory planning and decision-making device 10 determines the target avoidance trajectory that meets the set constraints based on the target reference line and the obstacle and road information of the surrounding environment of the vehicle 100.

[0111] In some embodiments, the aforementioned constraints can limit the travel range of the trajectory to prevent it from exceeding road boundaries or colliding with obstacles. Alternatively, the preset conditions can also limit constraints such as the maximum steering angle and maximum lateral acceleration to ensure that the trajectory can be actually executed by the host vehicle, such as vehicle 100. Therefore, a target avoidance trajectory that meets the set constraints is an avoidance trajectory that will not collide with any obstacles and possesses safety and reliability.

[0112] In some embodiments, obstacles in the surrounding environment of the vehicle 100 can be dynamic and static obstacles that will affect the driving space of the vehicle in its lane and adjacent lanes within a preset timeframe in the future. The road information can include lane line information (including lane line position, direction, type, etc.) of the vehicle 100 in its lane and adjacent lanes, as well as roadside information.

[0113] As an example, the trajectory planning and decision-making device 10 can filter out obstacles in the environment surrounding the vehicle 100 through the AES environmental target processing submodule 112.

[0114] As an example, the trajectory planning and decision-making device 10 can detect the road information of the main vehicle, i.e., vehicle 100, in the environment surrounding the main vehicle through the AES road information processing submodule 113.

[0115] In some embodiments, the trajectory planning and decision-making device 10 constructs a trajectory solution boundary based on the target reference line and obstacle and road information in the surrounding environment of the vehicle; and employs a preset algorithm to determine a target avoidance trajectory that satisfies the constraints within the trajectory solution boundary. For example, the preset algorithm can be the QP algorithm or the CILQR algorithm.

[0116] Thus, by dynamically constructing the trajectory solution boundary by combining the target reference line, surrounding obstacles, and road information, and using preset algorithms such as QP or CILQR within this boundary to generate a target avoidance trajectory that meets safety, dynamics, and traffic rule constraints, this method effectively ensures the feasibility and environmental adaptability of the trajectory, avoids collisions or control instability caused by ignoring local constraints, and simultaneously considers driving safety, comfort, and compliance, thereby improving the reliability and real-time performance of the autonomous driving system in complex scenarios.

[0117] In some embodiments, the trajectory solution boundary includes at least one of the following: the left and right driving boundaries of the vehicle, such as vehicle 100; the distance between the vehicle's avoidance trajectory and the avoidance point of the avoidance target is less than or equal to the minimum lateral distance; the vehicle's avoidance trajectory does not include the positions of vulnerable road users in the set of positions along the lateral movement direction; and the vehicle's avoidance trajectory does not include the positions of solid lane lines. Thus, by introducing constraints such as driving boundaries, minimum lateral avoidance distance, vulnerable road user movement areas, and solid lane lines into the trajectory solution boundary, collision risks and traffic violations are eliminated from the source, effectively improving the safety, compliance, and environmental adaptability of the avoidance trajectory, and enhancing the decision-making reliability of the autonomous driving system in complex scenarios.

[0118] Reference Figure 4B The diagram shown is a schematic representation of a vehicle reference line provided in an embodiment of this application. Figure 4B The reference line generation scene shown is similar to Figure 4A The only difference in the reference line generation scenario shown is that... Figure 4B The roads in the text are curved, meaning the road lines are curved; similarities will not be elaborated further.

[0119] Thus, before using a computationally intensive pre-defined algorithm such as CILQR to solve the problem, this embodiment uses reference lines to perform prior calculations on the avoidable space, direction, and rate of change, resulting in a more reasonable avoidance decision direction. Simultaneously, since fine-grained trajectory solving requires significant computational power, this solution uses coarse-grained pre-solution with reference lines to effectively avoid the unnecessary computational consumption caused by performing full a posteriori calculations on multiple avoidance directions, based on a pre-determined optimal avoidance direction. Furthermore, by combining structured road information, it can cover emergency avoidance scenarios in curved road conditions, and by utilizing pre-defined algorithms such as QP or CILQR, it can cover continuous avoidance scenarios, improving the scenario adaptability of the avoidance algorithm.

[0120] S305: Trajectory planning and decision-making device 10 controls vehicle 100 to steer around and avoid the target based on target avoidance trajectory.

[0121] As an example, the trajectory planning and decision-making device 10 can make decisions through the decision module 13 and control the vehicle 100 to steer around the target based on the target avoidance trajectory. For instance, when the AES decision processing submodule 133 detects that the target avoidance trajectory is safe and effective, the scenario confidence level meets the preset confidence threshold, and the driver has no clear intention to take over, the AES function decision is triggered, and the vehicle 100 is controlled to steer around the target based on the target avoidance trajectory, ensuring the safety and rationality of the AES function intervention.

[0122] It is understood that after detecting the obstacle avoidance target, this application generates at least two reference lines with different detour directions and determines a target reference line. Combining this with obstacle distribution and road information in the vehicle's surrounding environment, it further filters out target avoidance trajectories that meet multi-dimensional constraints such as dynamics, safety, and traffic rules, thereby achieving precise control of the vehicle's steering and obstacle avoidance. This solution significantly improves the vehicle's obstacle avoidance flexibility and path adaptability in complex dynamic scenarios, effectively avoiding collision risks or violations that may result from a single avoidance strategy. Simultaneously, the generated avoidance trajectory is smooth and continuous, balancing ride comfort and driving safety, and providing reliable and efficient local path replanning capabilities.

[0123] Thus, this application can use the target reference line to perform prior calculations on information such as the avoidable space (e.g., the space between the starting point and the ending point) and the detour direction, and then determine the avoidance trajectory based on the target reference line, ensuring more reasonable avoidance decisions. Meanwhile, since fine-grained trajectory solving requires high computational power, this scheme uses coarse-grained pre-solution of the reference line, effectively avoiding the unnecessary computational consumption caused by performing full a posteriori calculations for multiple avoidance directions, based on the pre-determined optimal avoidance direction. Furthermore, combining structured road information such as obstacles and curbs for trajectory planning can cover avoidance decisions in emergency avoidance scenarios in curves, improving the emergency avoidance effect in curve scenarios. Therefore, it improves the emergency avoidance effect in various scenarios.

[0124] Reference Figure 5 The diagram shown illustrates the target reference line selection process provided in this embodiment. The execution entity of this process can be the trajectory planning and decision-making device 10, and the process includes the following steps:

[0125] S501: The trajectory planning and decision-making device 10 generates M verification scores for each of at least two reference lines.

[0126] Among them, the M verification scores include at least one of the following: obstacle verification score, urgency verification score, smoothness verification score, lane keeping verification score, historical decision consistency verification score, roadside safety verification score, crossing solid line verification score, and vulnerable road user traversing direction verification score.

[0127] As an example, in obstacle detection, the intersection time when the main vehicle (e.g., vehicle 100) collides with other dynamic and static obstacles around it, excluding the target it is avoiding. The intersection time between the obstacle's trajectory and the reference line represented by the polynomial is used as an evaluation index. The longer the intersection time, the lower the probability of the reference line intersecting with the obstacle, and therefore the higher the obstacle verification score.

[0128] In some embodiments, the trajectory planning and decision-making device 10 determines the intersection time between each obstacle in the surrounding environment of the vehicle 100 and the reference line based on the type of obstacle. Furthermore, the trajectory planning and decision-making device 10 determines the obstacle verification score of the reference line based on the minimum intersection time and the maximum default intersection time.

[0129] Reference Figure 6 The diagram shown is a schematic representation of a reference line for verification based on static obstacles, provided in an embodiment of this application. Figure 6 As shown, there are moving targets A, stationary targets B, and moving targets C surrounding the main target vehicle. The algorithm iterates through all obstacle targets, verifying the intersection time between the predicted trajectory of each obstacle target and the polynomial reference line (i.e., the reference line where the endpoint L2 is located). Intersection is defined as the intersection of the target's trajectory with the reference line, and the intersection time is... There are two definitions:

[0130] Definition (1): The intersection point is based on the corresponding time of the predicted trajectory of the obstacle target. Sure.

[0131] Definition (2): The intersection point is based on the corresponding time of the reference line. Sure.

[0132] In some embodiments, when the obstacle is a dynamic obstacle, the trajectory planning and decision-making device 10 determines whether the obstacle intersects with a reference line. If the obstacle intersects with the reference line, a first time (i.e., the time mentioned above) is determined based on the predicted trajectory of the obstacle. ), and determine the second time (i.e., the time mentioned above) based on the reference line. Furthermore, if the difference between the first time and the second time is less than the set time difference, the minimum of the first time and the second time is determined as the corresponding intersection time; if the difference between the first time and the second time is greater than or equal to the set time difference, the default maximum time is determined as the corresponding intersection time. If the obstacle does not intersect the reference line, the default maximum time is determined as the corresponding intersection time.

[0133] As an example, if time With time If the time difference is less than the preset time difference (e.g., 3 seconds), it is considered that the obstacle target has a high probability of subsequent collision. Therefore, the intersection time corresponding to the obstacle target is... For time and time The minimum value in.

[0134] As an example, if time With time If the time difference is greater than 3 seconds or there is no intersection, it is considered that the obstacle target is unlikely to collide with the reference line later. Therefore, the intersection time is considered to be... Defined as the default maximum time (e.g., 6s).

[0135] In some embodiments, when the obstacle is a static obstacle, the trajectory planning and decision-making device 10 determines whether the obstacle intersects with the reference line; if the obstacle intersects with the reference line, the default minimum time is determined as the corresponding intersection time; if the obstacle does not intersect with the reference line, the default maximum time is determined as the corresponding intersection time.

[0136] As an example, if there is a static obstacle target that intersects with the trajectory of the reference line, then the intersection time of the obstacle target... Defined as 0s.

[0137] As an example, if there is a static obstacle target that intersects with the trajectory of the reference line, then the intersection time of the obstacle target... Defined as the default maximum time .

[0138] by Figure 6 For example, the intersection time of a dynamic obstacle target, target A, is 2.0s, while the corresponding time on the reference line is 2.5s. Target B is a stationary obstacle target with an intersection time of 0s, while the corresponding time on the reference line is 3.5s. Target C does not intersect with the current polynomial reference line, and its intersection time is the default maximum time. Furthermore, the trajectory planning and decision-making device traverses 10 times to obtain all obstacle targets around the main vehicle, and obtains the default minimum intersection time. The obstacle check score is calculated by the square shown in formula (8) below. The obstacle verification score The value range is [0,1].

[0139] (8)

[0140] If the obstacle verification score in formula (8) is lower than the corresponding preset threshold, such as 0.3 points (which can be calibrated), then the current polynomial reference line is considered to have an environmental collision risk, that is, the availability of the polynomial reference line is invalid.

[0141] In some embodiments, the urgency level verification score is calculated based on the maximum lateral acceleration corresponding to the polynomial reference line. The maximum lateral acceleration characterizes the urgency of the lane change or avoidance maneuver described by the reference line; a smaller value indicates a smoother trajectory change, safer driving behavior, and a higher urgency level verification score. The maximum lateral acceleration is further limited by the maximum lateral acceleration allowed by the adhesion limit of the vehicle's current road surface to ensure that the generated reference line conforms to vehicle dynamics feasibility.

[0142] In some embodiments, for each of at least two reference lines, the trajectory planning and decision-making device 10 can determine the maximum curvature among the curvatures of each sampling point in the reference line; determine the maximum lateral acceleration of the reference line based on the maximum curvature and a set vehicle speed; determine the acceleration difference between the square of the set lateral acceleration and the square of the maximum lateral acceleration; determine the ratio of the acceleration difference to the square of the set lateral acceleration; and determine the urgency verification score of the reference line.

[0143] As an example, for each polynomial reference line The trajectory planning and decision-making device 10 traverses multiple sampling points within a preset sampling interval, calculates the curvature at each sampling point, and determines the sampling point corresponding to the maximum curvature (i.e., the maximum curvature). For example, the curvature of the sampling point... Based on the following formula (9):

[0144] (9)

[0145] As an example, based on the maximum curvature and the vehicle's current longitudinal velocity, the maximum lateral acceleration corresponding to the polynomial reference line is calculated using the following formula (10). :

[0146] (10)

[0147] in, For maximum curvature, This represents the longitudinal speed of the vehicle.

[0148] As an example, the default lateral acceleration for setting the adhesion limit is... The urgency score is calculated according to the following formula (11). :

[0149] (11)

[0150] The default lateral acceleration value can be set according to actual needs, such as 7 meters per second squared (m / s²). Urgency level verification score. The value range is [0,1]. If the score is lower than the corresponding preset threshold (which can be calibrated, such as 0.1), the current polynomial reference line is determined to have a high planning risk, its availability is marked as invalid, and it will not participate in the subsequent target reference line selection process.

[0151] Thus, the scoring mechanism based on urgency verification effectively ensures that the selected reference line is within the vehicle dynamics feasible domain, avoiding the risk of loss of control due to excessive trajectory curvature, and significantly improving the safety and reliability of the avoidance path.

[0152] In some embodiments, the smoothness check score is evaluated based on the rate of change of lateral acceleration (i.e., lateral jerk, or lateral jerk) of the polynomial reference line.

[0153] In some embodiments, the smoothness verification score of the reference line is determined by: determining the root mean square of the lateral acceleration rate of change of the reference line; determining the root mean square of the default lateral acceleration rate of change of the reference line; determining the difference between the root mean square of the default lateral acceleration rate of change and the root mean square of the lateral acceleration rate of change, and determining the smoothness verification score of the reference line based on the ratio of the difference to the root mean square of the default lateral acceleration rate of change.

[0154] As an example, the trajectory planning and decision-making device 10 uses a polynomial reference line. The derivative of the lateral acceleration is used to calculate the lateral acceleration change rate (jerk) value corresponding to each sampling point within the preset sampling interval of the polynomial reference line, and the root mean square (RMS) value of the lateral jerk is further calculated. The RMS value characterizes the overall fluctuation of the lateral acceleration change rate and reflects the smoothness of the trajectory and the ride comfort.

[0155] As an example, a default rate of change of lateral acceleration is set. The smoothness check score of the reference line is 20 (calibrable), calculated using the following formulas (12) to (14). .

[0156] (12)

[0157] (13)

[0158] (14)

[0159] in, The root mean square of the rate of change of lateral acceleration. This represents the root mean square of the default rate of change of lateral acceleration. This represents the rate of change of lateral acceleration at the k-th sampling point. This represents the default rate of change of lateral acceleration, where n is the number of sampling points.

[0160] It's understandable that the smoothness verification score is negatively correlated with the root mean square value of the lateral acceleration rate of change: the smaller the root mean square value of the lateral acceleration rate of change, the smoother the trajectory change, the more comfortable the driving experience, and the higher the corresponding smoothness verification score. Conversely, if the lateral acceleration rate of change fluctuates drastically, the smoothness verification score decreases. The smoothness verification score is used to prioritize the trajectory with smoother kinematics and better occupant comfort among at least two reference lines, thereby improving the driving quality of the autonomous driving system.

[0161] In some embodiments, the lane keeping accuracy score is evaluated based on whether the endpoint of the reference line is located within the driver's lane. Specifically, if the polynomial reference line is the polynomial reference line containing endpoints L1 and R1, then that reference line is assigned a fixed bonus. This bonus mechanism aims to encourage obstacle avoidance trajectories to complete obstacle avoidance within the driver's lane, avoiding unnecessary lane-crossing behavior, thereby improving driving safety and traffic rule compliance.

[0162] In some embodiments, if the reference line is a reference line that detours to the left within the current lane (i.e., the reference line where the endpoint L1 is located) or a reference line that detours to the right within the current lane (i.e., the reference line where the endpoint R1 is located), the trajectory planning and decision-making device 10 determines the upper limit of the lane score (e.g., 1) as the lane keeping verification score of the reference line.

[0163] In other embodiments, if the reference line is a reference line that crosses the lane to the left (i.e., the reference line where the endpoint L2 is located) or a reference line that crosses the lane to the right (i.e., the reference line where the endpoint R2 is located), the trajectory planning and decision-making device 10 determines the lower limit of the lane score (e.g., 0) as the lane keeping verification score of the reference line, and the upper limit of the lane score is greater than the lower limit of the lane score.

[0164] As an example, refer to the following formula (15) for the lane keeping check score. According to the calculation formula, if the endpoint of the reference line is L1 or R1, then the lane keeping test score is... The score is 1; if the endpoint of the reference line is L2 or R2, the lane keeping check score is 1. It is 0.

[0165] (15)

[0166] In some embodiments, if the endpoint of the detour indicated by the reference line is the same as the endpoint of the detour corresponding to the historical detour decision information, the trajectory planning and decision-making device 10 determines the upper limit of the historical decision score as the historical decision information verification score of the reference line. If the endpoint of the detour indicated by the reference line is different from the endpoint of the detour corresponding to the historical detour decision information, the trajectory planning and decision-making device 10 determines the lower limit of the historical decision score as the historical decision information verification score of the reference line. The upper limit of the historical decision score (e.g., 0.5) is greater than the lower limit of the historical decision score (e.g., 0). The historical detour decision information includes historical reference lines generated for the avoidance target in the previous control cycle, and the endpoint of the detour indicated by the historical detour decision information is the endpoint of the detour indicated by the historical reference line.

[0167] It is understandable that the historical decision information verification score is based on a consistency assessment of historical avoidance decisions for the same avoidance target. For example, if the avoidance target in the current planning period is the same as that in the previous planning period, the avoidance position selected as the target reference line in the previous period (e.g., endpoint L2) is identified, and additional bonus points are assigned to the reference line ending at this historically high-scoring avoidance position in the current period. This bonus mechanism is used to enhance the weight of historically effective decisions in the current assessment, thereby improving the consistency of avoidance strategies over time.

[0168] As an example, refer to the historical decision information verification score shown in the following formula (16). According to the calculation formula, if the endpoint of the current reference line is the same as the endpoint of the previous reference line, then the historical decision information verification score is... The score is 0.5; if the endpoint of the current reference line is different from the endpoint of the previous reference line, the historical decision information verification score is [not specified]. It is 0.

[0169] (16)

[0170] In some embodiments, the safety verification score of the reference line on the roadside is determined by the trajectory planning and decision-making device 10 determining the minimum distance between the reference line and discrete points on the roadside in the vehicle's surrounding environment. If the minimum distance Greater than or equal to the set safety distance Then the upper limit of the curb score (e.g., 1) is determined as the curb-side safety check score of the reference line; if the minimum distance Less than the set safe distance or greater than the set danger distance Then determine the minimum distance and the set danger distance. The square of the difference is the first result, used to determine the set safety distance. minimum distance The square of the difference is the second result, and the ratio of the first and second results is determined as the safety check score for the roadside of the reference line; if the minimum distance Less than or equal to the set danger distance If the lower limit of the curb score (e.g., 0) is determined as the curb-side safety check score of the reference line.

[0171] It's understandable that curbside calibration is based on scoring reference lines along the curb, deducting points for potential collisions or reference line intersections. As an example, the curb is represented as a scatter plot, and the distance between each curb and the corresponding position on the polynomial reference line is checked one by one in the vehicle coordinate system. The minimum distance from the curb to the polynomial reference line is then obtained. Set the danger distance Set a safe distance Roadside verification score The calculation formula can be the following formula (17).

[0172] As an example,

[0173] (17)

[0174] As an example, if the safety verification score on the curb side is lower than the corresponding preset threshold, such as 0.1 points (which can be calibrated), then the planning of the polynomial reference line is considered to be too close to the curb, that is, the usability of the polynomial reference line is invalid.

[0175] Thus, this application calculates the minimum distance between the reference line and discrete points on the curb, and dynamically generates a curb-side safety verification score based on preset safe and dangerous distances, reasonably deducting points from reference lines that are close to or may collide with the curb. This mechanism effectively avoids the avoidance trajectory from getting too close to or encroaching on the curb area, reducing the risk of scraping or loss of control, and improving the safety and reliability of path planning in narrow or undefined lane marking scenarios.

[0176] In some embodiments, the cross-solid line verification score of the reference line is determined in the following manner: if the reference line is a reference line that crosses the solid line lane line, the trajectory planning and decision-making device 10 determines the lower limit of the cross-solid line score (e.g., -1) as the roadside safety verification score of the reference line; if the reference line is not a reference line that crosses the solid line lane line, the trajectory planning and decision-making device 10 determines the upper limit of the cross-solid line score (e.g., 0) as the roadside safety verification score of the reference line.

[0177] It is understandable that the above-mentioned reference line crossing solid line verification score can deduct points for reference lines whose trajectory crosses solid lines.

[0178] Refer to the following formula (18) for the cross-solid line verification score of the reference line. The calculation formula is as follows:

[0179] (18)

[0180] In some embodiments, the cross-traffic direction verification score of vulnerable road users (VRUs) on the reference line is determined as follows: the trajectory planning and decision-making device 10 determines the lateral movement direction of vulnerable road users in the surrounding environment of the vehicle; based on the lateral movement direction and the collision time window, a set of positions of vulnerable road users is determined at a set time interval; if the reference line includes positions within the set, the lower limit of the cross-traffic score (e.g., -1) is determined as the cross-traffic direction verification score of vulnerable road users on the reference line; if the reference line does not include positions within the set, the upper limit of the cross-traffic score (e.g., 0) is determined as the cross-traffic direction verification score of vulnerable road users on the reference line.

[0181] It's understandable that the VRU traversing motion direction verification significantly deducts points based on a reference line in the opposite direction of the lateral movement. VRU refers to pedestrians or cyclists. For VRU traversing scenarios with AES functionality, the target speed must not be too high, for example, below 8 kph.

[0182] As an example, for the position of the VRU (Vehicle Ruler) crossing the road within the Time to Collision (TTC), a static position is established at fixed time intervals, taking into account the leftmost and rightmost ends of the position aggregation point. Considering that pedestrian targets in real-world collision scenarios usually actively decelerate or move in the opposite direction to avoid collisions with the main vehicle, the following formula (19) is used to generate a cross-traffic direction verification score for pedestrian targets based on their movement direction. .

[0183] (19)

[0184] Reference Figure 7 The image shown is a schematic diagram of a reference line for traversing a dynamic obstacle, provided in an embodiment of this application. Figure 7 As shown, the VRU moves from right to left, so the avoidance point TR is the collision position when the VRU crosses (i.e., the predicted position corresponding to the collision time between the main vehicle and the crossing target, i.e., the VRU). Therefore, all reference lines involved with the avoidance point TR (i.e., reference lines with endpoints L1 and L2) will be penalized.

[0185] Thus, this application predicts the future position of vulnerable road users in the crossing direction and checks whether the reference line conflicts with them. Reference lines with collision risk are directly assigned low scores, thereby effectively avoiding potential conflicts between pedestrians and vehicles in the planning stage and significantly improving the proactive protection capability and safety of the avoidance trajectory for vulnerable road users such as pedestrians and cyclists.

[0186] S502: The trajectory planning and decision-making device 10 determines the target reference line from at least two reference lines based on the M verification scores of each reference line.

[0187] In some embodiments, the trajectory planning and decision-making device 10 determines a valid reference line from at least two reference lines. The first verification score of the valid reference line is greater than or equal to a corresponding preset threshold. The first verification score includes at least one of the following: obstacle verification score, urgency level verification score, or curbside safety verification score. Based on the M verification scores of each valid reference line, a corresponding total verification score is determined. The reference line with the highest total verification score is determined as the target reference line. In this way, by first screening valid reference lines that meet key safety thresholds (such as obstacle, urgency level, or curbside safety) and then comprehensively selecting the target reference line based on multi-dimensional verification scores, the basic safety of the avoidance path is ensured while also taking into account comfort and rationality, effectively improving the reliability and environmental adaptability of the autonomous driving system's decision-making.

[0188] It is understood that the description of the numerical relationship between the obstacle verification score, the urgency level verification score, or the curbside safety verification score and the corresponding preset threshold can be found in the relevant description in S501, and will not be repeated here.

[0189] In some embodiments, the values ​​of all the above verification scores have been normalized during the design process. Weights can be set for each verification score, and a polynomial reference line summary score can be performed based on these weights. Specific weights can be adjusted in practical applications. The formula for calculating the total verification score is as follows (20):

[0190] (20)

[0191] Among them, in formula (20) Represents a verification score The corresponding weight, the value of which can be determined according to actual needs, is not specifically limited here. Obstacle detection scores can be generated. Urgency level check score Smoothness check score Lane keeping test score Historical decision information verification score Roadside verification score Cross-solid line verification score Or, the score for checking the direction of movement of vulnerable road users crossing the road. .

[0192] Thus, this application comprehensively evaluates the feasibility and merits of each reference line based on multi-dimensional verification scores, thereby accurately determining the target reference line. This solution effectively integrates multi-source information such as environmental perception, high-precision maps, traffic rules, and behavior prediction, ensuring not only the geometric feasibility of the selected reference line but also achieving an optimized balance in terms of safety, compliance, comfort, and behavioral rationality. Especially in complex urban scenarios, it can prioritize avoiding high-risk behaviors such as crossing solid lines, approaching curbs, and encroaching on pedestrian paths, while also considering vehicle dynamics constraints and driving experience, significantly improving the robustness and safety of the autonomous driving system's obstacle avoidance decisions.

[0193] In some embodiments, the reference line generation process of this application includes phased evaluation and screening of candidate reference lines to achieve prediction of avoidance direction and pre-solution of feasible avoidance space.

[0194] Reference Figure 8 The diagram shown is a schematic representation of a reference line generation process framework provided in this application. Figure 8 As shown, the process first performs a preliminary validity assessment of each reference line based on key safety indicators (such as obstacle verification, urgency verification, and curbside safety verification). If any key indicator score is lower than the corresponding preset threshold, the reference line is deemed invalid and eliminated. Only when all the above key indicators meet the threshold requirements is the reference line retained as a valid reference line. Subsequently, for all reference lines that pass the validity verification, a comprehensive evaluation of multiple dimensions of verification scores is conducted, including but not limited to obstacle verification score, urgency score, smoothness score, lane keeping score, historical decision information score, curbside score, solid line crossing score, and vulnerable road user (VUR) traversing direction score. The scores of each sub-item are weighted and summed to obtain the total verification score for each reference line. Finally, the reference line with the highest total verification score is selected as the target reference line for subsequent trajectory planning and vehicle control. This two-stage evaluation process is as follows: Figure 8 As shown: In the first stage, high-risk paths are eliminated through "valid bit verification". In the second stage, the optimal solution is selected from the set of safe and feasible candidates. This optimizes the avoidance strategy and reasonably constrains the avoidance direction while ensuring safety, effectively improving the decision-making efficiency and robustness of the autonomous driving system in complex environments.

[0195] Reference Figure 9 The diagram shown illustrates a solution process for an obstacle avoidance trajectory provided in an embodiment of this application. This process is used to implement a target obstacle avoidance trajectory that satisfies constraints within the trajectory solution boundary using a preset algorithm. Specifically, the execution entity of this process is the trajectory planning and decision-making device 10, and this process is applied to the obstacle avoidance process of controlling the vehicle 100. The process includes the following steps:

[0196] S901: The trajectory planning and decision-making device 10 uses a preset algorithm to construct a state-space model and cost function.

[0197] For example, the default algorithm is illustrated using the CILQR algorithm.

[0198] In some embodiments, the state variables in the state-space model include: vehicle position coordinates, vehicle heading angle, and vehicle front wheel steering angle; the control variable in the state-space model is the rate of change of the vehicle front wheel steering angle.

[0199] As an example, the state-space model in this application can be based on a two-degree-of-freedom kinematic model, with the addition of the front wheel steering angle state, using the rate of change of steering angle as the increment. The state-space equations of the state-space model are given in Equations (21) and (22) below.

[0200] (twenty one)

[0201] (twenty two)

[0202] In formulas (21) and (22), X, Y, θ, and δ are state variables, where X and Y represent the vehicle's position coordinates, θ represents the vehicle's heading angle, and δ represents the front wheel steering angle. γ is the input to the state-space equation, i.e., the control variable, and represents the rate of change of the front wheel steering angle. v represents the vehicle speed, which is constant and is assumed to remain unchanged throughout the process. L represents the wheelbase, which is constant.

[0203] In some embodiments, the cost function described above includes at least one of the following: reference line tracking cost, lateral acceleration cost, driving boundary cost, steering angle boundary cost, or steering angular velocity cost.

[0204] The reference line tracking cost is used to characterize the degree of lateral deviation between each sampling point of the avoidance trajectory and the corresponding position of the target reference line, and the weights corresponding to a predetermined number of key sampling points at the end of the target reference line are greater than the first predetermined weight. For example, the reference line tracking cost can be calculated based on the square of the lateral position difference between each sampling point of the avoidance trajectory and the corresponding position of the target reference line in the y-axis direction.

[0205] Reference Figure 10 The diagram shown is a schematic diagram of a high-weight setting of sampling points based on the CILQR algorithm provided in an embodiment of this application. Figure 10This diagram illustrates the temporal structure of reference line generation and trajectory planning in an autonomous driving system. The vertical direction of the time axis (t0 to tn) represents the time series, from the initial time t0 to the future time tn, representing multiple sampling time points during the vehicle's journey. The reference line, composed of a series of black and white dots, represents the desired path generated by the system to avoid obstacles. Starting at time t0, the reference line extends forward, guiding the vehicle around obstacles ahead. Left and right boundary points (dark gray solid dots) are red dots distributed on both sides of the reference line, representing the left and right boundaries of the vehicle's drivable area defined at different time points, used to constrain the trajectory from exceeding or crossing the boundary. High-weight reference points (black dots) are located at critical positions on the reference line, typically concentrated at the ends (e.g., tk to tn). These points are given higher weight in the lateral direction during trajectory optimization to ensure accurate completion of the final avoidance maneuver, i.e., precise endpoint tracking. Low-weight reference points (white dots) are located in the middle section of the reference line. These points serve only as trend guides during optimization, with lower weights, allowing the trajectory to be flexibly adjusted while maintaining the overall direction, avoiding excessive constraints that could lead to uneven motion. It is understandable that the target avoidance trajectory represents the avoidance trajectory generated by the vehicle after optimization. In the early stage, it is close to the reference line, and as it approaches the endpoint, it gradually converges to the high-weight reference point to achieve a smooth and safe detour. Figure 10 The two gray rectangles represent static or dynamic obstacles that need to be avoided. The vehicle achieves effective avoidance through reference lines and trajectory planning. In this way, the segmented weight design strategy of the reference lines in the time dimension, with low weight for the front sampling points and high weight for the end sampling points, not only uses the reference lines to guide the avoidance direction, but also balances path tracking accuracy and motion smoothness through weight allocation, thus achieving efficient, safe and comfortable avoidance behavior.

[0206] It is understandable that the larger the square of the lateral position difference, the higher the corresponding reference line tracking cost, which characterizes the degree of lateral deviation between the avoidance trajectory and the reference line in the y-axis direction, i.e., the tracking accuracy of the avoidance trajectory to the reference line. Furthermore, this scheme does not apply the same weight to all sampling points, but only to specific key sampling points at the end of the reference line (e.g., Figure 10 The two or more black sampling points shown are given significantly higher weights. In this way, the reference line represents the future driving trend direction without being affected by boundary or lateral acceleration.

[0207] In some embodiments, the lateral acceleration cost is used to characterize the lateral acceleration corresponding to each sampling point in the obstacle avoidance trajectory. It is understood that the lateral acceleration cost is constructed based on the lateral acceleration values ​​calculated at each sampling point of the obstacle avoidance trajectory. Since excessive lateral acceleration may exceed the vehicle's adhesion limits or cause occupant discomfort, it is considered a characteristic of trajectory hazard. Therefore, this cost is adjusted during optimization by setting appropriate weights. For example, the higher the corresponding weight, the more the optimization algorithm, such as a preset algorithm, tends to generate a trajectory with lower lateral acceleration and a smoother, safer path. Thus, by introducing an adjustable-weighted lateral acceleration cost term, lateral dynamic risks are explicitly constrained in trajectory optimization, effectively suppressing aggressive or unstable avoidance behavior, and significantly improving driving safety and ride comfort while ensuring obstacle avoidance effectiveness.

[0208] In some embodiments, the driving boundary cost is used to characterize the lateral distance relationship between each sampling point in the avoidance trajectory and the left and right driving boundaries.

[0209] like Figure 11 The diagram shows the complete process and key constraints of an autonomous driving system when it faces obstacles (such as vehicles or vulnerable road users, VRUs) in front of it. Figure 11 The display shows a two-lane road with the main vehicle in the left lane. A target obstacle and a potentially vulnerable road user (VRU) crossing the road are directly ahead. The 4-second time distance corresponding to the main vehicle's current speed is marked as a dashed area, representing the system's predicted collision risk range within the next 4 seconds. Point E1 represents the endpoint of the avoidance trajectory, located within a safe zone, used to constrain the final path convergence. A reference line (gray dashed curve) is generated by the algorithm, serving as a guide line for the avoidance direction and reflecting the detour trend. The generated trajectory boundary (thick gray solid line) is a feasible space constructed based on road boundaries, obstacles, and dynamic predictions, serving as a physical constraint for trajectory optimization. The CILQR algorithm outputs the planned avoidance path (thin gray curve), the optimal avoidance trajectory solved by the CILQR algorithm under the reference line and boundary constraints. Starting from point H, it bypasses the target and avoids the VRU, ultimately converging to point E1.

[0210] The predicted location corresponding to the Time-to-Collision (TTC) between the vehicle and the target obstacle indicates that the avoidance maneuver must be completed before the collision occurs. Thus, by combining environmental perception, TTC prediction, reference line guidance, constraint boundary construction, and a pre-defined algorithm (CILQR), a safe and compliant avoidance trajectory is generated. By setting multiple constraints at key points (H, L1 / 2, R1 / 2), the system can achieve precise, smooth, and safe active avoidance behavior in complex dynamic scenarios.

[0211] It is understandable that during the optimization algorithm construction phase, the system dynamically generates feasible driving boundaries based on roadside locations, nearby obstacles, and other environmental constraints. When a trajectory point approaches or crosses this boundary, the corresponding boundary cost increases significantly. By introducing this term into the cost function, the optimization process is guided to maintain the trajectory within a safe and legal driving channel.

[0212] In some embodiments, the steering angle boundary cost characterizes the deviation between the steering angle corresponding to each sampling point in the avoidance trajectory and the steering angle boundary angle. It can be understood that during the optimization algorithm construction phase, the system sets the maximum permissible steering angle as a steering angle boundary limit based on the vehicle's physical characteristics; when the planned steering angle at a sampling point approaches or exceeds this boundary, its corresponding steering angle cost increases accordingly. By introducing this term into the cost function, the optimization process is guided to maintain the steering angle of each sampling point within the vehicle's actually executable steering angle range, thereby ensuring that the generated avoidance trajectory is dynamically feasible and conforms to the physical constraints of the vehicle's actuators.

[0213] In some embodiments, the steering angular velocity cost is used to characterize the steering angular velocity corresponding to each sampling point in the avoidance trajectory. It can be understood that the steering angular velocity cost is used to constrain the rate of change of the steering angle in the planned trajectory, thereby suppressing drastic fluctuations in the steering angle, improving the continuity and smoothness of the trajectory, and enhancing ride comfort. By adjusting the weight of this item in the trajectory optimization algorithm, its influence on the overall trajectory shape can be adjusted. For example, a higher corresponding weight results in a more gradual change in the steering of the generated trajectory and a more stable dynamic behavior.

[0214] Thus, by setting a multi-dimensional cost function that includes reference line tracking cost, lateral acceleration cost, driving boundary cost, steering angle boundary cost, and steering angular velocity cost, the avoidance trajectory can accurately follow the target reference line while taking into account comfort, vehicle dynamics limitations, and road boundary constraints, effectively improving the safety, smoothness, and feasibility of the trajectory.

[0215] S902: The trajectory planning and decision-making device 10 performs reverse calculation of the avoidance trajectory in the current iteration process based on the cost function to obtain the control gain of each sampling point. The control gain is related to the control quantity of the state space model.

[0216] In some embodiments, this application calculates the optimal feedback control gain for each sampling point by recursively calculating the control gain from the end point to the starting point of the avoidance trajectory in time. The control gain is directly related to control variables in the state-space model, such as the front wheel steering angle change rate. Thus, by solving in reverse using the avoidance trajectory, a locally optimal control strategy can be provided for generating a better trajectory in the next step.

[0217] S903: The trajectory planning and decision-making device 10 updates the avoidance trajectory in the current iteration process based on the control gain of each sampling point, and obtains the state and control variables corresponding to each sampling point in the updated state space model.

[0218] In some embodiments, by using the control gain obtained from the inverse solution, combined with the current state variables, and integrating along the time axis, the state variables and control variables can be updated to form new candidate avoidance trajectories, thereby gradually approaching the optimal solution.

[0219] S904: In this iteration, the trajectory planning and decision-making device 10 determines whether the convergence condition is met based on the change in the penalty value corresponding to the cost function.

[0220] If the condition is met, then proceed to S905; otherwise, re-enter S902.

[0221] As an example, convergence conditions include: the number of iterations reaches the maximum number of iterations, the cost difference decreases by a greater than a set percentage, and the absolute value of the cost difference is greater than a set absolute value.

[0222] The above convergence conditions are convergence judgment conditions, also known as inner loop iteration conditions. Reaching the maximum number of iterations prevents infinite loops. A sufficiently large decrease in the cost difference (greater than a set percentage) indicates that trajectory optimization is still effective. An absolute value greater than a set absolute value indicates that the absolute change in cost is still significant, preventing ineffective iterations in plateau regions. Thus, by using convergence judgment conditions, we can ensure that the preset algorithm stops within a reasonable time and guarantee that each iteration has a practical effect.

[0223] It can be understood that S901 to S904 is the inner loop iterative process for solving the avoidance trajectory.

[0224] As an example, during the inner loop iteration process, the weights corresponding to each cost item (including the steering angular velocity cost) are fixed values ​​and do not change with the increase of the number of iterations, so as to ensure the stability and convergence of the optimization process and avoid unnecessary trajectory oscillations or decision drifts caused by dynamic adjustment of weights.

[0225] It is understood that this application embodiment constructs a comprehensive objective function by weighted summation of multiple key performance indicators at all planned sampling points. This objective function guides a preset algorithm (such as QP or CILQR) to generate a safe and comfortable avoidance trajectory. Here, the reference line tracking cost is the primary control objective, ensuring the vehicle completes the avoidance maneuver according to the desired detour direction and endpoint. Lateral acceleration cost ensures driving safety and occupant comfort, preventing loss of control or discomfort due to sharp bends or abrupt lane changes. Driving boundary cost is used to prevent collisions and maintain the vehicle within the legally feasible area. Steering angle boundary cost constrains the steering angle at each sampling point to not exceed the vehicle's physical limits (such as maximum turning angle), ensuring the planning result matches the actuator's capabilities and avoiding the generation of unachievable commands. Steering angular velocity cost penalizes excessively rapid steering angle changes, i.e., limiting the steering wheel rotation rate to improve control smoothness and driving naturalness, and reduce execution jitter.

[0226] S905: The trajectory planning and decision-making device 10 determines whether the avoidance trajectory in the current iteration process meets the feasibility constraints if the convergence conditions are met.

[0227] Among them, the feasibility constraints are related to the vehicle's maximum steering angle and maximum lateral acceleration.

[0228] If the determination is yes, that is, the avoidance trajectory in this iteration process meets the feasibility constraint conditions, then proceed to S906; otherwise, proceed to S907.

[0229] In some embodiments, feasibility constraints serve as the feasibility judgment conditions for the avoidance trajectory. After meeting the convergence condition, the current avoidance trajectory is further checked to see if it meets the feasibility constraints of physical and safety constraints (such as not exceeding limits, not exceeding the steering angle, and lateral acceleration within the adhesion limits). Whether the feasibility constraints are met is indirectly or directly reflected by the value of the penalty term in the cost function; for example, a penalty value lower than the maximum steering angle or maximum lateral acceleration is considered satisfied. Thus, the feasibility of the avoidance trajectory is determined by distinguishing the constraints, ensuring that the output trajectory is executable and safe.

[0230] S906: The trajectory planning and decision-making device 10 determines the avoidance trajectory in the current iteration process as the target avoidance trajectory, provided that the feasibility constraints are met.

[0231] S907: When the penalty value of the cost function does not meet the feasibility constraints, the trajectory planning and decision-making device 10 adds a cost augmentation penalty term to the cost function.

[0232] Among them, the cost augmentation penalty term is related to the maximum steering angle constraint and / or the maximum lateral acceleration constraint.

[0233] S908: The trajectory planning and decision-making device 10 adjusts the penalty value of the cost function according to the cost augmentation penalty term until the penalty value of the cost function meets the feasibility constraint conditions, and takes the avoidance trajectory in the current iteration as the target avoidance trajectory.

[0234] As an example, S907 and S908 above describe the outer loop iteration process. It can be understood that an augmented constraint design is implemented for the outer loop iteration, specifically including a maximum steering angle constraint for physical limitations and a maximum lateral acceleration constraint for safety limitations. That is, a cost augmentation penalty term is added to the cost function. When the inner loop iteration ends, if the defined feasibility constraints are not met, the penalty coefficient for that portion of the cost is adjusted during cost calculation with each iteration of the outer loop, significantly increasing the cost of violating the constraints and improving their effectiveness.

[0235] It is understandable that even if the initial trajectory severely violates constraints (such as exceeding limits on sharp curves), the outer loop process in this scheme can gradually return to the feasible region through multiple rounds of augmented penalty guidance system, avoiding optimization collapse. Furthermore, all constraints (maximum steering angle, maximum lateral acceleration) are derived from vehicle physics characteristics, and the penalty coefficients can be calibrated, making them applicable to different vehicle models and road conditions.

[0236] S909: Trajectory planning and decision-making device 10 performs reverse verification of target avoidance trajectory.

[0237] In some embodiments, the aforementioned anti-verification includes at least one of the following: dynamic feasibility verification, collision detection verification, lane compliance verification, and curb safety distance verification. As an example, dynamic feasibility verification verifies whether the avoidance trajectory meets the vehicle's steering or acceleration requirements. Collision detection verification checks whether the trajectory conflicts with dynamic or static obstacles in time and space. Lane compliance verification determines whether the trajectory illegally crosses a solid line or enters a restricted area. Curb safety distance verification ensures that the trajectory maintains sufficient lateral distance from the curb. Thus, the final output target avoidance trajectory not only avoids obstacles but also meets multi-dimensional requirements such as dynamic constraints, road boundaries, and traffic rules.

[0238] Thus, traditional methods that directly apply hard constraints are prone to failure, while this scheme approximates hard constraints asymptotically through soft penalties, preserving optimization flexibility while ultimately ensuring safety. Furthermore, the inner loop converges quickly, and the outer loop typically requires only a few iterations (e.g., 2 to 5), keeping overall computational overhead manageable and meeting the real-time requirements of autonomous driving. Consequently, the final output target avoidance trajectory not only avoids obstacles but also satisfies multi-dimensional requirements such as dynamic constraints, road boundaries, and traffic rules.

[0239] like Figure 12As shown, the trajectory optimization process based on the CILQR algorithm is illustrated. It consists of two hierarchical structures: an outer loop and an inner loop, used to solve for the optimal avoidance trajectory that satisfies multiple constraints. The outer loop, located at the outermost layer, is primarily responsible for handling system constraints and updating relevant parameters. In each iteration, based on the current trajectory and environmental information (such as obstacles and road boundaries), points that do not meet the constraints are corrected or penalized, dynamically adjusting the constraint parameters in the cost function. After updating, the constraint parameters and other relevant information in the cost function are passed to the inner loop to guide the next round of optimization. The outer loop, by continuously adjusting constraints and parameters, gradually approaches a feasible and optimal solution, improving the robustness and convergence of the overall optimization. Figure 12 As shown, the inner loop uses an iterative approach to solve the linearized optimal control problem, comprising the following four key steps: (1) updating derivative information, (2) backward pass, forward pass, and convergence check. In the process of updating derivative information, the Jacobian and Hessian matrices of the cost function with respect to the state and control variables are calculated to provide gradient information for subsequent solutions. In the backward pass, based on the current cost function and the system dynamics model, a backward recursion is performed from the endpoint to the starting point to solve for the control gain (i.e., the optimal feedback control law) at each sampling point, used to generate the local optimal control input. In the forward pass, using the control gain obtained in the previous step, combined with the current state, the process proceeds forward along the time axis, updating the state trajectory and generating new candidate trajectories. In the convergence check, the cost change and state / control variable changes between the current iteration and the previous iteration are checked; if the convergence condition is met (e.g., the cost decrease is less than a threshold), the inner loop ends; otherwise, the process of updating derivative information is returned to continue iterating.

[0240] It is understandable that the above inner loop process iterates multiple times under fixed constraints (i.e. convergence conditions) until the trajectory converges; when the inner loop converges, the outer loop determines whether the global constraints (i.e. feasibility constraints) are met; if not, the constraints are reprocessed and the next round of the outer loop is entered; if met, the final avoidance trajectory is output and the process ends.

[0241] Thus, based on pre-defined algorithms such as CILQR, a two-layer iterative structure in autonomous driving trajectory planning achieves rapid convergence of local optimal control through the inner loop and gradually satisfies global constraints through the outer loop. This effectively balances computational efficiency and constraint feasibility, making it suitable for generating high-precision, safe, and reliable obstacle avoidance paths in complex dynamic environments.

[0242] It is understandable that the above Figure 3 , Figure 5 and Figure 9 The order of the steps is only one example, and other orders may be used in other embodiments. This application does not make specific limitations on this.

[0243] The hardware structure of the vehicle used in the embodiments of this application will be described next.

[0244] As an example, Figure 13 This is a schematic diagram of a possible functional framework for a vehicle 10 provided in an embodiment of this application. For example... Figure 13 As shown, the functional framework of vehicle 100 may include various subsystems, such as the sensor system 101, control system 102, one or more peripheral devices 103 (one is shown as an example), power supply 104, and computer system 105. Optionally, vehicle 100 may also include other functional systems, such as an engine system that provides power to vehicle 100, etc., which are not limited herein.

[0245] The sensor system 101 may include several detection devices that can sense the measured information and convert the sensed information into electrical signals or other desired forms of information output according to a certain rule. For example... Figure 13 As shown, these detection devices may include a global positioning system (GPS) 1011, a vehicle speed sensor 1012, an inertial measurement unit (IMU) 1013, a radar unit 1014, a laser rangefinder 1015, a camera unit 1016, a wheel speed sensor 1017, a steering sensor 1018, a gear position sensor 1019, or other components for automatic detection, etc., which are not limited in this application.

[0246] The Global Positioning System (GPS) 1011 is a system that uses GPS positioning satellites to perform real-time positioning and navigation globally. In this application, the GPS 1011 can be used to achieve real-time positioning of the vehicle 100, providing the vehicle 100's geographical location information. The vehicle speed sensor 1012 is used to detect the vehicle 100's speed. The inertial measurement unit 1013 may include a combination of an accelerometer and a gyroscope, and is a device for measuring the angular rate and acceleration of the vehicle 100. For example, during the vehicle 100's movement, the inertial measurement unit can measure the vehicle's position and angular changes based on the vehicle 100's inertial acceleration, such as measuring the vehicle 100's acceleration and angular rate.

[0247] Radar unit 1014, also known as a radar system, senses objects in the current environment where vehicle 100 is traveling using wireless signals. Optionally, the radar unit can also sense information such as the object's speed and direction of travel. In practical applications, the radar unit can be configured as one or more antennas for receiving or transmitting wireless signals. Laser rangefinder 1015 is an instrument that uses modulated laser light to measure the distance to a target object; that is, a laser rangefinder can be used to measure the distance to a target object. In practical applications, the laser rangefinder may include, but is not limited to, a combination of one or more of the following components: a laser source, a laser scanner, and a laser detector.

[0248] The camera unit 1016 is used to capture images, such as pictures and videos. In this application, during the movement of the vehicle 100 or after the camera device is activated, the camera device can acquire images of the environment in which the vehicle 100 is located in real time. For example, the camera device can capture images of the road surface where the vehicle 100 is located to detect the degree of friction between the road surface and the vehicle. As another example, during the process of the vehicle 100 entering or exiting a tunnel, the camera device can acquire corresponding images in real time and continuously. In practical applications, the camera device includes, but is not limited to, a dashcam, a camera, a camera, or other components used for taking pictures / photographs, and the number of such camera devices is not limited in this application.

[0249] Wheel speed sensor 1017 is a sensor used to detect the rotational speed of the wheels of vehicle 100, such as detecting the wheel speed of the inner rear wheel of vehicle 100. Commonly used wheel speed sensors 1017 may include, but are not limited to, magnetoelectric wheel speed sensors and Hall effect wheel speed sensors. Steering sensor 1018, also known as a steering angle sensor, can represent a system used to detect the steering angle of vehicle 100. In practical applications, steering sensor 1018 can be used to measure the steering wheel angle of vehicle 100 (e.g., 360 degrees), or to measure an electrical signal representing the steering angle of vehicle 100's steering wheel. Optionally, steering sensor 1018 can also be used to measure the steering angle of vehicle 100's tires (e.g., 20 degrees), or to measure an electrical signal representing the steering angle of vehicle 100's tires, etc., and this application is not limited thereto.

[0250] That is, the steering sensor 1018 can be used to measure any one or more of the following combinations: the steering angle of the steering wheel, an electrical signal representing the steering angle of the steering wheel, the steering angle of the wheel (the tire of vehicle 100), and an electrical signal representing the steering angle of the wheel.

[0251] The gear position sensor 1019 is used to detect the current gear of vehicle 100. Since different manufacturers produce vehicle 100, the gear positions may vary. Taking an autonomous vehicle 100 as an example, it supports six gears: P, R, N, D, 2, and L. P (parking) is used for parking; it uses a mechanical device to lock the brakes, preventing the vehicle from moving. R (reverse) is used for reversing. D (drive) is used for driving on the road. 2 (second gear) is also a drive gear, used to adjust the vehicle's speed. Second gear is typically used for going up or down slopes. L (low) limits the vehicle's speed. For example, on a downhill road, if vehicle 100 is put into L gear, the vehicle 100 will use engine power for braking while going downhill, so the driver does not have to keep the brake pedal pressed for a long time, which could cause the brake pads to overheat and cause danger.

[0252] The control system 102 may include several components, such as the steering unit 1021, braking unit 1022, lighting system 1023, automatic driving system 1024, map navigation system 1025, network time synchronization system 1026, obstacle avoidance system 1027, and cornering assistance system 1028 shown in the figure. Optionally, the control system 102 may also include components such as a throttle controller and an engine controller for controlling the driving speed of the vehicle 100, which is not limited in this application. The driver in this application may be the aforementioned engine controller. The user's acceleration operation on the vehicle 100 may be the user's operation of pressing the throttle controller (i.e., the accelerator pedal).

[0253] The steering unit 1021 may represent a system for adjusting the direction of travel of the vehicle 100, which may include, but is not limited to, a steering wheel or other structural device for adjusting or controlling the direction of travel of the vehicle 100.

[0254] Braking unit 1022 may represent a system for slowing down the speed of vehicle 100, or may be referred to as vehicle 100 braking system.

[0255] The lighting system 1023 is used to provide lighting or warning functions for the vehicle 100.

[0256] The autonomous driving system 1024 may include hardware and software systems for processing and analyzing data input to the autonomous driving system 1024 to obtain actual control parameters of various components in the control system 102, such as the desired braking pressure of the brake controller in the braking unit and the desired torque of the engine. This facilitates the control system 102 in implementing corresponding control and ensures the safe driving of the vehicle 100. Optionally, the autonomous driving system 1024 can also determine information such as obstacles faced by the vehicle 100 and the characteristics of the environment in which the vehicle 100 is located (e.g., the lane the vehicle 100 is currently traveling in, road boundaries, and traffic lights that it is about to pass) by analyzing the data. The data input to the autonomous driving system 1024 may be image data collected by a camera device or data collected by various components in the sensor system 101, such as the steering wheel angle provided by the steering angle sensor and the wheel speed provided by the wheel speed sensor. This application does not impose any limitations on these parameters.

[0257] The map navigation system 1025 is used to provide map information and navigation services for vehicle 100.

[0258] The network time system 1026 (NTS) is used to provide time synchronization services to ensure that the current system time of vehicle 100 is synchronized with the network standard time, which is beneficial to providing vehicle 100 with more accurate time information.

[0259] The obstacle avoidance system 1027 is used to predict obstacles that the vehicle 100 may encounter during driving, and then control the vehicle 100 to bypass or cross the obstacles in order to achieve normal driving of the vehicle 100.

[0260] The cornering assistance system 1028 may include hardware and software systems for processing and analyzing data input to the cornering assistance system 1028 to obtain actual control parameters of various components in the control system 102, such as the desired braking force of the brake controller (e.g., brake) in the braking unit and the desired torque of the engine. This facilitates the control system 102 in implementing corresponding control, ensuring that the vehicle 100 steers based on the AES function.

[0261] In some embodiments, the trajectory planning and decision-making device 10 of this application is implemented based on the control system 102, specifically based on the autonomous driving system 1024 in the control system 102, or based on the cornering assistance system 1028 in the control system 102, but not limited thereto. In this case, the control system 102, the autonomous driving system 1024, or the cornering assistance system 1028 can execute the trajectory planning and decision-making method described above.

[0262] Peripheral device 1016 may include several components, such as communication system 1031, touch screen 1032, user interface 1033, microphone 1034, and speaker 1035 as shown in the figure.

[0263] Several functions of the vehicle 100 are controlled and implemented by the computer system 105. The computer system 105 may include one or more processors 1051 (the figure shows one processor as an example) and a memory 1052 (also referred to as a storage device). In practical applications, the memory 52 may be located inside the computer system 105 or outside the computer system 105, for example, as a cache in the vehicle 100, etc., which is not limited in this application.

[0264] Processor 1051 may include one or more general-purpose processors, such as a graphics processing unit (GPU). Processor 1051 can be used to execute relevant programs or corresponding instructions stored in memory 1052 to implement the corresponding functions of vehicle 100. Memory 52 can be used to store a set of program code or corresponding instructions, so that processor 51 can call the program code or instructions stored in memory 52 to implement the corresponding functions of vehicle 100. This function includes, but is not limited to, […]. Figure 10 The schematic diagram of the functional framework of the vehicle 100 shown includes some or all of the functions. In this application, the memory 1052 can store a set of program code for controlling the vehicle 100, and the processor 1051 can call the program code to control the safe driving of the vehicle 100.

[0265] Optionally, in addition to storing program code or instructions, the memory 1052 may also store information such as road maps, driving routes, and sensor data. The computer system 105 can be combined with other components in the functional framework diagram of the vehicle 100, such as sensors in the sensor system and GPS, to realize the relevant functions of the vehicle 100. For example, the computer system 1052 can control the driving direction or speed of the vehicle 100 based on the data input from the sensor system 101; this application does not impose limitations on this.

[0266] Among them, this application Figure 13 The sensor system 101, control system 102, and computer system 105 shown are merely examples and do not constitute a limitation. In practical applications, vehicle 100 can combine several components according to different functions to obtain subsystems with corresponding functions. For example, vehicle 100 may also include an electronic stability program (ESP) and an electric power steering system (EPS), etc. Figure 13Not shown. The ESP system may consist of some sensors in sensor system 101 and some components in control system 102. Specifically, the ESP system may include wheel speed sensor 1017, steering sensor 1018, lateral acceleration sensor, and control units involved in control system 102. The EPS system may consist of some sensors in sensor system 101, some components in control system 102, and power supply 104. Specifically, the EPS system may include steering sensor 1018, generator and reducer involved in control system 102, battery power supply, etc.

[0267] It should be noted that the above Figure 13 This is merely a schematic diagram of one possible functional framework for vehicle 100. In practical applications, vehicle 100 may include more or fewer systems or components, and this application does not impose any limitations.

[0268] Various embodiments of the mechanisms disclosed in this application can be implemented in hardware, software, firmware, or combinations of these implementation methods. Embodiments of this application can be implemented as computer programs or program code executable on a programmable system, the programmable system including at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0269] Program code can be applied to input instructions to execute the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, the processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application-specific integrated circuit (ASIC), or a microprocessor.

[0270] The program code can be implemented using a high-level procedural language or an object-oriented programming language to communicate with the processing system. Assembly language or machine language can also be used when needed. In fact, the mechanisms described in this application are not limited to any particular programming language. In either case, the language can be a compiled language or an interpreted language.

[0271] In some embodiments, this application provides a readable medium storing instructions that, when executed on an electronic device (such as a trajectory planning and decision-making device), cause the electronic device to perform the trajectory planning and decision-making method described above.

[0272] In some embodiments, this application provides an electronic device (such as a trajectory planning and decision-making device), comprising: a memory for storing instructions executed by one or more processors of the electronic device, and a processor, one of the processors of the electronic device, for executing the trajectory planning and decision-making method described above.

[0273] In some embodiments, this application provides a computer program product including instructions for implementing the trajectory planning and decision-making method described above.

[0274] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored thereon on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, the instructions may be distributed via a network or through other computer-readable media. Therefore, machine-readable media may include any mechanism for storing or transmitting information in a machine-readable (e.g., computer-readable) form, including but not limited to floppy disks, optical disks, CD-ROMs, magneto-optical disks, read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic cards or optical cards, flash memory, or tangible machine-readable storage for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) using the Internet in the form of electrical, optical, acoustic, or other propagation signals. Therefore, machine-readable media include any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a machine-readable (e.g., computer-readable) form.

[0275] In the accompanying drawings, some structural or methodological features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. Rather, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Furthermore, the inclusion of structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0276] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0277] It should be noted that in the examples and description of this patent, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0278] Although this application has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A trajectory planning and decision-making method, characterized in that, Applications in trajectory planning and decision-making equipment include: During the vehicle's movement, a target that the vehicle was trying to avoid was detected; A target reference line is determined for the vehicle to detour around the avoidance target. The target reference line is used to indicate the detour direction and endpoint of the vehicle to detour around the avoidance target. The target reference line is one of at least two reference lines, and there are two reference lines with different detour directions among the at least two reference lines. Based on the target reference line and the obstacle and road information of the surrounding environment of the vehicle, a target avoidance trajectory that meets the set constraints is determined. The vehicle is steered to avoid the target based on the target avoidance trajectory.

2. The method according to claim 1, characterized in that, Determining the target reference line for the vehicle to circumvent the avoidance target includes: The vehicle is provided with at least two reference lines, which include at least one or more of the following: a reference line for detouring to the left within the current lane, a reference line for detouring to the left across a lane, a reference line for detouring to the right within the current lane, and a reference line for detouring to the right across a lane; wherein the endpoint of each of the at least two reference lines is spaced a first distance from the starting point of the vehicle in a first direction, and the endpoint of each reference line is spaced a second distance from the lane line associated with the detouring direction in a second direction; The target reference line is determined from the at least two reference lines.

3. The method according to claim 2, characterized in that, Determining the target reference line from the at least two reference lines includes: Generate M verification scores for each of the at least two reference lines; The target reference line is determined from the at least two reference lines based on the M verification scores of each reference line. The M verification scores include at least one of the following: obstacle verification score, urgency verification score, smoothness verification score, lane keeping verification score, historical decision consistency verification score, roadside safety verification score, crossing solid line verification score, and vulnerable road user traversing direction verification score.

4. The method according to claim 3, characterized in that, The step of determining the target reference line from at least two reference lines based on M verification scores of each reference line includes: A valid reference line is determined from the at least two reference lines. The first verification score of the valid reference line is greater than or equal to the corresponding preset threshold. The first verification score includes at least one of the following: obstacle verification score, urgency level verification score, or curbside safety verification score. The corresponding total verification score is determined based on the M verification scores of each of the valid reference lines. The reference line with the highest total score is determined as the target reference line.

5. The method according to any one of claims 1 to 4, characterized in that, The reference line is fitted with a set of polynomials, wherein the polynomials include at least one of the following: a polynomial for constraining the position of the starting point of the vehicle, a polynomial for constraining the position of the avoidance point of the vehicle, a polynomial for constraining the heading of the vehicle at the starting point, a polynomial for constraining the heading of the vehicle at the ending point, or a polynomial for constraining the curvature of the vehicle at the starting point. The location of the vehicle's avoidance point is associated with the location of the avoidance target and the detour direction, and the heading of the vehicle's destination is associated with the lane line associated with the detour direction.

6. The method according to claim 1, characterized in that, The step of determining the target avoidance trajectory that meets the set constraints based on the target reference line and the obstacle and road information of the vehicle's surrounding environment includes: Based on the target reference line and the obstacle and road information of the surrounding environment of the vehicle, the trajectory solution boundary is constructed; The target avoidance trajectory is obtained by using a preset algorithm and satisfying the set constraints within the trajectory solution boundary.

7. The method according to claim 6, characterized in that, The trajectory solution boundary includes at least one of the following: The left and right driving boundaries of the vehicle; The distance between the vehicle's avoidance trajectory and the avoidance point of the avoidance target is less than or equal to the minimum lateral distance; The vehicle's avoidance trajectory does not include the location of vulnerable road users in the set of positions along the lateral movement direction; The vehicle's avoidance trajectory does not include the location of the solid lane line.

8. The method according to claim 6 or 7, characterized in that, The set constraints include convergence conditions and feasibility constraints, and the target avoidance trajectory that satisfies the set constraints within the trajectory solution boundary using a preset algorithm includes: The state-space model and cost function are constructed using the preset algorithm; The avoidance trajectory in this iteration is solved in reverse according to the cost function to obtain the control gain of each sampling point. The control gain is related to the control quantity of the state space model. Based on the control gain of each sampling point, the avoidance trajectory in this iteration process is positively updated to obtain the state variables and control variables corresponding to each sampling point in the updated state space model. During this iteration, the convergence condition is determined based on the change in the penalty value corresponding to the cost function. The convergence condition includes: the number of iterations reaches the maximum number of iterations, the cost difference decreases by a greater than a set percentage, and the absolute value of the cost difference is greater than a set absolute value. If the convergence condition is met, it is determined whether the avoidance trajectory in this iteration process meets the feasibility constraint condition, which is related to the maximum steering angle and maximum lateral acceleration of the vehicle. If the feasibility constraints are met, the avoidance trajectory in this iteration process is determined as the target avoidance trajectory.

9. The method according to claim 8, characterized in that, The state variables in the state-space model include: vehicle position coordinates, vehicle heading angle, and vehicle front wheel steering angle; The control variable in the state-space model is the rate of change of the front wheel steering angle of the vehicle.

10. The method according to claim 8, characterized in that, The cost function includes at least one of the following: Reference line tracking cost, lateral acceleration cost, driving boundary cost, steering angle boundary cost, or steering angular velocity cost; The reference line tracking cost is used to characterize the degree of lateral deviation between each sampling point of the avoidance trajectory and the corresponding position of the target reference line, and the weight of a set number of key sampling points at the end of the target reference line is greater than the first set weight. The lateral acceleration cost is used to characterize the lateral acceleration corresponding to each sampling point in the avoidance trajectory; The driving boundary cost is used to characterize the lateral distance relationship between each sampling point in the avoidance trajectory and the left and right driving boundaries; The steering angle boundary cost is used to characterize the degree of deviation between the steering angle corresponding to each sampling point in the avoidance trajectory and the steering angle boundary angle. The steering angular velocity cost is used to characterize the steering angular velocity corresponding to each sampling point in the avoidance trajectory.

11. The method according to claim 10, characterized in that, The method further includes: If the penalty value of the cost function does not meet the feasibility constraint, a cost augmentation penalty term is added to the cost function, wherein the cost augmentation penalty term is related to the maximum steering angle constraint and / or the maximum lateral acceleration constraint. The penalty value of the cost function is adjusted according to the cost augmentation penalty term until the penalty value of the cost function satisfies the feasibility constraint condition, and the avoidance trajectory in the current iteration number is taken as the target avoidance trajectory.

12. The method according to any one of claims 6 to 11, characterized in that, The method further includes: The target avoidance trajectory is reverse-verified, and the reverse verification includes at least one of the following: dynamic feasibility verification, collision detection verification, lane compliance verification, and roadside safety distance verification.

13. A trajectory planning and decision-making device, characterized in that, include: A memory for storing instructions executed by one or more processors of the trajectory planning and decision-making device, and a processor, one of the processors of the trajectory planning and decision-making device, for performing the method of any one of claims 1 to 12.

14. A vehicle, characterized in that, Includes the trajectory planning and decision-making device as described in claim 13.

15. A readable medium, characterized in that, The readable medium stores instructions that, when executed on the trajectory planning and decision-making device, cause the trajectory planning and decision-making device to perform the method of any one of claims 1 to 12.

16. A computer program product, characterized in that, The computer program product includes: computer program code that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 12.

Images