Special vehicle trajectory planning method considering perception uncertainty

Abstract

The invention particularly relates to a special vehicle trajectory planning method considering perceptual uncertainty, and the method comprises the steps: obtaining the position information and uncertainty of an obstacle in an environment based on environment three-dimensional point cloud data, calculating the position uncertainty based on a preset motion model, obtaining a position uncertainty reference trajectory of the obstacle, and obtaining the position uncertainty reference trajectory of the obstacle; determining a collision probability based on the collision probability, and constructing a risk assessment model based on the collision probability according to a preset progressive decrease type dynamic confidence distribution strategy and a preset collision risk assessment condition; and sampling the at least one determined target state point by using a preset coordinate system to obtain a candidate trajectory set, constructing a chance constraint based on the collision probability, screening the candidate trajectory set based on a deterministic boundary constraint obtained by the chance constraint to obtain a plurality of trajectories, and selecting an optimal trajectory from the plurality of trajectories based on the risk assessment model. Therefore, the problems of sensing uncertainty and the like of the special vehicle caused by the environment are solved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a trajectory planning method for special vehicles that takes into account perceptual uncertainty. Background Technology

[0002] Autonomous driving technology, as a popular research direction in modern transportation, is receiving widespread attention and application globally, and shows broad application prospects in the field of special vehicles. Special vehicles are widely used in military, firefighting, medical, and rescue fields, and their missions require highly specialized and stable performance. The driving behavior and needs of special vehicles differ significantly from those of ordinary passenger vehicles. When performing tasks, they must consider special working environments and mission requirements. Therefore, how to develop effective trajectory planning algorithms that take into account the strong perceptual uncertainties in the environment and meet the special needs of special vehicles is a current research focus.

[0003] In related technologies, method (1) considers the uncertainty of the predicted trajectory of other vehicles and constructs POMDP (Partially Observable Markov Decision Process) based on the Markov assumption to solve the decision state point set of the vehicle and decouple it into a horizontal spatial decision set and a vertical temporal decision set. In the two sets, the drivable area, reference path, and reference speed curve are solved respectively to plan the optimal trajectory. Method (2) considers the probability distribution of the initial state of the vehicle and constructs a differentiable probability grid map by combining the prediction of the trajectory of surrounding vehicles. The collision probability is quantified by inferring the density change of the vehicle state along the reference trajectory through a data-driven density function deep neural network. Finally, the trajectory is solved by methods such as incremental optimization and incremental optimization.

[0004] However, method (1) is incomplete in its uncertainty transmission process, failing to consider the input and propagation of uncertainties caused by sensor errors and perception algorithms in the perception stage. It constructs a POMDP model based on a strong Markov assumption, which not only contradicts the actual decision-making process but also results in an excessively high-dimensional set of decision state points. This model can only find feasible solutions within a finite time for simple lane-changing decisions on structured roads, and is unsuitable for the dynamic and complex unstructured roads required by special vehicles. For the latter, although it considers environmental perception uncertainty and perception model error, it performs circular expansion processing on obstacle shapes, resulting in unquantifiable uncertainty and loss of accuracy. Furthermore, it obtains differentiable probabilities through grid discretization, and the two-step approximation introduces significant numerical uncertainty, increasing the collision probability estimation error, which urgently needs to be addressed. Summary of the Invention

[0005] This application provides a special vehicle trajectory planning method that considers perception uncertainty in order to solve problems such as perception uncertainty caused by the environment for special vehicles.

[0006] The first aspect of this application provides a special vehicle trajectory planning method considering perception uncertainty, including the following steps: Acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, and calculate the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period based on a preset motion model, and obtain the position uncertainty reference trajectory of the obstacle based on the position uncertainty. Based on the location uncertainty reference trajectory, the collision probability of the current vehicle is determined, and based on the collision probability, a risk assessment model is constructed according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. At least one target state point is determined, and the at least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set. Opportunity constraints are constructed based on the collision probability, and the candidate trajectory set is filtered based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. The optimal trajectory is then solved based on the risk assessment model and the multiple trajectories.

[0007] Optionally, before calculating the positional uncertainty of the obstacle in the 3D point cloud data within a preset time period based on a preset motion model, the method further includes: The covariance matrix is ​​constructed based on a pre-defined consensus clustering algorithm; The state representation in Gaussian distribution is constructed based on the covariance matrix and the spatial mean of obstacles in the three-dimensional point cloud data. The preset motion model is constructed based on the state representation in the form of the Gaussian distribution.

[0008] Optionally, the preset deterministic constraints include feasible region constraints and kinematic constraints.

[0009] Optionally, the opportunity constraint is: ; in, The set of obstacles perceived at the current moment. The bounding box of the current vehicle. The bounding box of the obstacle. For Minkowski difference operation, This is the preset lower limit of safety confidence. In order to be in The confidence threshold at any given time.

[0010] Optionally, the deterministic boundary constraints obtained from the chance constraints include: The original chance constraint is transformed into an upper bound constraint on the spatial coverage collision probability, wherein the spatial coverage collision probability is obtained by coordinate transformation based on a preset collision probability estimation strategy. Based on a preset numerical integration formula and a preset strategy to reduce computational load, the preset deterministic boundary constraints are obtained according to the spatial coverage collision probability.

[0011] A second aspect of this application provides a special vehicle trajectory planning device that considers perception uncertainty, comprising: The acquisition module is used to acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, and calculate the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period based on a preset motion model, and obtain the position uncertainty reference trajectory of the obstacle based on the position uncertainty. The evaluation module is used to determine the collision probability of the current vehicle based on the location uncertainty reference trajectory, and to construct a risk assessment model based on the collision probability according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. The solution module is used to determine at least one target state point, sample the at least one target state point using a preset coordinate system to obtain a candidate trajectory set, construct opportunity constraints based on the collision probability, and filter the candidate trajectory set based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. Based on the risk assessment model, the optimal trajectory is solved according to the multiple trajectories.

[0012] Optionally, before calculating the positional uncertainty of the obstacle in the 3D point cloud data within a preset time period based on a preset motion model, the acquisition module is further configured to: The covariance matrix is ​​constructed based on a pre-defined consensus clustering algorithm; The state representation in Gaussian distribution is constructed based on the covariance matrix and the spatial mean of obstacles in the three-dimensional point cloud data. The preset motion model is constructed based on the state representation in the form of the Gaussian distribution.

[0013] Optionally, the preset deterministic constraints include feasible region constraints and kinematic constraints.

[0014] Optionally, the opportunity constraint is: ; in, The set of obstacles perceived at the current moment. The bounding box of the current vehicle. The bounding box of the obstacle. For Minkowski difference operation, This is the preset lower limit of safety confidence. In order to be in The confidence threshold at any given time.

[0015] Optionally, the solution module is specifically used for: The original chance constraint is transformed into an upper bound constraint on the spatial coverage collision probability, wherein the spatial coverage collision probability is obtained by coordinate transformation based on a preset collision probability estimation strategy. Based on a preset numerical integration formula and a preset strategy to reduce computational load, the preset deterministic boundary constraints are obtained according to the spatial coverage collision probability.

[0016] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform a special vehicle trajectory planning method considering perceived uncertainty as described in the above embodiments.

[0017] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the special vehicle trajectory planning method considering perceived uncertainty as described in the above embodiments.

[0018] A fifth aspect of this application provides a computer program product storing a computer program that, when executed by a processor, implements the special vehicle trajectory planning method considering perceptual uncertainty as described in the above embodiments.

[0019] Therefore, this embodiment of the application acquires three-dimensional point cloud data of the environment within a preset range of the current vehicle, calculates the position uncertainty based on a preset motion model, obtains the reference trajectory of the obstacle's position uncertainty, determines the collision probability of the current vehicle based on this, and constructs a risk assessment model based on the collision probability according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. At least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set, and opportunity constraints are constructed based on the collision probability. Based on the deterministic boundary constraints obtained from the opportunity constraints, the candidate trajectory set is filtered to obtain multiple trajectories. Finally, based on the risk assessment model, the optimal trajectory is solved from the multiple trajectories. This solves the problem of perception uncertainty caused by the environment for special vehicles.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a special vehicle trajectory planning method considering perception uncertainty provided according to an embodiment of this application; Figure 2 This is a 3D bounding box 7-dimensional parameter diagram of a special vehicle trajectory planning method considering perceptual uncertainty according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the spatial uncertainty at time 𝑡=0 in a special vehicle trajectory planning method considering perceived uncertainty according to an embodiment of this application. Figure 4 This is a schematic diagram illustrating an example of position uncertainty propagation within [0, 9]s in a special vehicle trajectory planning method considering perceived uncertainty provided according to an embodiment of this application; Figure 5 This is a schematic diagram of a deterministic sampling trajectory solving method for a special vehicle trajectory planning method that considers perceptual uncertainty, according to an embodiment of this application; Figure 6 This is a schematic diagram illustrating an example of Minkowski difference calculation for a special vehicle trajectory planning method considering perceptual uncertainty provided in an embodiment of this application. Figure 7 This is an overall technical flowchart of a special vehicle trajectory planning method that takes into account perceptual uncertainty, according to an embodiment of this application; Figure 8 This is a schematic diagram of a special vehicle trajectory planning device that takes into account perceived uncertainty, according to an embodiment of this application. Figure 9 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0022] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0023] Special-purpose vehicles operate in complex and hazardous environments, demanding exceptionally stringent safety and reliability standards. Therefore, autonomous driving technology presents both immense potential and significant challenges in this field. Special-purpose vehicles face complex and ever-changing environments during missions, including varying terrain, weather conditions, and traffic situations, greatly increasing the difficulty of perception and decision-making. Furthermore, perception results are often highly uncertain due to factors such as sensor noise, environmental interference, or sensor malfunctions. However, current decision-making systems often overlook this uncertainty, simplifying perception results into deterministic outputs, thus failing to accurately assess and address potential safety risks. In addition, the driving behavior and needs of special-purpose vehicles differ significantly from those of ordinary passenger vehicles. Traditional autonomous driving systems are typically designed based on general driving behavior models, which limits the trajectory planning capabilities and autonomous driving performance of special-purpose vehicles. Special-purpose vehicles must consider unique working environments and mission requirements. For example, off-road vehicles performing special missions need to maintain strong maneuverability in environments lacking explicit traffic rules and road network constraints, and possess the ability to respond quickly to highly dynamic surrounding obstacles; or medical rescue vehicles need to be able to quickly reach their destination in emergency situations. Therefore, the current research focus is on how to develop effective trajectory planning algorithms that take into account the strong perceptual uncertainties in the environment and meet the special needs of special vehicles.

[0024] In conclusion, considering perceptual uncertainty in the trajectory planning stage of special vehicles can improve the adaptability of special vehicle autonomous driving systems to complex environments, reduce safety risks, and improve task execution efficiency.

[0025] However, the research methods in related technologies lack a good representation of uncertainty, cannot take into account real-time performance, accuracy and numerical stability, and cannot simultaneously solve the problems of safe interaction and dimensionality curse under unstructured roads when modeling trajectory planning problems, which urgently need to be addressed.

[0026] This application addresses the aforementioned problems by proposing a trajectory planning method for special vehicles that considers perceptual uncertainty. In this method, embodiments acquire 3D point cloud data of the environment within a preset range of the current vehicle, calculate positional uncertainty based on a preset motion model, obtain a reference trajectory for obstacle positional uncertainty, and determine the collision probability of the current vehicle. Based on the collision probability, a risk assessment model is constructed according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. At least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set. Opportunity constraints are constructed based on the collision probability, and the candidate trajectory set is filtered based on deterministic boundary constraints obtained from the opportunity constraints, resulting in multiple trajectories. Finally, the optimal trajectory is solved based on the risk assessment model and multiple trajectories. This solves the problem of perceptual uncertainty caused by the environment for special vehicles, enhancing the system's robustness and adaptability.

[0027] Specifically, Figure 1 This is a flowchart illustrating a special vehicle trajectory planning method that considers perceived uncertainty, as provided in an embodiment of this application.

[0028] like Figure 1 As shown, the special vehicle trajectory planning method considering perceived uncertainty includes the following steps: In step S101, three-dimensional point cloud data of the environment within a preset range of the current vehicle is acquired, and based on a preset motion model, the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period is calculated, and the position uncertainty reference trajectory of the obstacle is obtained based on the position uncertainty. The preset duration can be a threshold set by the user, a threshold obtained through a limited number of experiments, or a threshold obtained through a limited number of computer simulations; no specific limitation is made here.

[0029] Specifically, in unstructured road environments, special vehicles face perception uncertainties caused by factors such as sensor measurement errors, the complexity of obstacle shapes, and environmental occlusion. To address this issue, this invention designs a multi-model fusion-based method for estimating the uncertainty of LiDAR perception results, used to extract the statistical uncertainty of obstacle spatial position and orientation in each frame. LiDAR can acquire high-precision 3D point cloud data of the environment within a preset range of the vehicle, enabling the measurement of key information such as the position, shape, and distance of target objects, providing accurate perception capabilities for autonomous driving systems. In 3D target detection, obstacles are typically labeled with 3D bounding boxes (bboxes). Figure 2 As shown, Figure 2This is a 3D bounding box 7-dimensional parameter diagram of a special vehicle trajectory planning method considering perceptual uncertainty according to an embodiment of this application. Each bounding box is typically represented by a 7-dimensional vector. Each dimension can be viewed as a random variable following a specific distribution, and the mean and variance of this distribution can be derived by statistically analyzing the results of multiple detections of the same sample. Here, considering that trajectory planning is performed only within a two-dimensional map plane, and the length and width of obstacles remain constant, this invention focuses on three random variables. This allows for a reasonable simplification of the model derivation. The 3D object detection algorithm used in this invention is not limited to a specific method; general neural network-based methods can be used. The neural network used can be a model with the same structure but different parameters (e.g., different randomly initialized models under the same detection framework), or it can be multiple network architectures with different structures (e.g., CenterPoint, PointPillars, SECOND, etc.). After independent inference, each model outputs the 3D bounding box parameters of the obstacle, including its center point. ,size and orientation angle For the same laser frame, the set of output boxes generated by multiple models is denoted as . ,in Indicates the first The model detected the first A bounding box.

[0030] Furthermore, in the preset motion model, let the obstacle state variable be... The state evolution formula is constructed through a linear state transition model, that is...

[0031] in, The state transition matrix is ​​a first-order approximation of the system's kinematic equations, i.e.

[0032] The propagation of uncertainty follows the following covariance recursive formula:

[0033] This method allows for the acquisition of obstacles in the future. to The position uncertainty estimate at each time step is used for subsequent collision probability modeling and trajectory evaluation.

[0034] assumed At a certain moment, a car is detected by LiDAR-MIMO, and its state variables are as follows:

[0035] ; By sampling the Gaussian distribution that its center position follows 1000 times, we can obtain the following: Figure 3 The location indicated by the blue dot is uncertain. Figure 3 This application provides an embodiment of a special vehicle trajectory planning method that considers perceived uncertainty. A schematic diagram illustrating the spatial uncertainty at time =0; sampling the Gaussian distribution followed by its orientation angle 10 times yields the attitude uncertainty shown in the yellow bbox in the diagram.

[0036] Furthermore, such as Figure 4 As shown, Figure 4 This is a schematic diagram illustrating the propagation of position uncertainty within [0, 9]s in a special vehicle trajectory planning method considering perceived uncertainty according to an embodiment of this application. The derivation is performed using the aforementioned uncertainty diffusion model, and a time step is set. = 1, then the reference trajectory of the current vehicle's position uncertainty within the next 9 seconds can be deduced.

[0037] In step S102, the collision probability of the current vehicle is determined based on the position uncertainty reference trajectory, and a risk assessment model is constructed based on the collision probability according to the preset decreasing dynamic confidence allocation strategy and the preset collision risk assessment conditions.

[0038] Specifically, based on the reference trajectory with location uncertainty, this invention introduces the concept of trajectory collision probability; this metric attempts to characterize the trajectory from time [time value missing]. arrive Between these events, the vehicle encounters obstacles while following the planned trajectory. The overall probability of a collision occurring. Due to time... Since it is not a random variable, the instantaneous collision probability at multiple time points cannot be marginalized and solved using Bayesian integration, thus a closed-form expression for the trajectory collision probability cannot be obtained. This invention employs the following upper bound estimation method to construct the risk assessment model:

[0039] in, Indicates at time Instantaneous collision probability estimated by obstacle boundary penetration model Since the uncertainty of the obstacle state can be approximated as being mainly dominated by its position uncertainty, and the bounding box is relatively small, it can be estimated using the following Gaussian integral approximation: ; Since this integral is difficult to solve analytically over convex polygonal regions, the upper bound of the probability can be estimated using the value of the Gaussian kernel at the center point, i.e.: ; In practical engineering implementation, to balance efficiency and conservatism, this invention uses the Gaussian cumulative distribution function (CDF) of the obstacle's location relative to the vehicle's center for approximation. The vehicle's center point is denoted as... Then there is ; in, The cumulative distribution function representing the standard normal distribution; This represents the safety radius after the vehicle's bounding box expands; For the equivalent standard deviation, by The calculation is obtained using the following formula: ; This approximation essentially transforms the high-dimensional probability integral problem into an estimation of a univariate Gaussian CDF, which not only has clear probabilistic physical meaning but is also suitable for efficient implementation in complex real-time systems. For a more accurate penetration probability estimate, the Monte Carlo sampling method can be used to sample obstacle states and statistically analyze the proportion of obstacles falling into the vehicle's bounding box. However, considering real-time requirements, this invention adopts approximate analytical estimation as the primary strategy. To improve the safety performance of the trajectory optimizer within the feasible region, this upper bound on the trajectory collision probability is explicitly introduced into the objective function and weighted by a weighting factor, forming a risk cost in the following form: ; in, It serves as a weighting factor for risk cost, used to adjust the balance between risk aversion and trajectory performance.

[0040] Therefore, compared with the traditional empirical risk cost based on distance functions, the collision probability assessment of this invention has a clear probabilistic physical meaning and upper bound guarantee, which can more reasonably guide the trajectory planning results to move away from high-risk areas. By introducing the above-mentioned collision risk assessment mechanism, this invention not only restricts the collision probability at the constraint level, but also constructs a safety-driven preference term in the optimization objective, thereby achieving dual protection for the trajectory safety of special vehicles in unstructured environments.

[0041] Furthermore, according to the preset decreasing dynamic confidence level allocation strategy, a higher safety confidence level is required in the early stage of the trajectory (near time), while a lower confidence level is acceptable in the later stage of the trajectory (far time). This strategy can be defined as follows:

[0042] in, For at any time Confidence threshold; The starting time of the planned trajectory segment The required maximum confidence level is usually set to a high value; The end time of the planned trajectory segment The minimum confidence level that can be tolerated is usually set to a low value; As a confidence level deceleration rate control factor, determine The descent rate. In system design, this strategy can be enabled by default or dynamically adjusted based on special vehicle mission priority, scenario type, obstacle density, etc. For example, in high-density obstacle scenarios, the overall confidence curve can be raised, or additional... To decay more quickly .

[0043] In order to more comprehensively characterize and suppress potential collision risks during trajectory optimization, in addition to constructing opportunity constraints based on confidence constraints, this invention also introduces preset collision risk assessment conditions into the objective function, so that the optimization solution is not only limited by hard constraint boundaries, but also tends to a safer trajectory selection within the feasible solution space.

[0044] It should be noted that this invention aims to provide a trajectory planning method with uncertainty perception capabilities for special vehicles in unstructured road environments. To this end, after modeling the perceived uncertainty, the planning system needs to organically integrate the vehicle trajectory state, control input, objective requirements, and safety boundaries into a formalized trajectory optimization problem. To ensure the modeling's universality and solvability, this invention uses the Frenet coordinate system to model the vehicle's state relative to a reference trajectory, and on this basis, constructs a discrete-time optimal control system, coupled with a multi-objective optimization function and feasible region constraints, forming a complete system modeling framework.

[0045] ; in Indicates that the vehicle is in The lateral offset of time relative to the reference trajectory. This represents the vehicle's longitudinal offset along the reference trajectory. The velocities and accelerations of both are obtained through differential estimation and together describe the vehicle's state along the reference path. The control input is... , representing jerk in the horizontal and vertical directions, is used to adjust trajectory smoothness and dynamic response.

[0046] Based on the above definitions of state and control variables, the vehicle trajectory evolution can be represented by the following discrete linear state transition model: ; Where the state transition matrix With control matrix for: ; To achieve robust, smooth trajectory generation with risk avoidance capabilities, this invention designs the following multiple objective functions to balance requirements such as safety, feasibility, execution efficiency, and trajectory stability during the optimization process. First, the feasible region risk term represents the risk of the vehicle's position approaching the boundary of a non-passable area, i.e.: ; in, The shortest distance between the vehicle and the nearest boundary. This is the preset safety buffer distance.

[0047] Secondly, the obstacle collision risk item measures the collision risk between the vehicle and surrounding obstacles. The maximum probability of a collision occurring is: ; in, To deal with obstacles In state The probability of a collision.

[0048] To enhance the smoothness of the trajectory, the system introduces a trajectory smoothing term into the objective function, which limits drastic motion changes by controlling the energy loss of acceleration. ; To ensure the continuity of the trajectory across multiple frames and reduce the safety risks caused by sudden control changes, a trajectory continuity term is introduced, namely: ; in, For the previous trajectory generation in The state point at time, It is a weighted matrix.

[0049] Finally, to ensure that the trajectory satisfies the task objective as much as possible at the final state, an execution efficiency term is introduced: ; in, To preset the target state, To plan the final time step.

[0050] In step S103, at least one target state point is determined, and the at least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set. Opportunity constraints are constructed based on the collision probability, and the candidate trajectory set is filtered based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. The optimal trajectory is then solved based on the multiple trajectories using a risk assessment model.

[0051] Specifically, this invention proposes a trajectory generation mechanism based on target state point sampling. Given a starting state, it reduces the trajectory search space and improves the convergence efficiency of trajectory optimization by constructively sampling target state points. Target state sampling is performed in the Frenet coordinate system (a preset coordinate system), which uses a reference trajectory as a baseline and effectively expresses the vehicle's longitudinal position and lateral offset along the path. Specifically, the sampling space includes three dimensions: longitudinal position... Horizontal offset and the planning termination time This invention employs a uniform sampling strategy in each dimension to obtain multiple sets of candidate endpoint states. Each candidate point includes not only its spatial location but also its final velocity and acceleration, ensuring trajectory smoothness and executability. The sampling process fully integrates the state information at the end of the reference trajectory, expanding the sampling target within a limited range both laterally and longitudinally, thereby ensuring high accessibility and reasonableness.

[0052] Furthermore, after obtaining the target state set, this invention uses a quintic polynomial curve to connect the trajectories between each set of initial and target states. The quintic polynomial has high expressive power in trajectory generation; it not only satisfies position constraints but also simultaneously satisfies the continuity requirements of the first and second derivatives of velocity and acceleration, generating a smooth trajectory that conforms to the vehicle's dynamic characteristics. The current state of the vehicle is set as... The target state is Then, in the horizontal and vertical directions, we can construct fifth-order polynomial loci of the following forms:

[0053] ; coefficient The trajectory is uniquely determined by boundary conditions, satisfying the given initial and final state (position, velocity, acceleration) continuity requirements. This method avoids trajectory oscillations or abrupt changes, possesses strong dynamic executability and computational efficiency, and is particularly suitable for use in scenarios requiring high-frequency replanning within a short time domain. Each generated trajectory will enter subsequent modules for feasibility verification and constraint checks to ensure it meets conditions such as collision safety, kinematic limits, and feasible region constraints. However, for obstacle avoidance safety constraints expressed as chance constraints, while directly introducing probabilistic chance constraints into the trajectory planning problem has good physical meaning, it is often difficult to solve directly in applications with high real-time requirements due to its involvement of high-dimensional integrals of random variables or joint probability calculations. Therefore, to improve solution efficiency, this invention, after constructing collision probability chance constraints, further proposes an analytical deterministic transformation strategy to transform the original chance constraint problem into a deterministic constraint problem that is easy to solve numerically.

[0054] Therefore, the present invention adopts The concept of -shadow transforms chance constraints into analytical deterministic ones, thereby improving the efficiency of trajectory optimization while ensuring collision safety. -shadow refers to the shadowing of random state variables. The corresponding probability distribution, construct a set that satisfies area It is called of -shadow. This set can be obtained through the Minkowski sum of the probability ellipsoid formed by the nominal position and positional uncertainty of the obstacle and the geometric boundary.

[0055] Furthermore, after completing the sampling of trajectory target points and the generation of fifth-order polynomial trajectories, each candidate trajectory needs to undergo item-by-item verification of its constraints to ensure that it meets the requirements of vehicle motion capability and environmental safety throughout the entire planning time domain. The constraint checking process covers multiple deterministic and opportunistic constraints constructed in the system model, mainly including vehicle dynamics constraints (such as velocity, acceleration, and curvature boundaries), feasible region collision-free constraints, and constraints on the instantaneous spatial coverage collision probability.

[0056] Specifically, regarding the spatial constraints of the feasible region and obstacles, the aforementioned modeling method is used to calculate in real time the maximum probability value of the vehicle's bounding box being covered by the obstacle probability cloud at each moment on the trajectory. A set confidence threshold is used as the upper bound to ensure that the planned trajectory has acceptable collision safety in a dynamic environment. If a trajectory violates any of the above constraints at any point in time, it is considered an infeasible solution and is removed from the candidate trajectory set. After constraint checking, all the remaining trajectories will be used as the feasible solution set for subsequent trajectory optimization.

[0057] Furthermore, such as Figure 5 As shown, Figure 5 This diagram illustrates a deterministic sampling trajectory solution method for a special vehicle trajectory planning method considering perceived uncertainty, according to one embodiment of this application. After constraint checks, a trajectory evaluation mechanism is introduced to quantitatively assess the overall performance of each trajectory in order to select the final target trajectory for vehicle execution. The trajectory selection is based on the optimization objective function. The trajectory evaluation index design covers multiple dimensions, including feasible domain risk, obstacle collision risk, trajectory smoothness constraint, trajectory continuity, and target state deviation. By weighted summing of each index, the total cost function value of each trajectory over the entire planning time domain can be obtained. During the evaluation process, the state variables and control variables of each trajectory from the start time to the end time are substituted into the evaluation model for evaluation, and the results are summarized as the overall trajectory performance index. Finally, the trajectory with the smallest total cost function value among all feasible trajectories is selected as the optimal trajectory and output to the downstream vehicle control module for actual execution.

[0058] Optionally, in some embodiments, the preset deterministic constraints include feasible region constraints and kinematic constraints.

[0059] Understandably, feasible region constraints ensure that vehicles do not encroach on impassable areas during the planning process. This constraint can be constructed based on two types of information sources: if the higher-level path planning provides lateral boundary distance information, the geometric relationship between the vehicle's envelope contour and the left and right boundaries can be calculated to ensure: ;in, This indicates the set minimum forward safety distance (used to determine whether the obstacle spacing requirement is met). This represents the distance from the vehicle's center of gravity to the left and right obstacle boundaries, and its specific calculation method is as follows:

[0060]

[0061] ; in, Indicates the distance from the obstacle boundary to the reference trajectory; This represents the outer radius of the vehicle from its center of mass to its boundary (including vehicle dimensions and safety redundancy). This represents the distance from the vehicle's center of gravity to the center of the front axle (vehicle geometry parameter). This represents the equivalent overlap distance between the vehicle edge and the obstacle edge; This is represented as the lateral error boundary between the reference trajectory and the actual trajectory; This is a term related to the vehicle steering angle, used to correct the impact of vehicle orientation on boundary distance calculations; This is the distance from the front axle of the vehicle to its center of gravity. This indicates the vehicle's turning radius, which is inversely derived from the curvature. The angle between the vehicle's current heading and the tangent direction of the reference trajectory.

[0062] If the map is in raster format, the system will detect the overlap between the vehicle's geometric envelope and the raster obstacle region. This detection can be performed using the SAT (Separating Axis Theorem) method or the bounding point lookup method, balancing accuracy and efficiency.

[0063] Furthermore, kinematic constraints require that the velocity, acceleration, and curvature of the trajectory at each instant satisfy the vehicle's motion limit requirements, namely: .

[0064] In summary, this invention constructs a complete trajectory planning system model, combining Frenet trajectory state description, discrete-time system dynamics, optimal control structure with multiple risk objective functions, and explicit safety and physical constraints to form a trajectory optimization problem with safety, smoothness, and task completion as objectives.

[0065] Optionally, in some embodiments, the chance constraint is: ; in, The set of obstacles perceived at the current moment. The bounding box of the current vehicle. The bounding box of the obstacle. For Minkowski difference operation, This is the preset lower limit of safety confidence. In order to be in The confidence threshold at any given time.

[0066] It is understandable that in special vehicle trajectory planning tasks, the vehicle perceives surrounding dynamic obstacles in real time while following the planned trajectory. Due to factors such as perception errors, occlusion, and changes in lighting, the position and attitude of obstacles are uncertain. Therefore, traditional Boolean collision detection methods are difficult to effectively model such probabilistic risks and cannot reflect a controllable collision constraint space in the optimization process. To address this, this invention establishes a probabilistic collision risk model and introduces a chance constraint expression to formally limit the probability of unsafe trajectories occurring. This invention uses the following form of chance constraint to limit the probability of collision between the vehicle and surrounding obstacles during trajectory execution: ; In the formula, This represents the set of obstacles perceived at the current moment. and These are the bounding boxes for the vehicle and the obstacles, respectively. This represents the Minkowski difference operation. This indicates the preset lower limit of safety confidence. Obstacles. The state is a random variable with uncertainty, and its bounding box position vector It follows a Gaussian distribution.

[0067] Furthermore, such as Figure 6 As shown, Figure 6 This is a schematic diagram illustrating an example of Minkowski difference calculation for a special vehicle trajectory planning method considering perceived uncertainty according to an embodiment of this application. Geometrically, the Minkowski difference... This represents the set of vector differences between all points in the vehicle's bounding box and the obstacle's bounding box. If this set covers the origin, the two convex bounding boxes overlap, resulting in a collision. Based on the GJK (Gilbert-Johnson-Keerthi) collision detection theory, whether two convex sets overlap is equivalent to whether their Minkowski difference contains the origin. This constraint constitutes the core safety condition in the trajectory optimization problem, providing a clear and controllable risk boundary for the subsequent generation of the optimal trajectory. However, when within an execution time... When defining such joint constraints, the probability of collision events being jointly true at all time points is often low, resulting in overly conservative chance constraints, which significantly compresses the feasible domain of the trajectory optimization problem and seriously affects the quality and real-time performance of the solution.

[0068] Therefore, this invention proposes an improved opportunity constraint mechanism to address the characteristics of trajectory planning, such as continuous replanning of the vehicle's trajectory and the fact that the actual path of the vehicle is not strictly equal to the planned trajectory: at each discrete time point Independent constraints are constructed on the upper bound, and a dynamic confidence threshold is introduced. Thus, the expression for the opportunity constraint is obtained.

[0069] Optionally, in some embodiments, the deterministic boundary constraints obtained from the chance constraints include: transforming the original chance constraint into an upper bound constraint on the spatial coverage collision probability, wherein the spatial coverage collision probability is obtained by coordinate transformation based on a preset collision probability estimation strategy; and obtaining a preset deterministic boundary constraint based on the spatial coverage collision probability according to a preset numerical integration formula and a preset strategy for reducing computational load.

[0070] It is understood that, in this invention, in order to transform the above description into geometric reachability constraints for the vehicle, it is considered to calculate any point on the vehicle's bounding box. The maximum probability of being blocked by obstacles is used to estimate the maximum spatial coverage risk of a trajectory point. This leads to the original chance constraint formula: ; This can be transformed into the following upper bound constraint on spatial coverage collision probability: ; The method for estimating the collision probability can be expressed as: ; The indicative function is defined as follows: ; To enable the integration region to accept the integration kernel function for coordinate transformation, the following variable substitution is defined: ; The collision probability expression can be transformed into:

[0071] Because this integral is a triple unclosed Gaussian integral, the computational load is enormous, and conventional solutions require more than a few seconds, which cannot meet the real-time requirements of trajectory optimization. Therefore, this invention further introduces the Gauss-Hermite numerical integration method (a preset numerical integration formula), the basic form of which is:

[0072] in, Hermite polynomial The root, The weighting coefficients are used. After performing a 5th-order Gauss-Hermite expansion of the triple integral, the number of basic operations can be reduced to 125. Furthermore, to meet the real-time requirements of trajectory planning at a frequency of 10Hz, this invention reduces computational load through preset strategies, including: replacing the single integral with the C++ built-in error function erf(); reducing the sampling of obstacle orientation angles from 5 times to 2 times; and calculating only the spatial occupancy probability of the four corner points and the center point of the vehicle boundary.

[0073] Therefore, the number of basic operations required for each trajectory evaluation is reduced to a constant level (<10 times), achieving efficient collision probability estimation within an acceptable accuracy range, thus effectively transforming the probabilistic chance constraints in trajectory optimization into real-time solvable deterministic boundary constraints.

[0074] Optionally, in some embodiments, before calculating the positional uncertainty of obstacles in the 3D point cloud data within a preset time period based on a preset motion model, the method further includes: constructing a covariance matrix based on a preset consensus clustering algorithm; constructing a Gaussian distribution state representation based on the covariance matrix and the spatial mean of obstacles in the 3D point cloud data; and constructing a preset motion model based on the Gaussian distribution state representation.

[0075] Understandably, to achieve consistent judgment and fusion of detection results from multiple models, this invention employs a consensus clustering algorithm based on the IoU distance metric to automatically identify the detection results of the same object in multiple models and classify them into the same category. Based on this, the center of the detection box within each cluster category is determined. Orientation Perform parameter sampling and construct the following covariance matrix:

[0076] in, , , These represent obstacles at... direction, Direction and orientation angle The variance on the current frame reflects the variance of the current frame. The uncertainty of the spatial state of obstacles at any given moment.

[0077] The covariance matrix, along with the spatial mean of the obstacles, Together they form the state representation of a Gaussian distribution. The above distribution is used to characterize the uncertainty of the spatial position and orientation of obstacles estimated by the perception system in the current frame.

[0078] Due to the lack of clear road signage and traffic rule constraints, the driving behavior of road users in unstructured roads is often highly dynamic and unpredictable. Special vehicles also possess a certain degree of off-road capability, allowing them to traverse a wider range of road conditions compared to passenger cars. Furthermore, driving intentions in off-road environments are difficult to predict and model, and there is currently no large-scale and complete trajectory prediction dataset for unstructured roads. Trajectory prediction research methods in related technologies are also difficult to apply to this unstructured scenario.

[0079] To further assess the trend of uncertainty evolving over time, this invention introduces a motion model (a preset motion model) based on the assumption of constant acceleration, propagating the aforementioned Gaussian state to several future moments, and compensating for the uncertainty of the vehicle's intention by shortening the prediction time domain and increasing the prediction frequency.

[0080] To facilitate those skilled in the art to further understand the special vehicle trajectory planning method considering perceived uncertainty in the embodiments of this application, the following is combined with... Figure 7 The embodiments shown will be described in detail.

[0081] Specifically, such as Figure 7 As shown, Figure 7This is a flowchart illustrating the overall technical process of a special vehicle trajectory planning method considering perceptual uncertainty, according to one embodiment of this application. This embodiment quantifies the spatial uncertainty of surrounding obstacles in real time by fusing detection results from multiple neural network outputs, and extrapolates the future short-term uncertainty diffusion trend based on the target motion model. On this basis, a risk modeling mechanism centered on collision probability is constructed, and a trajectory planning problem framework is introduced through chance constraints. Furthermore, a trajectory evaluation index system combining dynamic constraints and safety constraints is designed. Considering the difficulty in solving optimal control problems with probabilistic constraints in practical systems and the challenge in meeting response time limits, this invention transforms chance constraints into deterministic expressions and introduces a sampling mechanism based on the target state to construct a candidate trajectory set. This achieves trajectory planning under the premise of satisfying collision risk control, thereby enhancing the response capability of special vehicles to perceptual uncertainty and dynamic risks in unstructured environments.

[0082] Furthermore, the system in this embodiment consists of three stages: quantification of perceived uncertainty, modeling of the trajectory planning problem based on chance constraints, and solving the trajectory planning problem based on sampling. These three stages process sensor data input, risk modeling, and trajectory optimization tasks sequentially, ultimately outputting the optimal driving trajectory that meets the collision probability control requirements.

[0083] Therefore, the trajectory safety constraint mechanism based on uncertainty modeling proposed in this invention, while fully considering perception errors and target behavior uncertainties, characterizes the collision risk of the trajectory in a probabilistic form, overcoming the conservative problem of feasible region caused by rigid distance constraints in traditional methods, and providing a solid guarantee for the efficient and safe passage of special vehicles in complex unstructured environments. The dynamic confidence constraint mechanism proposed in this invention, by dynamically allocating the collision confidence level in the trajectory time domain, enables the system to more flexibly balance multiple factors such as target point distance, feasible region boundary distance, and obstacle distribution while taking into account both real-time performance and safety, effectively expanding the trajectory solution space in high-risk environments and improving the adaptability and robustness of path optimization. The opportunity-constraint deterministic transformation mechanism based on the -shadow concept avoids the high latency problem caused by high-dimensional integral calculations. It transforms complex probabilistic constraints into simple and clearly expressed distance-domain analytic functions, improving the real-time response capability of the trajectory planning module and meeting the stringent requirements of trajectory planning periodicity for special vehicles in dynamic environments. The trajectory safety probabilistic modeling method proposed in this invention, from a spatial coverage perspective, directly estimates the upper bound of risk covered by the probability cloud of dynamic obstacles at the vehicle boundary. This abandons traditional collision detection methods based on Mahalanobis distance or geometric discrimination, making collision risk modeling more physically interpretable and laying the foundation for subsequent soft constraint modeling of risk levels in the objective function. The trajectory generation and filtering mechanism constructed in this invention has engineering advantages such as clear structure, unified interface, and strong adaptability. It can be flexibly embedded into different types of special vehicle task management and path control systems, and is suitable for various high-requirement scenarios such as complex field scenes, unmanned task scheduling, and urban emergency transportation, possessing broad application value.

[0084] According to the trajectory planning method for special vehicles considering perception uncertainty proposed in this application, this application acquires three-dimensional point cloud data of the environment within a preset range of the current vehicle, calculates the position uncertainty based on a preset motion model, obtains a reference trajectory for the position uncertainty of obstacles, and determines the collision probability of the current vehicle based on this. Based on the collision probability, a risk assessment model is constructed according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. At least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set. Opportunity constraints are constructed based on the collision probability, and the candidate trajectory set is filtered based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. Finally, based on the risk assessment model, the optimal trajectory is solved from the multiple trajectories. This solves the perception uncertainty of special vehicles caused by the environment.

[0085] Next, referring to the accompanying drawings, a special vehicle trajectory planning device considering perceived uncertainty is described according to an embodiment of this application.

[0086] Figure 8 This is a block diagram of a special vehicle trajectory planning device that takes into account perceived uncertainty, according to an embodiment of this application.

[0087] like Figure 8 As shown, the special vehicle trajectory planning device 10 that takes into account perceived uncertainty includes: an acquisition module 100, an evaluation module 200, and a solution module 300.

[0088] The acquisition module 100 is used to acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, and calculate the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period based on the preset motion model, and obtain the position uncertainty reference trajectory of the obstacle based on the position uncertainty. The evaluation module 200 is used to determine the collision probability of the current vehicle based on the location uncertainty reference trajectory, and to construct a risk assessment model based on the collision probability according to the preset decreasing dynamic confidence allocation strategy and the preset collision risk assessment conditions. The solver module 300 is used to determine at least one target state point, sample at least one target state point using a preset coordinate system to obtain a candidate trajectory set, construct opportunity constraints based on collision probability, and filter the candidate trajectory set based on deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. Based on the risk assessment model, the optimal trajectory is determined according to the multiple trajectories.

[0089] Optionally, before calculating the positional uncertainty of obstacles in the 3D point cloud data within a preset time period based on a preset motion model, the acquisition module 100 is further configured to: construct a covariance matrix based on a preset consensus clustering algorithm; construct a state representation in Gaussian distribution form based on the covariance matrix and the spatial mean of obstacles in the 3D point cloud data; and construct a preset motion model based on the state representation in Gaussian distribution form.

[0090] Optionally, the preset deterministic constraints include feasible region constraints and kinematic constraints.

[0091] Optionally, the opportunity constraint is: ; in, The set of obstacles perceived at the current moment. The bounding box of the current vehicle. The bounding box of the obstacle. For Minkowski difference operation, This is the preset lower limit of safety confidence. In order to be in The confidence threshold at any given time.

[0092] Optionally, the solver module 300 is specifically used to: transform the original chance constraint into an upper bound constraint for spatial coverage collision probability, wherein the spatial coverage collision probability is obtained by coordinate transformation based on a preset collision probability estimation strategy; and based on a preset numerical integration formula and a preset strategy for reducing computational load, a preset deterministic boundary constraint is obtained according to the spatial coverage collision probability.

[0093] It should be noted that the foregoing explanation of the special vehicle trajectory planning method embodiment considering perception uncertainty also applies to the special vehicle trajectory planning device considering perception uncertainty in this embodiment, and will not be repeated here.

[0094] According to the special vehicle trajectory planning device considering perception uncertainty proposed in the embodiments of this application, the embodiments of this application acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, calculate the position uncertainty based on a preset motion model, obtain the reference trajectory of the obstacle's position uncertainty, determine the collision probability of the current vehicle based on this, and construct a risk assessment model based on the collision probability according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions; sample at least one determined target state point using a preset coordinate system to obtain a candidate trajectory set, construct opportunity constraints based on the collision probability, and filter the candidate trajectory set based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories; and solve for the optimal trajectory based on the risk assessment model and multiple trajectories. Thus, the perception uncertainty of special vehicles caused by the environment is solved.

[0095] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0096] When the processor 902 executes the program, it implements the special vehicle trajectory planning method that takes into account perceived uncertainty provided in the above embodiments.

[0097] Furthermore, electronic devices also include: Communication interface 903 is used for communication between memory 901 and processor 902.

[0098] The memory 901 is used to store computer programs that can run on the processor 902.

[0099] The memory 901 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0100] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0101] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0102] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0103] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described special vehicle trajectory planning method considering perceived uncertainty.

[0104] This application also provides a computer program product that stores a computer program that, when executed by a processor, implements the above-mentioned special vehicle trajectory planning method that takes into account perceived uncertainty.

[0105] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0106] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0107] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0108] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0109] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A special vehicle trajectory planning method considering perceptual uncertainty, characterized in that, The method includes the following steps: Acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, and calculate the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period based on a preset motion model, and obtain the position uncertainty reference trajectory of the obstacle based on the position uncertainty. Based on the location uncertainty reference trajectory, the collision probability of the current vehicle is determined, and based on the collision probability, a risk assessment model is constructed according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. At least one target state point is determined, and the at least one target state point is sampled using a preset coordinate system to obtain a candidate trajectory set. Opportunity constraints are constructed based on the collision probability, and the candidate trajectory set is filtered based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. Based on the risk assessment model, the optimal trajectory is determined according to the multiple trajectories.

2. The method according to claim 1, characterized in that, Before calculating the positional uncertainty of obstacles in the 3D point cloud data within a preset time period based on a preset motion model, the method further includes: The covariance matrix is ​​constructed based on a pre-defined consensus clustering algorithm; The state representation in Gaussian distribution is constructed based on the covariance matrix and the spatial mean of obstacles in the 3D point cloud data. The preset motion model is constructed based on the state representation in the form of the Gaussian distribution.

3. The method according to claim 1, characterized in that, The preset deterministic constraints include feasible region constraints and kinematic constraints.

4. The method according to claim 1, characterized in that, The opportunity constraint is: ； in, The set of obstacles perceived at the current moment. The bounding box of the current vehicle. The bounding box of the obstacle. For Minkowski difference operation, This is the preset lower limit of safety confidence. In order to be in The confidence threshold at any given time.

5. The method according to claim 1, characterized in that, The deterministic boundary constraints obtained from the chance constraints include: The original chance constraint is transformed into an upper bound constraint on the spatial coverage collision probability, wherein the spatial coverage collision probability is obtained by coordinate transformation based on a preset collision probability estimation strategy. Based on a preset numerical integration formula and a preset strategy to reduce computational load, the preset deterministic boundary constraints are obtained according to the spatial coverage collision probability.

6. A special vehicle trajectory planning device considering perceptual uncertainty, characterized in that, include: The acquisition module is used to acquire three-dimensional point cloud data of the environment within a preset range of the current vehicle, and calculate the position uncertainty of the obstacle in the three-dimensional point cloud data within a preset time period based on a preset motion model, and obtain the position uncertainty reference trajectory of the obstacle based on the position uncertainty. The evaluation module is used to determine the collision probability of the current vehicle based on the location uncertainty reference trajectory, and to construct a risk assessment model based on the collision probability according to a preset decreasing dynamic confidence allocation strategy and preset collision risk assessment conditions. The solution module is used to determine at least one target state point, sample the at least one target state point using a preset coordinate system to obtain a candidate trajectory set, construct opportunity constraints based on the collision probability, and filter the candidate trajectory set based on the deterministic boundary constraints obtained from the opportunity constraints to obtain multiple trajectories. Based on the risk assessment model, the optimal trajectory is solved according to the multiple trajectories.

7. The apparatus according to claim 6, characterized in that, Before calculating the positional uncertainty of obstacles in the 3D point cloud data within a preset time period based on the preset motion model, the acquisition module is further configured to: The covariance matrix is ​​constructed based on a pre-defined consensus clustering algorithm; The state representation in Gaussian distribution is constructed based on the covariance matrix and the spatial mean of obstacles in the 3D point cloud data. The preset motion model is constructed based on the state representation in the form of the Gaussian distribution.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the special vehicle trajectory planning method considering perceived uncertainty as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the special vehicle trajectory planning method considering perceived uncertainty as described in any one of claims 1-5.

10. A computer program product, said computer program product storing a computer program, characterized in that, When the program is executed by the processor, it implements the special vehicle trajectory planning method considering perceived uncertainty as described in any one of claims 1-5.

Images