An improved risk field-based multi-objective trajectory optimization method for unmanned vehicles

Abstract

The application discloses an improved risk field-based multi-target trajectory optimization method for an unmanned vehicle, and belongs to the technical field of intelligent traffic and automatic driving.The method comprises the following steps: three risk fields are respectively constructed, i.e., a driving environment risk field which is combined with road shape, static obstacles and environmental conditions; a vehicle risk field which considers vehicle size, position, driving direction and simultaneously considers the motion interaction between the vehicle and obstacles; and a predicted trajectory risk field which evaluates future risks in advance according to the possible driving trajectory of surrounding vehicles, combines trajectory credibility and uncertainty, and integrates the three risk fields to form a comprehensive risk field, constructs a three-dimensional space-time model based on the risk field, and finally outputs an executable trajectory by considering driving safety, efficiency, comfort and economy through reasonable safety strategies and multi-target optimization rules.The application has high risk evaluation accuracy in a dynamic traffic environment and good trajectory planning effect.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and autonomous driving technology, and in particular to a method for optimizing the multi-objective trajectory of unmanned vehicles based on an improved risk field. Background Technology

[0002] Trajectory planning for autonomous vehicles requires generating driving trajectories that satisfy multiple constraints in real time within dynamic traffic environments, and is one of the core technologies for achieving high-level autonomous driving. Effective trajectory optimization relies on accurate risk quantification and multi-objective coordination.

[0003] In terms of risk assessment, existing methods can be mainly divided into four categories: deterministic assessment methods ignore motion uncertainty and have limited application scenarios; probabilistic assessment methods are more reasonable in quantification but have higher computational costs; reachability set-based methods provide comprehensive assessment but have high computational complexity and poor real-time performance; potential field-based methods can consider multiple scenario elements at the same time, but existing potential field models are difficult to accurately characterize the anisotropic characteristics of vehicle risk and lack effective characterization of the uncertainty of predicted trajectory.

[0004] In trajectory planning, sampling methods, artificial potential field methods, numerical optimization methods, and deep learning and reinforcement learning methods each have their own advantages and disadvantages. As technology matures, the research focus has shifted from single safety assurance to multi-objective collaborative optimization, but existing methods lack effective coordination mechanisms in the collaborative optimization of safety, efficiency, comfort, and economy.

[0005] Chinese invention patent document CN113188556B discloses a method and device for trajectory planning of intelligent connected vehicles based on a driving safety field. The method calculates the risk posed by the traffic environment to the intelligent connected vehicle based on the driving safety field, where the risk is obtained by combining the field strengths of the potential energy field, kinetic energy field, and behavioral field. A sparse searchable region is generated based on the surrounding road traffic environment, the vehicle speed, and the prediction time domain. A lane selection trajectory is generated in the searchable region based on the risk, and the sparse search region is refined based on the lane selection trajectory. The optimal path is obtained by using the gradient descent method, and the speed is planned by using an adaptive following method. Finally, the final planned trajectory is obtained by dynamic simulation in the prediction time domain.

[0006] Chinese invention patent document CN112622932B discloses an autonomous driving lane-changing trajectory planning algorithm based on potential energy field heuristic search. This method establishes potential energy field functions for dynamic vehicles, lane lines, and lane boundaries based on the characteristics of various elements in the road traffic environment; on the basis of the risk potential energy field, it calibrates the minimum unacceptable risk threshold for vehicles, divides the non-intrusive area of ​​vehicle trajectory, and uses the A-star heuristic search algorithm to search for the path with the minimum risk value; it designs constraints based on model predictive control, including safety constraints and comfort constraints, and solves a safe, comfortable, and executable autonomous driving lane-changing trajectory with reference to the A-star planned trajectory.

[0007] Chinese invention patent document CN112644486B discloses an intelligent vehicle obstacle avoidance trajectory planning method based on a driving safety field. The method establishes a driving safety field model, including the potential energy field of stationary objects, the kinetic energy field of moving objects, and the behavior field. The potential energy field of stationary objects is further divided into a first type of stationary object (objects that would cause great damage if they collide) and a second type of stationary object (objects that constrain driving behavior, such as road markings), which are modeled separately. Based on the magnitude of the field value at various points in the driving safety field, the obstacle avoidance trajectory is planned, and an index function based on comfort, lane change distance, stability, and continuity is established to select the optimal obstacle avoidance trajectory.

[0008] Chinese invention patent document CN119761582B discloses a long-term driving risk identification method and system based on multimodal trajectory prediction. This method performs unified vectorization processing on the historical trajectories of the target vehicle and its adjacent vehicles, inputting them into a hierarchical interactive feature extraction network to obtain global hierarchical multi-scale interactive features; it obtains the multimodal driving intention distribution of the target vehicle based on a joint model, selecting high-probability candidate intention target points; it calculates the acceleration and front wheel steering angle distribution parameters of the target vehicle at various future times, inputting them into the vehicle kinematics model to generate a high-quality multimodal predicted trajectory set; it calculates a comprehensive driving risk field based on the predicted trajectory, covering the static obstacle risk field, the road risk field, and the multimodal predicted dynamic obstacle risk field, and identifies the driving risk distribution in the future time domain based on this risk field.

[0009] Chinese invention patent document CN115503700B discloses an active collision avoidance method and vehicle electronic device based on a multi-risk fusion potential field. The method acquires information about the vehicle and its surrounding environment, predicts the driving trajectory of the vehicle and other vehicles, and determines the positional relationship of the trajectories. If the trajectories intersect, the intersection point is calculated; if the trajectories are parallel and there is a possibility of collision, the rear-end collision point is calculated. The method uses an artificial potential field method to construct risk domains for lane boundary lines, static obstacles, dynamic obstacles, predicted trajectory intersection points, and collision points. The collision avoidance path is subject to real-time curvature constraints, and the optimal virtual force in the current time period is determined based on the principle of minimizing driving risk, thus planning the real-time optimal local collision avoidance trajectory.

[0010] Chinese invention patent document CN115416656B discloses an autonomous driving lane-changing method, device, and medium based on multi-objective trajectory planning. The method constructs a collision risk model and a traffic efficiency model; obtains lane sets for different traffic scenarios; constructs a multi-objective planning model based on the lane sets, collision risk model, and traffic efficiency model; uses collision risk, comfort, traffic efficiency, and path consistency between preceding and following times as optimization objectives; solves the multi-objective planning model using a non-dominated sorting dynamic programming algorithm to obtain a Pareto solution set; compares the Pareto solution sets of various modes to determine the target lane; and generates a smooth path based on the path set using a quadratic programming method as a guide line for local trajectory planning.

[0011] While the aforementioned patents have made valuable explorations in trajectory planning based on potential or risk fields, they still have the following limitations: First, in terms of risk field modeling, existing methods mainly employ isotropic potential field models or simple elliptical Gaussian distributions, which are insufficient to accurately characterize the anisotropic spatial distribution characteristics of vehicle risk caused by factors such as vehicle size and driving direction. In particular, they fail to establish a unified anisotropic risk field model that integrates spatial orientation and motion direction. Second, regarding the handling of prediction uncertainty, although some patents consider multimodal trajectory prediction, they lack an effective mechanism for characterizing the cumulative effect of prediction trajectory uncertainty over time and space, and fail to couple prediction confidence with risk field distribution. Third, in terms of multi-objective optimization, existing methods mostly employ simple weighted summation or Pareto optimization strategies, primarily focusing on safety, comfort, and traffic efficiency, lacking a systematic consideration of economy (energy consumption), and failing to establish a safety cost function that integrates risk gradient information to achieve proactive risk avoidance. Therefore, there is an urgent need to propose a trajectory planning technology for autonomous vehicles that can accurately quantify the anisotropic distribution of dynamic risks, effectively characterize the cumulative effect of prediction uncertainty, and achieve multi-objective collaborative optimization of safety, efficiency, comfort, and economy. Summary of the Invention

[0012] To address the shortcomings of existing technologies, the present invention aims to provide a multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field, which solves the problems of difficulty in quantifying the spatial distribution of risk in dynamic multi-vehicle interaction environments and difficulty in coordinating the optimization of multiple objectives such as safety, efficiency, comfort, and energy consumption.

[0013] To achieve the above objectives, the present invention adopts the following technical solution: a multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field, comprising the following steps:

[0014] Step 1: Construct the driving environment risk field, which integrates the road geometric potential field, the static obstacle risk field, and environmental condition influencing factors. The road geometric potential field includes the boundary potential field and the lane line potential field. The static obstacle risk field is modeled separately according to the geometric characteristics of the obstacles, which are divided into three categories: point, line, and region. Step 2: Construct a vehicle risk field, introduce spatial orientation and motion direction factors to characterize the anisotropic characteristics of vehicle risk, combine equivalent mass and vehicle condition factors to construct an objective risk field, and integrate collision trend factors and vehicle response factors to realize the transformation of objective risk into effective risk. Step 3: Construct a predicted trajectory risk field. Based on the multimodal predicted trajectories of surrounding vehicles and their confidence levels, introduce a height attenuation coefficient and dynamic diffusion width to transform the predicted trajectory into a spatiotemporal risk distribution. A forward-looking risk assessment is achieved through confidence-weighted fusion. Step 4: Construct a comprehensive driving risk field by weighting and fusing the vehicle risk field, the predicted trajectory risk field, and the driving environment risk field to obtain a unified comprehensive driving risk field; Step 5: Multi-objective trajectory optimization based on 3D spatiotemporal situation map. Construct a 3D spatiotemporal situation map to transform the trajectory planning problem into an optimal path search problem on a discrete graph. Identify safe areas through dynamic thresholds, design a multi-objective cost function that integrates risk gradients to achieve coordinated optimization of safety, efficiency, comfort and economy. Use a heuristic graph search method to solve for the optimal trajectory and output an executable trajectory after smoothing.

[0015] Step 1 specifically includes:

[0016] Step 1.1, Construct the road geometric potential field Including the boundary potential field Potential field with lane dividing line : ; The boundary potential field A piecewise exponential function is used to constrain vehicle movement within the road area and guide vehicles to travel along the center of the lane: ; in, Let y be the road width, y be the lateral position, and k1, k2, and k3 be the boundary potential field parameters. The potential field of the lane dividing line forms a risk "threshold" at the lane dividing line, preventing vehicles from crossing the lane dividing line at will: ; in, This is the risk intensity coefficient for the dividing line. This represents the width of the risk distribution at the dividing line.

[0017] Step 1.2: Based on geometric features, static obstacles are categorized into three types: point-like, line-like, and region-like, and modeled separately to generate a static obstacle risk field. ; ; The risk field of the point-like obstacles, such as cones and warning signs, is as follows: ; in, The distance from a point in space to an obstacle. The risk intensity coefficient for point-like obstacles. The distance decay exponent, For numerical smoothing terms; The risk field of the linear obstacles, such as guardrails and construction barriers, is as follows: ; in, It represents the shortest distance from a point in space to the line segment. The risk intensity coefficient for linear obstacles. The distance decay exponent, For numerical smoothing terms; For obstacles in the area, such as construction areas and hazardous areas, the area with the highest risk value is designated as a no-entry zone, while the external risk decreases exponentially with distance. ; In the formula, This represents the regional risk intensity coefficient. The risk attenuation coefficient outside the area is used, and the maximum risk value inside the area is used to form a restricted zone. The external risk decreases exponentially with distance, so as to guide vehicles to avoid the risk in advance.

[0018] Step 1.3: Construct the environmental condition influencing factor Kenv, taking into account the effects of ground adhesion coefficient, road curvature and visibility. Specifically, the environmental condition influence factor Kenv is defined as: ; in: The ground adhesion coefficient, The absolute value of the road curvature. Normalized visibility.

[0019] Step 1.4: Integrate the road geometric potential field, static obstacle risk field, and environmental condition influencing factors to generate the driving environment risk field: .

[0020] Step 2 specifically includes: Step 2.1: Construct an objective risk field by integrating spatial orientation influence factors, motion direction influence factors, equivalent mass, and vehicle condition influence factors. ; The spatial orientation influence factor This characterizes the impact of the obstacle vehicle's shape and orientation on risk distribution; when the obstacle vehicle travels longitudinally along the road (heading angle) When =0°: ; Where (x, y) are the coordinates of any point. , , where L and W are the x and y coordinates of the target vehicle's center of gravity, respectively, and L and W are dimensional coefficients related to the vehicle's length and width. , The longitudinal and lateral effective length corrections are given, K is the overall risk intensity adjustment factor, and v is the target vehicle speed. , Here, ε is the velocity sensitivity coefficient, and ε is the numerical smoothing term; When the obstacle vehicle's heading angle φo ≠ 0°, a coordinate transformation is required. Substitute into the formula This allows us to obtain the spatial orientation influence factor at any heading angle: ; The motion direction influencing factor This characterizes the characteristic that the risk in the area in front of the obstacle vehicle's direction of movement is higher than that in the area behind it: ; Where v is the speed of the obstacle vehicle, The velocity direction sensitivity coefficient, The angle between the spatial point and the direction of movement of the obstacle vehicle; The equivalent mass Taking into account the impact of obstacle mass, speed of movement, and other attributes on potential risks: ; Where m is the mass of the obstacle vehicle, T is the collision time coefficient, and v is the speed of the obstacle vehicle; Vehicle condition influencing factors The risk field intensity is adaptively adjusted based on vehicle type, dynamic state, and driving behavior. ; in, For vehicle type coefficients, , These are the lateral and longitudinal accelerations, respectively. , For acceleration influence coefficient, , Here, represents the standard deviation of the uncertainty in heading and speed, respectively, and σ is the uncertainty sensitivity coefficient. This is a driving behavior correction factor.

[0021] Step 2.2: Integrate the objective risk field, collision trend factor, and vehicle response factor to generate the vehicle risk field: ; Among them, collision trend factor This characterizes the impact of the relative motion direction between the vehicle and the obstacle vehicle on collision risk. ; in, For the speed of the obstacle vehicle, For the vehicle's speed, The angle between the direction of the obstacle vehicle's velocity and the line connecting them is... The angle between the direction of the vehicle's velocity and the line connecting them is... This is the collision trend sensitivity coefficient; The vehicle response factor This reflects the impact of the vehicle's mass and speed on the consequences of a collision. ; in, For the sake of vehicle quality, This is the collision time coefficient for the vehicle. This refers to the vehicle's speed.

[0022] Step 3 specifically includes:

[0023] Step 3.1: Based on the height attenuation coefficient and dynamic diffusion width, construct the single-trajectory prediction risk field: Specifically, a height attenuation coefficient is introduced. To describe how risk diminishes as the predicted trajectory extends, a dynamic diffusion width is introduced. Describes the cumulative effect of prediction uncertainty as the trajectory extends, at any point in space. The predicted risk field under the influence of trajectory Tm is: ; The height attenuation coefficient is: ; Among them, forward aiming distance The forward aiming time With current vehicle speed Determined, λ is the attenuation coefficient, and the trajectory length is: ; The dynamic diffusion width is: ; in Based on Gaussian width, is the diffusion growth coefficient.

[0024] Step 3.2: Obtain the multimodal predicted trajectories of surrounding vehicles. and corresponding confidence level Filter the valid trajectory subset The confidence levels are then normalized. A multimodal predicted trajectory risk field is generated by weighting and superimposing each single trajectory risk field according to its normalized confidence level. ; in, This represents the normalized confidence level.

[0025] Step 4 specifically includes: Vehicle risk field Predicting trajectory risk field and driving environment risk field Weighted fusion is performed to generate a comprehensive driving risk field: ; in, To determine the total field strength of the comprehensive driving risk field, , , These are the weighting coefficients corresponding to the vehicle risk field, the predicted trajectory risk field, and the driving environment risk field, respectively, where n is the number of surrounding vehicles.

[0026] Step 5 specifically includes: Step 5.1: Construct a three-dimensional spatiotemporal situation map, transforming the trajectory planning problem into an optimal path search problem on a discrete graph; and identify safe zones; Specifically, the three-dimensional search space is defined as S=X×Y×T, where X, Y, and T are the longitudinal, transverse, and temporal dimensions, respectively. The state transition reachable set satisfies the dynamic constraint function to ensure the physical executability of the planned trajectory. The primary security region is identified using dynamic thresholds. The primary security region is defined as follows: ; Dynamic security threshold Adaptively adjust based on overall traffic conditions and vehicle status: ; in, Based on the threshold, The current number of vehicles in the surrounding area. Based on the number of vehicles, The critical number of vehicles. For the vehicle's speed, The maximum permissible speed is represented by α and β, which are adjustment coefficients. When traffic is dense or high-speed, the threshold is lowered to make the planning more conservative; when traffic is sparse and low-speed, the threshold is raised to improve efficiency. Connectivity analysis is performed on the filtered safe regions, and connected regions that meet all constraints are integrated into reachable safe regions.

[0027] Step 5.2: Design a multi-objective cost function that integrates risk gradients; Specifically, a weighted summation method is used to transform the multi-objective problem into a scalar optimization problem: ; in, , , , These are the weighting coefficients for safety, efficiency, economy, and comfort, respectively. By integrating risk gradient information into the security cost model, proactive risk avoidance can be achieved. ; in, As a direct risk item, For gradient guiding terms, It is a second-order gradient term. Let λ represent the unit tangent vector of the trajectory at point s, and λ and μ be the weighting coefficients. Efficiency costs are used to establish a spatiotemporally coupled optimization model: ; in, This is a time efficiency term, measuring the degree of deviation between the actual driving speed and the expected speed; This is a spatial efficiency term that measures the deviation between the actual path and the reference path. , These are the weighting coefficients; Economic considerations: An energy consumption model is established based on vehicle dynamics principles. ; Total power consumption includes traction power and braking losses: ; Traction power includes rolling resistance, gradient resistance, air resistance, and acceleration resistance, while braking loss power takes into account the efficiency of the energy recovery system. ; The cost of comfort is assessed using a multi-dimensional evaluation model based on human physiology. ; in, To accelerate, Let be the radius of curvature. Angular acceleration, , , These are the weighting coefficients for longitudinal comfort, lateral comfort, and angular comfort, respectively.

[0028] Step 5.3: Based on the three-dimensional spatiotemporal situation map and the multi-objective cost function, a heuristic graph search method is used to solve the problem and obtain the original optimal trajectory. .

[0029] Step 5.4: Perform spatial path smoothing on the original optimal trajectory; Specifically, a quintic spline curve is used for spatial path smoothing to ensure C² continuity: ; in, For spline coefficients, The parameter is denoted as , and the optimization objective is to minimize the rate of change of curvature. ; The constraints are: ,in This is for spatial deviation tolerance.

[0030] Step 5.5: Replan the velocity profile based on the smooth path; output the executable trajectory; Specifically, the optimization objective for the velocity profile is: ; in, , , These are the weighting coefficients for time, acceleration, and jerk, respectively. For longitudinal acceleration, For longitudinal acceleration; The constraints include: (1) Velocity upper bound constraint: ; (2) Longitudinal dynamic constraints: ; (3) Time consistency constraint: ; in, For maximum speed, To limit the speed, κ(s) is the curvature. For maximum lateral acceleration, For the maximum longitudinal acceleration, εtime is the maximum jerk, and εtime is the time deviation tolerance. Output executable trajectory It contains position, velocity, and acceleration information.

[0031] The beneficial effects of this invention are as follows: An improved risk field model incorporating spatial orientation and motion direction influence factors is constructed, achieving accurate characterization of vehicle risk anisotropy. A height attenuation coefficient and dynamic diffusion width are introduced to construct a predicted trajectory risk field, effectively characterizing the cumulative effect of prediction uncertainty over time and space. A comprehensive driving risk field is constructed by fusing the vehicle risk field, predicted trajectory risk field, and driving environment risk field, achieving comprehensive risk assessment of dynamic traffic environments. An adaptive safe zone identification strategy is designed based on a three-dimensional spatiotemporal situation map and dynamic safety thresholds. A multi-objective cost function incorporating risk gradients is constructed, achieving synergistic optimization of safety, efficiency, comfort, and economy. This invention overcomes the shortcomings of traditional methods in risk anisotropy characterization, prediction uncertainty characterization, and multi-objective synergistic optimization, significantly improving trajectory planning performance in complex dynamic traffic environments and providing effective technical support for autonomous driving technology. Attached Figure Description

[0032] Figure 1 The system architecture diagram of a multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field provided by the present invention is shown below. Figure 2 This invention provides a two-dimensional distribution map of the driving environment risk field. Figure 3 The present invention provides a three-dimensional distribution map of the driving environment risk field; Figure 4 The objective risk field distribution map of the vehicle provided by this invention (a) v=0, =0; Figure 5 The objective risk field distribution diagram (b) provided by this invention has v=0. =45°; Figure 6 The objective risk field distribution map (c) provided by this invention has a v=5. =0; Figure 7 The objective risk field distribution map (d) provided by this invention has a v=5. =45°; Figure 8 The vehicle risk field distribution diagram (a) provided by the present invention has a vehicle heading angle of 0°. Figure 9 The vehicle risk field distribution diagram (b) provided by the present invention has a vehicle heading angle of 45°. Figure 10 The single-trajectory prediction risk field distribution map provided by this invention (a) shows a vehicle changing lanes to the left; Figure 11 The single-trajectory prediction risk field distribution map (b) provided by this invention shows continuous lane changes by vehicles; Figure 12 This invention provides a risk field distribution map for multimodal predicted trajectories. Figure 13 The comprehensive driving risk field strength distribution map provided by this invention; Figure 14 This is a schematic diagram of the trajectory optimization results provided by the present invention. Detailed Implementation

[0033] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. It should be noted that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] Please see Figures 1-14 This invention provides a method for optimizing the multi-objective trajectory of autonomous vehicles based on an improved risk field. This method can be applied to intelligent transportation systems, autonomous vehicles, and vehicle-road cooperative scenarios.

[0035] See Figure 1 This is a schematic diagram of a multi-objective trajectory optimization system architecture for autonomous vehicles based on an improved risk field, provided by an embodiment of the present invention. The present invention adopts a three-layer progressive architecture of "risk modeling-assessment-optimization," and the system includes two core modules: an improved risk field construction module and a multi-objective trajectory optimization module. The method mainly includes the following steps: Step 1: Construct the driving environment risk field, which integrates the road geometric potential field, the static obstacle risk field, and environmental condition influencing factors. The road geometric potential field includes the boundary potential field and the lane line potential field. The static obstacle risk field is modeled separately according to the geometric characteristics of the obstacles, which are divided into three categories: point, line, and region. Step 1 specifically includes: Step 1.1, Construct the road geometric potential field Including the boundary potential field Potential field with lane dividing line : ; The boundary potential field A piecewise exponential function is used to constrain vehicle movement within the road area and guide vehicles to travel along the center of the lane: ; in, Let y be the road width, y be the lateral position, and k1, k2, and k3 be the boundary potential field parameters.

[0036] The lane dividing lines create a risk "threshold" at the lane boundaries, preventing vehicles from crossing the lane dividing lines at will. ; in, This is the risk intensity coefficient for the dividing line. This represents the width of the risk distribution at the dividing line.

[0037] Step 1.2: Based on geometric features, static obstacles are categorized into three types: point-like, line-like, and region-like, and modeled separately to generate a static obstacle risk field. ; ; The risk field of the point-like obstacles, such as cones and warning signs, is as follows: ; in, The distance from a point in space to an obstacle. The risk intensity coefficient for point-like obstacles. The distance decay exponent, This is a numerical smoothing term.

[0038] The risk field of the linear obstacles, such as guardrails and construction barriers, is as follows: ; in, It represents the shortest distance from a point in space to the line segment. The risk intensity coefficient for linear obstacles. The distance decay exponent, This is a numerical smoothing term.

[0039] For obstacles in the area, such as construction areas and hazardous areas, the area with the highest risk value is designated as a no-entry zone, while the external risk decreases exponentially with distance. ; In the formula, This represents the regional risk intensity coefficient. This represents the risk attenuation coefficient outside the designated area. The area with the highest risk value is designated as a restricted zone, while external risks decrease exponentially with distance, enabling vehicles to avoid obstacles in advance.

[0040] Step 1.3, Constructing Environmental Condition Influencing Factors Taking into account the effects of ground adhesion coefficient, road curvature and visibility; Specifically, environmental condition influencing factors Defined as: ; in: The ground adhesion coefficient, The absolute value of the road curvature. Normalized visibility.

[0041] Step 1.4: Integrate the road geometric potential field, static obstacle risk field, and environmental condition influencing factors. (See [link / reference]). Figure 2 , Figure 3 Generate a driving environment risk field: ; Step 2: Construct a vehicle risk field, introduce spatial orientation and motion direction factors to characterize the anisotropic characteristics of vehicle risk, combine equivalent mass and vehicle condition factors to construct an objective risk field, and integrate collision trend factors and vehicle response factors to realize the transformation of objective risk into effective risk. Step 2 specifically includes: Step 2.1, integrate the spatial orientation influence factor, motion direction influence factor, equivalent mass, and vehicle condition influence factor, see [link / reference]. Figure 4 , Figure 5 , Figure 6 and Figure 7 Construct an objective risk field: ; The spatial orientation influence factor This characterizes the impact of the obstacle vehicle's shape and orientation on risk distribution; when the obstacle vehicle travels longitudinally along the road (heading angle) When =0°: ; Where (x, y) are the coordinates of any point. , , where L and W are the x and y coordinates of the target vehicle's center of gravity, respectively, and L and W are dimensional coefficients related to the vehicle's length and width. , The longitudinal and lateral effective length corrections are given, K is the overall risk intensity adjustment factor, and v is the target vehicle speed. , ε is the velocity sensitivity coefficient, and ε is the numerical smoothing term.

[0042] When the obstacle vehicle's heading angle When the angle is ≠ 0°, a coordinate transformation is required. Substitute into the formula This allows us to obtain the spatial orientation influence factor at any heading angle: ; The motion direction influencing factor This characterizes the characteristic that the risk in the area in front of the obstacle vehicle's direction of movement is higher than that in the area behind it: ; Where v is the speed of the obstacle vehicle, The velocity direction sensitivity coefficient, The angle between the spatial point and the direction of movement of the obstacle vehicle.

[0043] The equivalent mass Taking into account the impact of obstacle mass, speed of movement, and other attributes on potential risks: ; Where m is the mass of the obstacle vehicle, T is the collision time coefficient, and v is the speed of the obstacle vehicle.

[0044] Vehicle condition influencing factors The risk field intensity is adaptively adjusted based on vehicle type, dynamic state, and driving behavior. ; in, For vehicle type coefficients, , These are the lateral and longitudinal accelerations, respectively. , For acceleration influence coefficient, , Here, represents the standard deviation of the uncertainty in heading and speed, respectively, and σ is the uncertainty sensitivity coefficient. This is a driving behavior correction factor.

[0045] Step 2.2, integrate the objective risk field, collision trend factor, and vehicle response factor, see [link / reference]. Figure 8 , Figure 9 Generate vehicle risk field: ; Among them, collision trend factor This characterizes the impact of the relative motion direction between the vehicle and the obstacle vehicle on collision risk. ; in, For the speed of the obstacle vehicle, For the vehicle's speed, The angle between the direction of the obstacle vehicle's velocity and the line connecting them is... The angle between the direction of the vehicle's velocity and the line connecting them is... This represents the collision trend sensitivity coefficient.

[0046] The vehicle response factor This reflects the impact of the vehicle's mass and speed on the consequences of a collision. ; in, For the sake of vehicle quality, This is the collision time coefficient for the vehicle. This refers to the vehicle's speed.

[0047] Step 3: Construct a predicted trajectory risk field. Based on the multimodal predicted trajectories of surrounding vehicles and their confidence levels, introduce a height attenuation coefficient and dynamic diffusion width to transform the predicted trajectory into a spatiotemporal risk distribution. A forward-looking risk assessment is achieved through confidence-weighted fusion. Step 3 specifically includes: Step 3.1: Construct a single-trajectory prediction risk field based on the height attenuation coefficient and dynamic diffusion width.

[0048] Specifically, a height attenuation coefficient is introduced. To describe how risk diminishes as the predicted trajectory extends, a dynamic diffusion width is introduced. Describe the cumulative effect of prediction uncertainty as the trajectory extends. (Any point in space) See Figure 10 , Figure 11 The predicted risk field under the influence of trajectory Tm is: ; The height attenuation coefficient is: ; Among them, forward aiming distance The forward aiming time With current vehicle speed Let λ be the attenuation coefficient, and the trajectory length be: .

[0049] The dynamic diffusion width is: ; Based on Gaussian width, is the diffusion growth coefficient.

[0050] Step 3.2: Obtain the multimodal predicted trajectories of surrounding vehicles. and corresponding confidence level Filter the valid trajectory subset The confidence levels are then normalized. A multimodal predicted trajectory risk field is generated by weighting and superimposing the risk fields of each individual trajectory according to their normalized confidence levels. (See [link to relevant documentation]). Figure 12 : ; in, This represents the normalized confidence level.

[0051] Step 4: Construct a comprehensive driving risk field by weighting and fusing the vehicle risk field, the predicted trajectory risk field, and the driving environment risk field to obtain a unified comprehensive driving risk field; Step 4 specifically includes: See Figure 13 Vehicle risk field Predicting trajectory risk field and driving environment risk field Weighted fusion is performed to generate a comprehensive driving risk field: ; in, To determine the total field strength of the comprehensive driving risk field, , , These are the weighting coefficients corresponding to the vehicle risk field, the predicted trajectory risk field, and the driving environment risk field, respectively, where n is the number of surrounding vehicles.

[0052] Step 5: Multi-objective trajectory optimization based on 3D spatiotemporal situation map. Construct a 3D spatiotemporal situation map to transform the trajectory planning problem into an optimal path search problem on a discrete graph. Identify safe areas through dynamic thresholds, design a multi-objective cost function that integrates risk gradients to achieve coordinated optimization of safety, efficiency, comfort and economy. Use a heuristic graph search method to solve for the optimal trajectory and output an executable trajectory after smoothing.

[0053] Step 5 specifically includes: Step 5.1: Construct a three-dimensional spatiotemporal situation map, transforming the trajectory planning problem into an optimal path search problem on a discrete map; and identify safe areas.

[0054] Specifically, the three-dimensional search space is defined as S = X × Y × T, where X, Y, and T represent the vertical, horizontal, and temporal dimensions, respectively. The reachable set of the state transition satisfies the dynamic constraint function to ensure the physical executability of the planned trajectory.

[0055] The primary security region is identified using dynamic thresholds. The primary security region is defined as follows: ; Dynamic security threshold Adaptively adjust based on overall traffic conditions and vehicle status: ; in, Based on the threshold, The current number of vehicles in the surrounding area. Based on the number of vehicles, The critical number of vehicles. For the vehicle's speed, The threshold represents the maximum permissible speed, and α and β are adjustment coefficients. Lowering the threshold during periods of dense traffic or high speeds makes the planning more conservative; raising the threshold during periods of sparse traffic and low speeds improves efficiency.

[0056] Connectivity analysis is performed on the filtered safe regions, and connected regions that meet all constraints are integrated into reachable safe regions.

[0057] Step 5.2: Design a multi-objective cost function that integrates risk gradients; Specifically, a weighted summation method is used to transform the multi-objective problem into a scalar optimization problem: ; in, , , , These are the weighting coefficients for safety, efficiency, economy, and comfort, respectively.

[0058] By integrating risk gradient information into the security cost model, proactive risk avoidance can be achieved. ; in, As a direct risk item, For gradient guiding terms, It is a second-order gradient term. Let λ represent the unit tangent vector of the trajectory at point s, and let λ and μ be the weighting coefficients.

[0059] Efficiency costs are used to establish a spatiotemporally coupled optimization model: ; in, This is a time efficiency term, measuring the degree of deviation between the actual driving speed and the expected speed; This is a spatial efficiency term that measures the deviation between the actual path and the reference path. , These are the weighting coefficients.

[0060] Economic considerations: An energy consumption model is established based on vehicle dynamics principles. ; Total power consumption includes traction power and braking losses: ; Traction power includes rolling resistance, gradient resistance, air resistance, and acceleration resistance, while braking loss power takes into account the efficiency of the energy recovery system. .

[0061] The cost of comfort is assessed using a multi-dimensional evaluation model based on human physiology. ; in, To accelerate, Let be the radius of curvature. Angular acceleration, , , These are the weighting coefficients for longitudinal comfort, lateral comfort, and angular comfort, respectively.

[0062] Step 5.3: Based on the three-dimensional spatiotemporal situation map and the multi-objective cost function, a heuristic graph search method is used to solve the problem and obtain the original optimal trajectory. .

[0063] Step 5.4: Perform spatial path smoothing on the original optimal trajectory; Specifically, a quintic spline curve is used for spatial path smoothing to ensure C² continuity: ; in, For spline coefficients, The parameter is denoted as . The optimization objective is to minimize the rate of change of curvature. ; The constraints are: ,in This is for spatial deviation tolerance.

[0064] Step 5.5: Replan the velocity profile based on the smooth path; output the executable trajectory.

[0065] Specifically, the optimization objective for the velocity profile is: ; in, , , These are the weighting coefficients for time, acceleration, and jerk, respectively. For longitudinal acceleration, This refers to longitudinal acceleration.

[0066] The constraints include: (1) Velocity upper bound constraint: ; (2) Longitudinal dynamic constraints: ; (3) Time consistency constraint: ; in, For maximum speed, To limit the speed, κ(s) is the curvature. For maximum lateral acceleration, For the maximum longitudinal acceleration, εtime represents the maximum jerk, and εtime represents the time deviation tolerance.

[0067] See Figure 14 Output executable trajectory It contains position, velocity, and acceleration information.

[0068] The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field provided in this embodiment achieves efficient trajectory planning for autonomous vehicles in complex dynamic traffic environments by improving the construction of the risk field, fusing comprehensive driving risk fields, searching for three-dimensional spatiotemporal situation maps, and optimizing multi-objective cost functions.

[0069] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0071] The parts of this invention not described in detail are prior art.

Claims

1. A multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field, characterized in that, Includes the following steps: Step 1: Construct the driving environment risk field, which integrates the road geometric potential field, the static obstacle risk field, and environmental condition influencing factors. The road geometric potential field includes the boundary potential field and the lane dividing line potential field. The static obstacle risk field is divided into three categories based on the geometric characteristics of the obstacles: point, line, and region, and modeled separately. Step 2: Construct a vehicle risk field, introduce spatial orientation and motion direction factors to characterize the anisotropic characteristics of vehicle risk, combine equivalent mass and vehicle condition factors to construct an objective risk field, and integrate collision trend factors and vehicle response factors to realize the transformation of objective risk into effective risk. Step 3: Construct a predicted trajectory risk field. Based on the multimodal predicted trajectories of surrounding vehicles and their confidence levels, introduce a height attenuation coefficient and dynamic diffusion width to transform the predicted trajectory into a spatiotemporal risk distribution. A forward-looking risk assessment is achieved through confidence-weighted fusion. Step 4: Construct a comprehensive driving risk field by weighting and fusing the vehicle risk field, the predicted trajectory risk field, and the driving environment risk field to obtain a unified comprehensive driving risk field; Step 5: Multi-objective trajectory optimization based on 3D spatiotemporal situation map. Construct a 3D spatiotemporal situation map to transform the trajectory planning problem into an optimal path search problem on a discrete graph. Identify safe areas through dynamic thresholds, design a multi-objective cost function that integrates risk gradients to achieve coordinated optimization of safety, efficiency, comfort and economy. Use a heuristic graph search method to solve for the optimal trajectory and output an executable trajectory after smoothing.

2. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field as described in claim 1, characterized in that, In step 1: The boundary potential field adopts a piecewise exponential function form, which is constrained according to the lateral position relative to the road boundary and lane line, and guides the vehicle to travel along the center of the lane. The lane dividing line potential field forms a risk "threshold" to prevent vehicles from crossing the lane dividing line at will; The point obstacle risk field adopts a distance inverse decay model; the linear obstacle risk field is modeled based on the shortest distance from a point to a line segment; the regional obstacle risk field is set to the maximum risk value of the restricted area inside the region, and decays exponentially with distance outside the region. The environmental condition influencing factors comprehensively consider the effects of ground adhesion coefficient, road curvature, and visibility.

3. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, In step 2: The spatial orientation influence factor is based on the size, position and heading angle of the obstacle vehicle. An anisotropic distribution function is used to characterize the influence of the vehicle shape on the spatial distribution of risk. When the heading angle is non-zero, a coordinate rotation transformation is performed. The motion direction influence factor adopts an exponential function form to characterize the characteristic that the risk in the area in front of the obstacle vehicle's motion direction is higher than that in the area behind it. The equivalent mass takes into account the impact of the obstacle vehicle's mass and speed on the collision consequences. The vehicle condition influencing factor adaptively adjusts the risk field intensity based on vehicle type, lateral and longitudinal acceleration, state uncertainty, and driving behavior. The collision trend factor quantifies the collision risk trend based on the relative motion direction of the vehicle and the obstacle vehicle; The vehicle response factor reflects the impact of vehicle mass and speed on the consequences of a collision.

4. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, In step 3: The height attenuation coefficient describes how the risk intensity decreases exponentially as the ratio of the predicted trajectory length to the forward-looking aiming distance increases. The dynamic diffusion width describes the cumulative effect of prediction uncertainty as the trajectory extends, and its value increases linearly with the trajectory length. The multimodal predicted trajectory risk field is obtained by screening effective predicted trajectories, normalizing the confidence scores, and then weighting and superimposing the risk fields of each single trajectory according to the normalized confidence scores.

5. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, The expression for the comprehensive driving risk field in step 4 is: ； in , , These are the weighting coefficients. For vehicle risk areas, To predict the trajectory risk field, This is a high-risk driving environment.

6. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, Step 5, which involves constructing a three-dimensional spatiotemporal situation map, includes: The three-dimensional search space is defined as the Cartesian product of the vertical, horizontal, and time dimensions, and the state transition satisfies the vehicle dynamics constraints. Primary safety zones are identified by dynamic safety thresholds, which are adaptively adjusted based on the number of surrounding vehicles and the speed of the vehicle itself. When traffic is dense or high-speed, the threshold is lowered to make the planning more conservative, and when traffic is sparse and low-speed, the threshold is raised to improve efficiency. Perform connectivity analysis on the primary security zones and integrate them into reachable security zones.

7. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, The multi-objective cost function in step 5 is designed as follows: ； in, , , , These are the weighting coefficients for safety, efficiency, economy, and comfort, respectively. The security cost integrates risk field value, risk gradient, and second-order gradient information to achieve proactive risk avoidance. The efficiency cost includes a time efficiency term and a space efficiency term, which respectively measure the deviation between the actual speed and the expected speed, and the deviation between the actual path and the reference path. The economic cost is based on an energy consumption model established by vehicle dynamics, including traction power and braking loss power considering energy recovery efficiency; The comfort cost is comprehensively assessed based on the effects of longitudinal jerk, lateral acceleration, and angular acceleration.

8. The multi-objective trajectory optimization method for autonomous vehicles based on an improved risk field according to claim 1, characterized in that, Step 5, which involves finding the optimal trajectory and smoothing the process, includes: Based on the three-dimensional spatiotemporal situation map and multi-objective cost function, a heuristic graph search method is used to solve for the original optimal trajectory; The original trajectory is smoothed spatially using a quintic spline curve, with the goal of minimizing the rate of curvature change, ensuring C² continuity, and satisfying spatial deviation tolerance constraints. Velocity profile optimization is performed based on a smooth path, taking into account the weighted costs of time, acceleration, and jerk, while satisfying the upper bound constraint of velocity, longitudinal dynamics constraint, and time consistency constraint. The output is an executable trajectory containing position, velocity, and acceleration information.

Images