A method for modeling vehicle micro-behavior in intersection scenarios

Abstract

This invention relates to the cross-disciplinary integration of autonomous driving and smart tourism, and discloses a method for modeling the microscopic behavior of vehicles in intersection scenarios. The method includes the following steps: performing dynamic light projection from the driver's viewpoint to extract the line-of-sight cutoff interface and determine the blind spot range; constructing a topologically anisotropic cognitive entropy flow field based on road lane flow direction to simulate the non-uniform distribution of risk along the lanes; introducing a second-order line-of-sight coupling mechanism, combining the visible range of neighboring vehicles and behavioral confidence to calculate a trust inhibition factor, and dynamically correcting the basic entropy field; constructing a composite objective function containing physical operating costs and cognitive entropy costs, and using model predictive control to solve for the optimal control sequence. This invention can transform the potential risks of blind spots into calculable cognitive loads, endowing autonomous vehicles with human-like defensive driving capabilities and social game intelligence in scenarios with obstructed vision, balancing driving safety and traffic efficiency.

Description

Technical Field

[0001] This invention relates to the field of cross-border integration technology of autonomous driving and smart cultural tourism, specifically a method for modeling the micro-behavior of vehicles in intersection scenarios. Background Technology

[0002] With the rapid development of autonomous driving technology, environmental perception and decision-making have become core components for achieving fully autonomous driving. In complex urban road traffic scenarios, autonomous vehicles often face the problem of line-of-sight obstruction caused by static obstacles or dynamic traffic flow. Vehicle sensors, limited by physical line-of-sight, cannot directly acquire environmental information behind obstructed areas, thus creating perception blind spots. How to make safe and efficient trajectory decisions under conditions of obstructed vision is one of the most serious challenges currently facing the deployment of autonomous driving technology.

[0003] As "intelligent connected self-driving tours" become a new trend in cultural tourism consumption, autonomous vehicles are frequently entering unstructured roads around historical cities and mountain scenic areas. These areas typically feature complex road conditions, severe vegetation obstruction, and irregular intersections. However, as "outsiders," these autonomous vehicles, lacking familiarity with the local road conditions, often adopt extremely conservative strategies when facing blind spots, frequently braking abruptly or remaining stationary for extended periods. This not only reduces road traffic efficiency but also severely impacts the smoothness of sightseeing and the comfort of passengers, easily triggering motion sickness or anxiety.

[0004] To address these issues, some existing technologies attempt to model perceived uncertainty using partially observable Markov decision processes (MIDPMs) in an effort to find the optimal strategy within the probability space. However, MIDPMs suffer from a severe state-space explosion problem; their computational load increases exponentially with scene complexity, making it difficult to meet the stringent millisecond-level real-time requirements of in-vehicle computing platforms. On the other hand, while risk assessment methods based on artificial potential fields offer high computational efficiency, existing potential field models typically construct isotropic repulsive force fields based on Euclidean distance, assuming risk spreads uniformly in all directions. This isotropic assumption ignores the constraints of road structure on risk propagation, failing to accurately reflect the physical law that risk should extend along the lane flow direction. This results in a risk field that does not match the actual traffic flow characteristics, leading to abrupt vehicle avoidance maneuvers or unreasonable path planning.

[0005] Furthermore, most existing blind spot decision-making schemes treat the vehicle as an isolated perceptual entity, neglecting the potential value of other road users as "social sensors" in traffic scenarios. In the game-theoretic behavior of human drivers, drivers often indirectly infer the safety situation in the blind spot by observing the behavior of nearby vehicles with a wider field of vision. However, current technologies lack this social reasoning mechanism based on second-order line-of-sight coupling, and cannot effectively utilize the implicit interaction information of surrounding vehicles to dynamically correct the perception of blind spot risks. This results in autonomous vehicles lacking human-like flexibility and environmental adaptability when facing blind spots. Therefore, there is an urgent need for an autonomous driving blind spot decision-making method that can take into account road topology characteristics, possess social reasoning capabilities, and be computationally efficient. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for modeling the micro-behavior of vehicles in intersection scenarios. This method solves the technical problem that autonomous vehicles make overly conservative decisions and seriously affect the travel experience of tourists in unfamiliar tourist road environments due to obstructed visibility and lack of local road experience.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for modeling the microscopic behavior of vehicles in intersection scenarios, the method comprising the following steps:

[0008] Step S1: Obtain the static topology data of the intersection scene and the motion status information of all vehicles in the scene at the current moment. The static topology data includes lane centerlines, lane connectivity, and static obstacle outlines; the vehicle motion status information includes position, speed, acceleration, and heading angle.

[0009] Step S2: Perform dynamic ray projection with the driver's viewpoint as the origin. Based on the intersection results of the rays and obstacles in the scene, determine the visible area at the current moment and extract the line-of-sight cutoff interface that distinguishes the visible area from the blind spot. Specifically, this step involves emitting discrete rays within the driver's field of vision and calculating the nearest intersection point of each ray with a static obstacle or dynamic vehicle bounding box. If the effective length of a ray is less than the maximum set viewing distance, it is determined that there is occlusion in the direction of that ray. The set of terminal intersection points of all occluded rays is defined as the line-of-sight cutoff interface, which dynamically divides the intersection space into a defined visible area and an uncertain blind spot.

[0010] Step S3: Based on the relationship between the road lane flow direction and the spatial distribution of blind spots, a topological anisotropic cognitive entropy flow field is constructed in the blind spot after the line of sight cutoff interface. The topological anisotropic cognitive entropy flow field is used to characterize the potential risk uncertainty of the spatial location in the blind spot.

[0011] To address the shortcomings of traditional models that simply assume no risk in the blind zone or a uniform risk distribution, this invention employs the following innovative mechanism when constructing the topological anisotropic cognitive entropy flow field:

[0012] First, identify spatial points within the blind spot. Identify the potential lanes that belong to or are adjacent to, and obtain the lane at the point. Tangent vector at point Spatial points relative to the view cutoff interface The relative position is decomposed into longitudinal projected distances along the tangent vector direction. and the lateral projection distance perpendicular to the tangent vector direction .

[0013] Calculation of spatial points based on Gaussian decay model Topological fundamental entropy The specific formula is as follows:

[0014]

[0015] In the formula, For the relevant lane set, This is the base entropy magnitude. Set the longitudinal attenuation coefficient. Greater than the transverse attenuation coefficient This causes the generated virtual artifact morphology to be constrained by the lane geometry and topology, exhibiting anisotropic characteristics extending along the lane.

[0016] Furthermore, to simulate the social interaction mechanism by which drivers use the behavior of neighboring vehicles to infer risk, this step also includes introducing a second-order line-of-sight coupling mechanism to dynamically correct the field strength. The trust inhibition factor is then calculated. :

[0017]

[0018] In the formula, This refers to the group of other nearby vehicles within the vehicle's field of vision. As an indicator function, when a specific area within the blind zone... Located in the adjacent car The value is 1 when it is within the visible area; For neighboring cars Behavioral confidence; This is the social impact weighting coefficient.

[0019] The final synthesized spatial cognitive entropy field for:

[0020]

[0021] This mechanism ensures that when nearby vehicles have a clear view of the blind spot and maintain stable driving behavior, the entropy field intensity within the blind spot can be significantly reduced.

[0022] Step S4: Construct a trajectory decision objective function that includes physical operation cost and cognitive entropy cost. Solve for the future control sequence of the vehicle by minimizing the objective function. The cognitive entropy cost is calculated based on the cumulative intensity of the vehicle's planned trajectory in the topological anisotropic cognitive entropy flow field.

[0023] The trajectory decision objective function Defined as:

[0024]

[0025] Among them, physical operation cost This includes velocity tracking error, acceleration smoothness, and obstacle avoidance penalties. Cognitive entropy cost. Specifically, it is defined as the geometric region covered by the planned trajectory of the vehicle in the prediction time domain. The sum of integrals in the topologically anisotropic cognitive entropy flow field:

[0026]

[0027] By using model predictive control to solve the above objective function, when the entropy flow field intensity in the blind zone is high, the cognitive entropy cost term will drive the solver to output deceleration or lateral offset commands, thereby enabling the vehicle to exhibit defensive driving behavior that actively avoids areas of high uncertainty.

[0028] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0029] The technical solution provided by this invention accurately captures the blind spot boundary from the driver's perspective using light projection technology, and constructs a cognitive entropy flow field with anisotropic characteristics by combining road topology, making the expression of blind spot risk conform to the physical extension law of traffic flow. Simultaneously, a second-order line-of-sight coupling mechanism is introduced to quantify the dynamic impact of neighboring vehicle behavior on the driver's risk perception. Finally, through a unified cost function, physical collision avoidance and information-based risk avoidance are integrated into the same decision-making framework, which can more realistically reflect the micro-behavioral characteristics of vehicles in intersection scenarios with limited visibility and complex game dynamics.

[0030] This invention provides a method for modeling the microscopic behavior of vehicles in intersection scenarios. It has the following beneficial effects:

[0031] 1. Overcoming the bottleneck of behavioral modeling under obstructed vision and improving simulation realism. This invention uses dynamic ray projection technology to extract the visual obstruction interface in real time, and constructs a heterogeneous spatial model that distinguishes between the "visible zone" and the "blind zone". Compared with traditional simulation methods that rely solely on geometric distance, this solution gives the virtual vehicle the perceptual limitation of "not being able to see". This allows the model to spontaneously generate defensive driving behaviors such as hesitation and peeking due to obstructed vision, which greatly improves the realism of micro-behavioral simulation in complex intersection scenarios, especially in scenarios with large vehicles mixed in traffic.

[0032] 2. Utilizing a topological anisotropic entropy field to accurately quantify blind spot risk distribution. This invention innovatively combines road topology to construct a cognitive entropy flow field. By using differentiated settings of longitudinal and lateral attenuation coefficients, the virtual risks within the blind spot exhibit a "fluid" distribution extending along the lane. This mechanism avoids the geometric mismatch problem of traditional circular field in road scenarios, ensuring that vehicles are only alert to potential oncoming vehicles in the direction of lane extension, while ignoring high-entropy interference from irrelevant directions such as roadside buildings. This achieves a precise mathematical replication of the driver's spatial risk perception logic.

[0033] 3. Introducing a second-order line-of-sight coupling mechanism to reproduce socialized interactive reasoning capabilities. By establishing a trust inhibition factor based on the neighboring vehicle's field of vision, this invention, for the first time, endows vehicles with the ability to indirectly perceive using "the eyes of others" in a micro-behavioral model. When the model detects that a neighboring vehicle has a clear view of the blind spot and is confident in passing through, it automatically reduces the risk assessment of that area. This socialized cognitive mechanism effectively solves the problem of traditional obstacle avoidance algorithms being too conservative, leading to low intersection traffic efficiency, and realistically reproduces the group trust and herd mentality of human drivers in congestion games.

[0034] 4. Constructing a unified physical and cognitive decision-making framework to achieve end-to-end closed-loop control. This invention incorporates the cognitive entropy cost representing uncertainty and the physical operational cost representing kinematic characteristics into the same objective function, performing multi-objective collaborative optimization within a model predictive control framework. This design breaks down the separation between the perception layer and the decision-making layer, enabling vehicles to automatically balance the needs of "safety avoidance" and "efficient passage" during continuous mathematical optimization without relying on complex rule bases, significantly improving the model's generalization ability and robustness under unknown conditions.

[0035] 5. Supports extreme scenario testing and verification of autonomous driving algorithms. Thanks to the high-fidelity simulation of blind spot risks and irrational game behavior, the model constructed in this invention can serve as an advanced background traffic flow generator for virtual testing of autonomous driving. It can generate adversarial traffic flows with high uncertainty and cognitive intelligence, providing a highly challenging extreme testing environment for the perception blind spot handling capabilities and game planning algorithms of autonomous driving systems, thereby helping to improve the safety of real-world autonomous driving deployments. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0037] Figure 2 This is a schematic diagram illustrating the principle of dynamic field-of-view ray projection and line-of-view interception interface extraction of the present invention;

[0038] Figure 3 This is a schematic diagram illustrating the generation principle and field strength distribution of the topological anisotropic cognitive entropy flow field of the present invention.

[0039] Figure 4 This is a schematic diagram illustrating the second-order line-of-sight coupling mechanism and the calculation principle of the trust inhibition factor of the present invention.

[0040] Figure 5 This is a flowchart of the trajectory planning and decision-making closed-loop method based on model predictive control of the present invention.

[0041] Figure 6 This is a schematic diagram of the hardware structure of the electronic device of the present invention. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please see the appendix Figures 1 to 6 This invention provides a method for modeling the micro-behavior of vehicles in intersection scenarios. It should be noted that although this embodiment uses the behavior modeling of general autonomous vehicles in intersection scenarios as an example for detailed explanation, this method is particularly applicable to cross-regional autonomous driving and unmanned shuttle services in scenic areas in smart cultural tourism scenarios.

[0044] In such tourism application scenarios, the "vehicles" mentioned below can specifically refer to autonomous passenger vehicles or park sightseeing vehicles carrying tourists; the "intersections" and "obstacles" mentioned below can specifically refer to complex intersections in unfamiliar tourist cities, corners of ancient town streets, or unstructured road nodes in scenic areas that are obscured by vegetation and landscape.

[0045] This invention utilizes the following topological entropy field and social game mechanism to solve the problem of conservative decision-making and poor experience caused by obstructed vision in unfamiliar and complex environments for out-of-town tourist vehicles or sightseeing vehicles.

[0046] This method first requires constructing a semantically enabled digital twin environment for the intersection and initializing vehicle states, specifically including the following steps:

[0047] Step S1: Obtain the static topology data of the intersection scene and the motion status information of all vehicles in the scene at the current moment.

[0048] In this embodiment, the intersection scenario is abstracted as a directed graph structure containing geometric features and topological relationships. The system first loads high-precision map data, parses it, and constructs a static road network model. .

[0049] in, This represents the set of lane centerlines. Each lane... It is parameterized as a series of discrete reference waypoint sequences. Each waypoint contains global location coordinates. and the lane tangential azimuth at that point This parameterized representation not only describes the geometry of the road but also implies the permitted direction of traffic flow, providing a reference coordinate system for the subsequent construction of anisotropic entropy fields.

[0050] This represents the set of connectivity relationships between lanes, defining the preceding, succeeding, and adjacent relationships of each lane. This topological connectivity information is used in simulations to quickly retrieve potential driving paths of vehicles and determine the specific flow direction of blind spot risk propagation along the road network.

[0051] This represents the set of static, impassable obstacles in the scene, including roadside green belts, building walls, traffic islands, and pillars. These obstacles are stored as polygonal outlines and used as input for occlusion in subsequent ray casting algorithms.

[0052] After loading the static environment, the system instantiates the traffic participants in the scene. For any target vehicle in the simulation scene... A kinematic model is established. Considering that vehicles typically travel at low to medium speeds and experience minimal lateral slippage in intersection scenarios, this embodiment employs a kinematic bicycle model to describe the vehicle's motion evolution.

[0053] At any simulation moment ,vehicle System state vector The definition is as follows:

[0054]

[0055] The physical meaning and definition of each symbol in this state vector are as follows:

[0056] and : Indicates the horizontal and vertical coordinates of the center point of the vehicle's rear axle in the global Cartesian coordinate system;

[0057] : Indicates the longitudinal speed of travel at the center of the rear axle of the vehicle;

[0058] : Represents the heading angle of the vehicle body, defined as the angle between the vehicle body's longitudinal axis and the X-axis of the global coordinate system;

[0059] : Indicates the front wheel steering angle, which is constrained by the vehicle's mechanical steering limit.

[0060] Accordingly, the vehicle's control input vector Defined as:

[0061]

[0062] in, For longitudinal acceleration, This is the rate of change of the front wheel steering angle, i.e., the steering angular velocity.

[0063] Based on the above definition, the time update of the vehicle state follows the following discrete-time state equation:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] In the formula, The time step of the simulation system, This refers to the vehicle's wheelbase.

[0070] In addition to its own motion state, the system also needs to acquire the bounding box information of all other dynamic obstacles in the scene in real time. Each dynamic obstacle... It is represented as a oriented rectangular bounding box. Its parameters include center coordinates, length and width dimensions, and current heading angle. The bounding box set of all dynamic obstacles. With static obstacle set Together, they form the input source for the subsequent visual computing module, used to determine whether the line of sight is obstructed.

[0071] Through the above steps, this invention constructs a simulation base environment containing accurate road topology semantics and complete vehicle dynamics description, enabling subsequent visual perception and cognitive reasoning algorithms to perform calculations based on real and physically consistent data.

[0072] In this invention, to simulate the perceptual limitations of human drivers caused by physical environmental occlusion, instead of directly providing the vehicle model with global God's-eye view data, a local visual perception model centered on the driver is constructed. First, the driver's viewpoint position is defined. This location is based on the vehicle's centroid coordinates. and the driver's relative offset in the vehicle coordinate system Perform the calculations and transform to the global coordinate system using a rotation matrix:

[0073]

[0074] Set the driver's horizontal field of view to The maximum effective line of sight is In the simulation calculation, the continuous field of view is discretized into... A scanning ray. (Number) ray launch direction angle From the current vehicle heading angle and scanning index Joint decision:

[0075]

[0076] For each generated ray The algorithm performs an intersection test to find the nearest obstruction point on the ray path. The intersection object includes a set of static obstacles in the scene. and the collection of all dynamic traffic participants other than this vehicle. The bounding box.

[0077] For the A ray, whose corresponding ray equation is defined as: ,in It is a unit direction vector. The distance parameter is used. The set of intersections between the ray and all obstacle boundary segments is calculated, and the distance viewpoint is selected from this set. recent and The effective intersection point. Let the distance parameter at the nearest intersection point be... .

[0078] If at the maximum viewing distance No intersections were detected within the range, i.e. If the line of sight is clear in that direction, then the effective viewing distance is determined. The corresponding terminal point is denoted as the "virtual infinity point".

[0079] If an intersection is detected, that is If the ray is blocked by an obstacle, then the effective line of sight is determined to be limited. At this point, the actual point of impact of the ray on the surface of the obstacle. The coordinates are calculated as follows:

[0080]

[0081] By iterating through all A ray was used to obtain an ordered set of terminal points. These points, connected in angular order, form the boundary of the driver's visible polygon at the current moment. The area inside the visible polygon is the defined visible area. Vehicle behavior within this area is considered fully observable.

[0082] Building upon this, the key aspect of this embodiment lies in extracting the "line-of-sight cutoff interface" used to generate the subsequent risk field. Not all ray termination points constitute a line-of-sight cutoff interface; only those intersections that fall on the surface of an obstacle and prevent the line of sight from penetrating have occlusion properties.

[0083] The system categorizes and labels the set of endpoints. For each endpoint... Check whether it is located on the physical boundary of a static obstacle or a dynamic vehicle. If If the point of impact is located on the boundary of an entity, then that point is marked as an "occlusion point". Connecting consecutive adjacent occlusion points forms a set of line segments that interrupt the line of sight. .

[0084] Each line segment cut off This represents a "window" that cannot be seen through, and the fan-shaped area behind it is the blind spot. Compared to simple Boolean visibility checks, this method extracts the specific view cutoff interface. The ability to clearly define the geometric starting boundary of the blind zone provides a precise geometric benchmark for calculating the decay gradient of the risk field extending backward from this boundary in subsequent steps. For example, when a large truck obstructs the view, the extracted cutoff interface is the outline segment of the side or rear of the truck, and the subsequent cognitive entropy field will spread backward into the blind zone from this segment.

[0085] In this invention, to address the uncertainty of risks within blind spots, the commonly used Euclidean distance isotropic attenuation model in traditional potential field methods is abandoned. Instead, an anisotropic computation mechanism constrained by road topology is proposed. The core idea of ​​this mechanism is that the driver's fear of blind spots is not uniformly diffused, but rather has a greater psychological projection distance along the lane extension direction, while in the direction perpendicular to the lane it is subject to psychological constraints from structures such as curbs.

[0086] Specifically, for blind spots Any spatial sampling point within The system first retrieves the nearest potential lane to the given point. Lanes are obtained through road network topology data. At point Unit tangent vector at the projection ,in This represents the lane flow direction angle at that point. Simultaneously, the calculation is performed with... Perpendicular unit normal vector .

[0087] Next, calculate the spatial points. To the corresponding view cutoff interface geometric distance vector To accurately quantify the anisotropy of risk propagation, vectors will be used. Decomposed into components along the lane tangent and the component along the lane normal The decomposition calculation formula is as follows:

[0088]

[0089]

[0090] In the formula, " " indicates the vector dot product operation. It represents the depth of the blind spot, which is the distance that a potentially at-risk vehicle travels along the lane. It characterizes lateral deviation, that is, the degree to which a potential risk deviates from the center of the lane.

[0091] Based on the above decomposition distance, an improved Gaussian kernel function is used to construct points. Topological fundamental entropy at the location This entropy value represents the basic cognitive stress caused solely by environmental geometric occlusion, without any additional social interaction information.

[0092]

[0093] in, The set basic entropy amplitude constant represents the maximum uncertainty intensity at the line-of-sight cutoff interface (i.e., the starting edge of the blind zone).

[0094] In the formula and These are the spatial attenuation coefficients for the longitudinal and lateral directions, respectively. In this embodiment, they are set as follows: This parameter differentiation setting has a clear physical meaning: larger... This causes the entropy field intensity to decay slowly along the lane longitudinally, simulating the driver's psychological expectation that "a vehicle may suddenly emerge from deep in the blind spot"; while the smaller This causes the entropy field intensity to decay rapidly in the lateral direction, limiting the ineffective diffusion of the risk field into non-lane areas.

[0095] In addition, for spatial points located in areas with overlapping lanes, such as the center of an intersection... If it is simultaneously affected by the flow direction of multiple lanes, then by combining the relevant lane sets... Perform a summation or maximum value calculation to aggregate the combined risks from each lane. The general form is:

[0096]

[0097] Through the above calculations, an invisible but numerically real "virtual artifact" is formed within the blind spot. This artifact is no longer a circular "fog," but rather a long, narrow, or fan-shaped potential energy distribution flowing along the lane. When the simulated vehicle plans a path and attempts to traverse this area, regardless of whether there are actual obstacles in the blind spot, this topological entropy field exerts a mathematical "damping" or "repulsive force" on the vehicle, forcing the vehicle model to exhibit a tendency to decelerate or detour at the numerical solution level. This thus replicates the defensive driving logic of a human driver at the microscopic level.

[0098] In this invention, considering the significant social attributes of actual driving behavior, drivers often indirectly infer environmental risks by observing the behavior of other vehicles. For example, when the driver's view is obstructed, but a vehicle is accelerating past, the driver will infer that the area is safe. To mathematically represent this complex psychological game process, this embodiment constructs a second-order line-of-sight coupling model.

[0099] First of all, this car Traverse the set of other nearby vehicles that can be observed within its field of view. For any adjacent car in the set The system needs to determine specific spatial points within the blind zone. Is it located in the adjacent vehicle? Within the visible range. This is equivalent to the neighboring car. Using the new viewpoint origin, repeat the ray projection process described above. Define a second-order line-of-sight indicator function. If point Not relative to adjacent vehicles If an obstacle blocks the view and is within the effective line of sight, then Otherwise, it is 0. This step establishes a three-way geometric relationship between "this vehicle - neighboring vehicles - blind spot".

[0100] After confirming the neighboring vehicle blind spot Once visibility is achieved, the system further evaluates neighboring vehicles. The credibility of the information conveyed by the current action, i.e., the confidence level of the action. The confidence level calculation depends on the longitudinal acceleration of the neighboring vehicle. and speed Generally, steady acceleration or constant speed suggests a safe environment, while sudden braking or sharp deceleration suggests potential danger. This embodiment uses the following normalized mapping function to calculate the confidence level:

[0101]

[0102] In the formula, A threshold constant for determining acceleration and deceleration behavior; The scaling factor used to control the steepness of the sigmoid function; This is the road speed limit. This formula ensures that when a neighboring vehicle accelerates at a high speed, ... The confidence level approaches 1 (high confidence); conversely, if the adjacent vehicle significantly decelerates, It will drop rapidly and may even tend to turn negative.

[0103] By combining visibility assessment and behavioral confidence, calculations are performed for spatial points. Trust inhibitors This factor is presented in a multiplicative form to reflect the group trust enhancement effect resulting from the combined influence of multiple neighboring vehicles:

[0104]

[0105] in, This is a social influence weighting coefficient used to adjust the degree to which interactive information interferes with one's own cognition. When one or more neighboring vehicles clearly see the blind spot and exhibit high-confidence passing behavior, The value will be significantly less than 1, which plays a role in "noise reduction" or "suppression" of the entropy value.

[0106] Finally, the calculated trust inhibition factor is used to evaluate the generated topological entropy value. After weighted correction, the spatial cognitive entropy field strength used as the final decision input is obtained. :

[0107]

[0108] Through this correction process, blind spots that were originally judged to be high-risk geometrically may become safe at the cognitive level due to the "exploratory" behavior of neighboring vehicles. This mechanism gives simulated vehicles human-like reasoning abilities, enabling them to flexibly follow the vehicle in front or neighboring vehicles through blind spots at congested intersections, without getting stuck or stopping excessively due to rigid visual obstruction judgments.

[0109] In this invention, the microscopic behavioral decisions of the vehicle are modeled as a finite-time domain... This presents an optimal control problem within the blind spot. To simultaneously consider driving efficiency, physical safety, and the cognitive and psychological burden caused by blind spots, this embodiment designs a composite cost function. This function is about time. The integral form covers the period from the current moment. End of prediction horizon The whole process:

[0110]

[0111] In the formula, For the cost of conventional physical operation, The weighting coefficients for cognitive entropy cost. This is a cognitive entropy cost term unique to this invention.

[0112] Physical operating cost It consists of three parts: efficiency, comfort, and obstacle avoidance. Its mathematical expression is as follows:

[0113]

[0114] in, Indicates the vehicle's desired cruising speed; and These are the penalty weights for speed tracking error and acceleration amplitude, used to constrain the vehicle to maintain a stable and efficient driving state; the last term is a potential field obstacle avoidance term based on an inverse proportional function. and These represent the time intervals at which the vehicle and the detected physical obstacles are predicted. The position vector, To prevent errors caused by dividing by zero on small positive numbers, For obstacle avoidance weights.

[0115] Cognitive entropy cost This is used to quantify the psychological stress caused by a vehicle entering a region of high uncertainty. In this embodiment, it is defined as the physical space occupied by the vehicle at each moment on the predicted trajectory. In real-time cognitive entropy field Overlap integral in:

[0116]

[0117] In numerical computation, the rectangular region occupied by the vehicle is typically discretized into several sampling points, and the integral described above is approximated by summing the field strength values ​​at these sampling points. Because Since it incorporates topological anisotropy and socialization inhibition factors, this cost term can keenly reflect whether the blind zone is dangerous and whether it is worth taking the risk.

[0118] Based on the aforementioned cost function, a Model Predictive Control (MPC) solver is constructed. The solver searches for the optimal control input sequence within each simulation step. This minimizes the total cost JJ while satisfying the vehicle's kinematic constraints. The optimization problem is formalized as follows:

[0119]

[0120] in This refers to the discrete-time state equation described in Example 1.

[0121] Once the solver converges, the system extracts only the first element from the control sequence. Apply to the vehicle dynamics model to update the vehicle to the next time step. status .

[0122] As the vehicle's position changes, the driver's viewpoint... Displacement occurs, causing changes in the projection of light and obscuring the viewpoint. This leads to movement or deformation, which in turn results in a cognitive entropy field. The distribution is reconstructed. This closed-loop cycle of "perception-modeling-decision-motion" repeats itself until the simulation ends.

[0123] Through this mechanism, when the vehicle detects a high-entropy blind spot ahead, in order to minimize... The optimizer will automatically tend to reduce speed or adjust course, thus naturally giving rise to anthropomorphic defensive driving behaviors such as slowing down to observe and peering out, without the need for manually writing cumbersome rule logic.

[0124] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for modeling the microscopic behavior of vehicles in intersection scenarios, characterized in that, Includes the following steps: Step S1: Obtain the static topology data of the intersection scene and the motion state information of all vehicles in the scene at the current moment; Step S2: Perform dynamic ray projection with the driver's viewpoint as the origin. Based on the intersection of the ray and obstacles in the scene, determine the visible area at the current moment and extract the line-of-sight cutoff interface that distinguishes the visible area from the blind spot. Step S3: Based on the relationship between the road lane flow direction and the spatial distribution of blind spots, a topological anisotropic cognitive entropy flow field is constructed in the blind spot after the line of sight cutoff interface. The topological anisotropic cognitive entropy flow field is used to characterize the potential risk uncertainty of the spatial location in the blind spot. Step S4: Construct a trajectory decision objective function that includes physical operation cost and cognitive entropy cost. Solve the future control sequence of the vehicle by minimizing the objective function. The cognitive entropy cost is calculated based on the cumulative intensity of the vehicle's planned trajectory in the topological anisotropic cognitive entropy flow field. The construction of the topological anisotropic cognitive entropy flow field in step S3 also includes introducing a second-order line-of-sight coupling mechanism to dynamically correct the field strength, specifically including: Detect other nearby vehicles within the vehicle's field of vision; Determine whether a specific area within the blind spot is within the visible area of ​​the adjacent vehicle; If the specific area is located within the visible area of ​​the neighboring vehicle, a trust inhibition factor is generated based on the confidence level of the driving behavior of the neighboring vehicle. The trust inhibition factor is used to reduce the entropy field intensity of the specific region, thereby completing the social interaction correction. Calculate the trust inhibition factor : ； In the formula, This refers to the group of other nearby vehicles within the vehicle's field of vision. As an indicator function, when a specific area within the blind zone... Located in the adjacent car The value is 1 when it is within the visible area; For neighboring cars Behavioral confidence; This is the social impact weighting coefficient; The formula for confidence level is as follows: ； In the formula, A threshold constant for determining acceleration and deceleration behavior; The scaling factor used to control the steepness of the sigmoid function; This refers to the road speed limit. It is longitudinal acceleration; For speed.

2. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 1, characterized in that, The dynamic light projection process in step S2 specifically includes: Within the driver's field of vision, emit discrete rays and calculate the nearest intersection point of each ray with the bounding box of a static obstacle or a dynamic vehicle. If the effective length of the ray is less than the maximum set viewing distance, it is determined that there is an obstruction in the direction of the ray, and the set of all the terminal intersections of the obstructed rays is defined as the line of sight cutoff interface. The line-of-sight cutoff interface divides the intersection space into a defined visible area and a blind spot with uncertain visibility.

3. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 1, characterized in that, The construction of the topological anisotropic cognitive entropy flow field in step S3 specifically includes: Identify potential lanes to which or adjacent to spatial points within blind spots; obtain the tangent vector of the potential lane at the spatial point; The relative position of the spatial point with respect to the line-of-sight cutoff interface is decomposed into a longitudinal projection distance along the tangent vector direction and a lateral projection distance perpendicular to the tangent vector direction. Based on the longitudinal projection distance and the lateral projection distance, the topological basic entropy value of the spatial point is calculated, wherein the rate of decay of the entropy value is greater in the lateral direction than in the longitudinal direction, so that the field strength distribution exhibits anisotropic characteristics extending along the lane.

4. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 3, characterized in that, The calculation of the topological entropy value adopts a Gaussian decay model, setting the longitudinal decay coefficient to be greater than the lateral decay coefficient, so that the generated virtual artifact shape is constrained by the lane geometric topology.

5. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 1, characterized in that, The generation logic of the trust inhibition factor is as follows: If a nearby vehicle has a clear view of the blind spot and maintains a stable acceleration or constant speed passing behavior, a high-intensity trust inhibition factor is generated to significantly reduce the entropy value. If a nearby vehicle brakes suddenly or swerves to avoid a collision, a low-intensity or negative trust inhibition factor is generated, maintaining or enhancing the entropy value within the blind spot.

6. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 1, characterized in that, The construction of the trajectory decision objective function in step S4 specifically includes: Define the physical operation cost, which includes a velocity tracking error term, an acceleration smoothness term, and a distance-inverse penalty term to physical obstacles; Define cognitive entropy cost as the sum of integrals of the geometric region covered by the planned trajectory of the vehicle in the prediction time domain within the topological anisotropic cognitive entropy flow field; The total cost function is obtained by weighted summing of the physical operation cost and the cognitive entropy cost.

7. A method for modeling vehicle micro-behavior in intersection scenarios according to claim 6, characterized in that, In step S4, the future control sequence of the vehicle is solved by using model predictive control to find the optimal control input that minimizes the total cost function while satisfying the vehicle kinematic constraints. When the topological anisotropic cognitive entropy flow field intensity is high in the blind zone, the cognitive entropy cost drives the solver to output deceleration or lateral offset commands to avoid the high entropy region.

8. The method for modeling vehicle micro-behavior in intersection scenarios according to claim 1, characterized in that, This method is a closed-loop iterative process. After executing the control command at the current moment, steps S2 to S4 are re-executed based on the updated position and heading angle of the vehicle to reconstruct the line-of-sight cutoff interface and the topological anisotropic cognitive entropy flow field in real time.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 8.

Images