Non-structured intersection traffic safety assessment method based on dynamic safety potential field

Abstract

The application relates to the technical field of unstructured road intersection mixed traffic safety management, in particular to a mixed traffic passing strategy safety evaluation method based on a dynamic safety potential field of an unstructured intersection, which comprises the following steps: for a given passing priority strategy, a multi-vehicle simulation trajectory of mixed traffic is generated through a priority-aware forward simulator; a dynamic safety potential field integrating multi-source threats is constructed at each discrete time; an instantaneous safety index is extracted based on the safety potential field and is accumulated along the time domain to obtain a comprehensive safety score; the comprehensive safety score is used as a safety measurement of the passing strategy and is used for comparing the safety performance of different candidate strategies; and a three-level conversion link from an abstract passing strategy to a simulation trajectory, from the simulation trajectory to a time-space potential field and from the time-space potential field to a scalar safety score is established, so that the technical problem that a systematic strategy safety quantitative mechanism is lacked in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of mixed traffic safety management technology at unstructured road intersections, and particularly to a method for assessing the safety of traffic strategies when connected autonomous vehicles (CAVs) and manually driven connected vehicles (CMVs) coexist in unstructured intersections with no signals, no lane markings, and irregular road boundaries (such as open-pit mines and ports). Specifically, it is a traffic strategy safety assessment method based on a dynamic safety potential field. By constructing a spatiotemporally continuous safety potential field that integrates environmental threats and vehicle motion threats, it transforms abstract traffic priority strategies into quantifiable safety indicators, providing a safety assessment basis for the allocation of right-of-way in mixed traffic at unstructured intersections. Background Technology

[0002] In unstructured road environments such as open-pit mines and ports, intersections are critical bottlenecks in traffic systems. These intersections lack traffic lights and standard lane markings, have irregular road boundaries, and often feature undulating terrain, posing significantly higher safety risks than structured intersections in cities. With the advancement of autonomous driving technology, unstructured intersections are evolving from a conservative model that only allows single-vehicle passage to a multi-vehicle cooperative passage model. During this transitional phase, CAVs (Continuously Operated Vehicles) and manually driven CMVs will operate in a mixed manner for a long period, and the rational allocation of right-of-way directly determines operational safety and efficiency.

[0003] The core of right-of-way allocation lies in determining the order in which multiple vehicles pass through a conflict zone, i.e., the traffic strategy. However, how to assess the safety of a given traffic strategy is a crucial prerequisite: only by establishing a reliable safety assessment mechanism can the safest option be selected from multiple candidate strategies. Existing safety assessment methods mainly include time-of-collision (TTC) based indicators and minimum distance based indicators. These indicators only focus on the binary relationship between vehicles and do not incorporate road boundary risks, terrain risks, and uncertainties in human driving into a unified assessment framework, resulting in insufficient assessment in unstructured mixed traffic environments.

[0004] In the prior art, invention patent CN116580548A discloses a vehicle collision warning method based on safety field theory, which assesses collision risk by constructing a safety field around the vehicle. Invention patent CN115376333B discloses a vehicle scheduling method at intersections, which determines conflicts and allocates right-of-way based on projected headway distances. Invention patent CN114283607B discloses a multi-vehicle cooperative planning method based on distributed swarm intelligence learning.

[0005] However, the above methods have the following technical shortcomings in assessing the safety of traffic strategies at unstructured mixed traffic intersections: (1) Existing safety field or safety index methods only consider the collision risk between vehicles and do not include the collision risk of road boundaries and the rollover risk caused by terrain in a unified assessment framework. The technical reason is that in a structured road environment, the boundaries are regular and the terrain is flat, so the risks of boundaries and terrain can be ignored. However, the irregular boundaries and undulating terrain of unstructured intersections are important causes of accidents, and considering only the collision risk between vehicles cannot fully reflect the true safety level of traffic strategies.

[0006] (2) Existing methods fail to effectively quantify the impact of uncertainty in human driving behavior on safety assessment when evaluating traffic strategies involving CMVs. The technical reason is that CMVs are driven by humans, and their trajectories have inherent randomness and unpredictability. Existing deterministic assessment indicators cannot capture this uncertainty, resulting in an underestimation of the risk of traffic strategies involving CMVs.

[0007] (3) Existing methods lack a systematic mechanism to transform abstract traffic priority policies into quantifiable safety indicators. The technical reason is that traffic policies themselves are a set of abstract priority rankings, which cannot be directly measured for safety. An intermediate medium is needed to map the policy to an assessable physical quantity (such as vehicle trajectory), but existing methods have not established a complete transformation link from policy to trajectory and from trajectory to safety indicators.

[0008] In summary, in the mixed traffic management of unstructured intersections, the safety assessment of traffic strategies is a key prerequisite for right-of-way allocation. Existing safety assessment methods only focus on the risk of collisions between vehicles while ignoring boundary and terrain risks, fail to quantify the impact of uncertainties in manual driving, and lack a systematic transformation mechanism from abstract strategies to quantifiable safety indicators. Summary of the Invention

[0009] In view of this, the present invention provides a traffic policy safety assessment method based on a dynamic safety potential field. First, the abstract traffic policy is transformed into a multi-vehicle simulation trajectory through a priority-aware forward simulator. Then, a dynamic safety potential field is constructed that integrates the boundary threat potential, terrain threat potential, and vehicle motion threat potential. The vehicle motion threat potential introduces a driving behavior fluctuation coefficient to CMV to reflect human driving characteristics. Finally, the safety potential field value is accumulated in the time domain along the simulation trajectory to obtain a comprehensive safety score for the traffic policy, providing a quantitative decision-making basis for right-of-way allocation.

[0010] A safety assessment method for unstructured intersection mixed traffic strategies based on dynamic safety potential fields includes the following steps: 1) For a given traffic priority strategy, generate multi-vehicle simulation trajectories for mixed traffic using a priority-aware forward simulator; 2) Construct a dynamic security potential field that integrates multiple sources of threats at each discrete moment; 3) Extract instantaneous safety indicators based on the safety potential field and accumulate them along the time domain to obtain a comprehensive safety score; 4) Use the comprehensive security score as a security metric for the prevailing strategy to compare the security performance of different candidate strategies. After the comparison, select the optimal strategy and output it.

[0011] Furthermore, in step 1), the priority-aware forward simulator transforms the abstract traffic priority strategy into an evaluable multi-vehicle simulation trajectory, specifically including: The priority-aware forward simulator described in 1.1 treats each vehicle as a force-bearing point mass and applies two types of forces to it: Path following guidance force: attracts the vehicle to a forward guidance point on a preset reference path. The guidance force intensity increases with the deviation distance and decreases when approaching. Environmental repulsion force: originates from other vehicles, road boundaries, and terrain extremes, and increases rapidly at close range to trigger evasive maneuvers.

[0012] The path-following guidance force is defined as follows: the vehicle's current position, speed, and heading angle are denoted as:

[0013] In the formula, The vehicle's current location. and These represent the x-axis and y-axis coordinates of vehicle i at time t, respectively. The vehicle's current speed. and These represent the x-axis and y-axis velocity components of vehicle i at time t, respectively. The heading angle of vehicle i at time t.

[0014] For the i-th vehicle, its look-ahead points on the reference path Recorded as:

[0015] The unit direction vector of the vehicle pointing towards the forward guidance point is:

[0016] Path following guidance Defined as:

[0017] in, This represents the maximum strength coefficient of the path guiding force. This is the adjustment coefficient for the guiding force as it varies with the deviation distance. To prevent extremely small positive numbers with a denominator of zero, the guiding force increases as the vehicle deviates further from the reference path; the guiding force gradually decreases as the vehicle approaches the forward guidance point.

[0018] 1.2 Traffic priority is achieved by modulating the asymmetry of the repulsive forces between vehicles, including: make This represents the passage priority identifier for the i-th vehicle, and A larger value indicates higher priority. Define the priority modulation function:

[0019]

[0020] when > When, it means that the i-th car has a higher priority than the j-th car. =0, thus the repulsive force between the j-th vehicle and the i-th vehicle is set to zero, and the high-priority vehicle is not affected by the low-priority vehicle; conversely, the low-priority vehicle is subjected to the full repulsive force of the high-priority vehicle and decelerates to avoid it.

[0021] By changing the priority allocation scheme, the simulator can generate multi-vehicle simulation trajectories corresponding to different traffic strategies.

[0022] 1.3 The design of the repulsive force between vehicles comprehensively considers factors such as the mass of the two vehicles, the relative distance, the relative speed, and the vehicle dimensions: the intensity of the repulsive force is directly proportional to the product of the masses of the two vehicles and inversely proportional to the square of the distance between the boundaries of the two vehicles (analogous to the inverse square law of gravity), and multiplied by a distance-based Gaussian attenuation term to suppress unreasonable interference from long distances; when the relative speed between the two vehicles is high, the repulsive force is amplified through a speed-sensitive factor to reflect the increased urgency of the collision; when two large vehicles approach each other at close range, the repulsive force is further enhanced through a size coupling factor. Specifically: The repulsive force between vehicles is defined as follows: the center distance r between vehicle i and vehicle j. ij (t) and boundary spacing d ij (t) are respectively:

[0023]

[0024] and Let be the envelope radii of vehicle i and vehicle j, respectively.

[0025] The unit repulsion direction from vehicle j to vehicle i for:

[0026] The repulsive force between vehicles without priority modulation is:

[0027] The repulsive force between vehicles after considering traffic priority is:

[0028] in, , The weights of the two vehicles are respectively. , These are the envelope radii of the two vehicles, This is the repulsive force coefficient between vehicles. This is the distance attenuation coefficient. The relative velocity sensitivity coefficient, The vehicle size coupling coefficient. This is the vehicle's maximum speed.

[0029] The design of the boundary repulsion force is as follows: Select several discrete boundary points closest to the vehicle, calculate the exponentially decaying repulsion force for each point (direction from the boundary point to the vehicle, intensity decreases exponentially with distance), and take the average of the repulsion forces of all boundary points as the boundary repulsion force experienced by the vehicle.

[0030] The design of the terrain repulsion force is as follows: based on the digital elevation data of the intersection, local protrusions and depressions are identified as extreme points. Each extreme point is used as a repulsion source. The intensity of the repulsion force is proportional to the deviation of the point from the average road surface height. It decreases with distance according to the Gaussian function. The repulsion forces of all extreme points are superimposed to form the terrain repulsion force.

[0031] Specifically, the boundary repulsion force is designed as follows: select the n discrete points of the road boundary closest to the i-th vehicle, denoted as... 1, 2,..., n The distance from the i-th vehicle to the k-th boundary point is:

[0032] Boundary repulsion Defined as:

[0033] in, The boundary repulsion force intensity coefficient, This is an adjustment coefficient for the range of action of the boundary repulsion force. The direction of this force is from the boundary point towards the vehicle, causing the vehicle to move away from the road boundary.

[0034] The terrain repulsion force is defined as follows: based on the digital elevation data of the intersection, Λ local convex or concave extreme points are identified, denoted as... 1, 2,..., Λ It satisfies:

[0035] in, The elevation of the extreme point. The average elevation of the intersection. The elevation difference threshold is given. The repulsion strength at the m-th terrain extremum point is:

[0036] Terrain repulsion Defined as:

[0037] in, The topographic repulsion coefficient, It is the spatial attenuation scale of topographic repulsion.

[0038] 1.5 The resultant force is obtained through vector summation, projected onto the vehicle's current direction of motion to obtain the acceleration command, and the steering angle command is obtained by comparing the deviation between the resultant force direction and the vehicle's current heading. The control command is truncated based on vehicle kinematic constraints (velocity upper and lower limits, acceleration upper and lower limits, maximum steering angle, etc.) before updating the vehicle state. The process iterates step-by-step until all vehicles have passed, outputting the pose and velocity sequence of each vehicle at each time step. Specifically: The net force acting on the i-th vehicle is the vector sum of the path-following guiding force, the inter-vehicle repulsion force, the boundary repulsion force, and the terrain repulsion force:

[0039] Vehicle current heading unit vector for:

[0040] Projecting the resultant force onto the vehicle's current direction of motion yields the longitudinal acceleration command:

[0041] After incorporating acceleration constraints:

[0042] and These are the lower and upper limits of the acceleration constraint, respectively.

[0043] Resultant force direction angle and steering angle The instructions are as follows:

[0044]

[0045]

[0046] respectively resultant force Components in the x and y directions.

[0047] The resultant force direction angle coefficient, This is the steering angle coefficient. These are the upper and lower limits of the steering angle constraint, respectively.

[0048] Vehicle status is updated according to discrete time step Δt:

[0049]

[0050]

[0051]

[0052] Among them, L i Let represent the vehicle wheelbase. `clip(·)` indicates truncation based on the vehicle's speed, acceleration, and steering angle upper and lower limits. Iterate this process step-by-step until all vehicles have passed through the intersection, outputting the position, speed, and heading angle sequence of each vehicle at each time step.

[0053] Furthermore, the process of constructing the dynamic safety potential field in step 2) is as follows: 2.1 Vehicle Motion Threat Potential Construction: Each vehicle is modeled as an envelope disk, with the envelope radius of the CAV determined by geometric parameters such as vehicle width, front overhang, and rear overhang. For CMVs, a driving behavior fluctuation coefficient is introduced. The envelope radius is enlarged to reflect the inherent uncertainty of the human driving trajectory. This coefficient is determined by collecting historical CMV operation data at the intersection and comprehensively calculating the normalized value of the speed variance, the normalized value of the speed range, and the weighted combination of acceleration fluctuation amplitude. The larger the value, the stronger the uncertainty of human driving.

[0054] In the vehicle-fixed coordinate system, the motion threat is decomposed into a component along the vehicle direction and a component perpendicular to the vehicle direction. The threat intensity along the vehicle direction is inversely proportional to the distance from the front / rear boundaries of the vehicle and is modulated by the current speed and acceleration (the faster the speed and the greater the acceleration, the stronger the forward threat). A direction enhancement factor is applied in front of the vehicle to reflect the high probability of a forward collision. The threat intensity in the perpendicular direction is only related to the distance from the side of the vehicle and decays linearly. The threat components in the two directions are synthesized using the Euclidean norm and their reciprocals are taken to obtain the scalar value of the vehicle's motion threat potential. This inverse distance form ensures that the threat intensity is maximized near the vehicle and decays with distance. For CMVs, an additional fluctuation coefficient is introduced in the calculation of the component along the vehicle direction to amplify its threat range.

[0055] 2.2 Boundary Threat Potential Construction: For irregular road boundaries at unstructured intersections, the physical boundary is first contracted inwards by the vehicle envelope radius to obtain an equivalent safety boundary. When the vehicle center falls into the area between the physical boundary and the equivalent boundary, a collision is unavoidable, and the threat value is set to the upper limit of 1. Outside the equivalent boundary, the threat intensity is calculated according to a power-law decay form based on the distance from the equivalent boundary. A large decay exponent is used to ensure that the threat rapidly decreases to a negligible level as it moves away from the boundary.

[0056] 2.3 Topographic Threat Potential Construction: Based on discrete elevation data of intersection pavements, local extreme points (protrusions and depressions) exceeding the elevation difference threshold are identified, with each extreme point serving as a threat source. The threat intensity of a single extreme point is proportional to the absolute value of its deviation from the average pavement height, and decreases with distance according to a Gaussian function (the threat decreases with distance from the extreme point). The threat contributions of all extreme points are spatially superimposed to constitute the topographic threat potential.

[0057] 2.4 Multi-source potential field superposition: Based on the principle of vector field superposition, the motion threat potential, boundary threat potential and terrain threat potential of all vehicles are superimposed in space to form the total distribution of dynamic safety potential field at that moment.

[0058] Furthermore, in step 3), the process of constructing a multi-layered security indicator system includes: 3.1 Instantaneous Safety Indicator Extraction: At each discrete moment, the intersection space is divided into a uniform grid. Based on the safety potential field obtained in step 2, the average potential field value within each grid is calculated to form a potential field matrix. Three complementary instantaneous safety indicators are extracted from this matrix: peak threat (maximum value of the matrix, reflecting the most dangerous local area at the current moment), mean threat (mean value of the matrix, reflecting the overall risk level of the intersection), and distribution concentration (information entropy of matrix elements; high entropy indicates that the risk is concentrated in a local area, and low entropy indicates that the risk is evenly distributed). The three indicators are weighted according to preset weights to form an instantaneous comprehensive safety indicator, with the peak threat having the largest weight because it directly reflects the probability of a collision.

[0059] 3.2 Comprehensive Security Score: The comprehensive security score S of the passage strategy is obtained by summing the instantaneous comprehensive security indices at each discrete moment along the time domain (equivalent to the discrete approximation of time domain integration). The lower the score S, the smaller the cumulative security risk and the better the security performance under the strategy.

[0060] By comparing the S-values ​​of different candidate passage strategies, the passage scheme with the best security can be determined.

[0061] The parameters involved in the above method are obtained from real-time observation by on-board sensors or cloud control platforms, task instructions issued by the scheduling system, fixed parameters calibrated at the vehicle factory, intersection geometry and vehicle kinematic limit parameters, etc. The weight coefficients and other parameters can be customized by the user according to actual needs. All of the above parameters can be obtained through observation, calculation or customization.

[0062] The advantages of the technical solution of this invention compared with the prior art are as follows: Safety assessment of traffic strategies at unstructured intersections faces three major challenges: the multi-source nature of environmental risks, the uncertainty of human driving, and the inability to directly quantify strategies. This invention, based on a newly developed multi-source fusion dynamic safety potential field, integrates vehicle motion threats, boundary collision threats, and terrain rollover threats into a spatiotemporally continuous potential field framework. Through the principle of field superposition, vector superposition is performed in the same spatial coordinate system to form a comprehensive potential field distribution that can fully characterize the safety situation at unstructured intersections. This overcomes the technical deficiency of existing safety assessment methods that only focus on inter-vehicle collision risks while ignoring the unique boundary and terrain risks inherent in unstructured environments.

[0063] Based on the invented driving behavior fluctuation coefficient mechanism, the trajectory uncertainty parameter of CMV is calibrated by collecting historical operation data of CMV, and the parameter is embedded in the construction process of vehicle motion threat potential to amplify the envelope radius and threat range of CMV. This makes the threat potential generated by CMV naturally greater than that generated by CAV under the same conditions. Thus, the risk difference between controllable and uncontrollable vehicles is implicitly distinguished in a unified potential field framework, overcoming the technical defect that existing deterministic assessment indicators cannot capture the randomness of human driving.

[0064] Based on the invented priority-aware forward simulator and multi-layered security index system, a three-level transformation link is established, from abstract passage strategy to simulation trajectory (encoding priority through asymmetric repulsion mechanism), from simulation trajectory to spatiotemporal potential field (constructed through multi-source potential field), and from spatiotemporal potential field to scalar security score (through instantaneous index extraction and time domain accumulation). This overcomes the technical deficiency of existing methods that lack a systematic strategy security quantification mechanism. Attached Figure Description

[0065] To more clearly illustrate the content and specific embodiments of this invention, the accompanying drawings required for the description process are introduced. The following drawings are merely for the purpose of illustrating the subject matter of this invention. Those skilled in the art can readily provide other similar drawings based on the drawings in this invention without any inventive effort.

[0066] Figure 1 This is a flowchart of the overall process for a security assessment method for access strategies based on dynamic security potential fields. Detailed Implementation

[0067] The present invention will now be described in detail with reference to embodiments and accompanying drawings. However, it should be understood that the embodiments and drawings are for illustrative purposes only and do not constitute any limitation on the scope of protection of the present invention. All reasonable modifications and combinations included within the inventive spirit of the present invention fall within the scope of protection of the present invention.

[0068] A safety assessment method for mixed traffic strategies at unstructured intersections based on dynamic safety potential fields is proposed. This method is mainly used to address the technical problem of quantitatively assessing the safety of a given traffic priority strategy when CAV and CMV are mixed at unstructured road intersections such as open-pit mines and ports.

[0069] like Figure 1 As shown, the method of the present invention mainly includes the following execution steps: Step 1: Traffic Strategy Representation and Simulation Trajectory Generation For m CAVs and n CMVs simultaneously approaching an intersection, a traffic strategy can be represented as a priority arrangement of all vehicles, i.e., a permutation specifying the order in which each vehicle passes through the conflict zone. The given traffic strategy is then input into a priority-aware forward simulator.

[0070] The simulator operates as follows: At each discrete time step, a guiding force (generated by a look-ahead guide point on the reference path, attracting the vehicle back to the reference path) and an environmental repulsive force (generated by other vehicles, road boundaries, and terrain extremities, driving the vehicle away from danger zones) are calculated for each vehicle. The asymmetry of the repulsive force between vehicles is controlled by a priority flag—high-priority vehicles are immune to the repulsive force of low-priority vehicles, while low-priority vehicles bear the full repulsive force of high-priority vehicles. The resultant force is projected onto the direction of motion to obtain acceleration commands, which are then combined with kinematic constraints to update the vehicle state.

[0071] Taking two vehicles as an example, the asymmetric repulsion mechanism is illustrated as follows: When the strategy specifies that vehicle A has a higher priority than vehicle B, the repulsion force of vehicle B on vehicle A is zero, and vehicle A's normal driving is unaffected; the repulsion force of vehicle A on vehicle B remains intact, and vehicle B decelerates to avoid the collision due to the reduced net force, naturally creating a time interval that allows vehicle A to pass through the conflict zone first. This mechanism results in different simulation trajectories for different priority arrangements.

[0072] The simulation iterates step by step until all vehicles have passed through, outputting the position, speed, and heading angle sequence of each vehicle at each time point, which serves as the input for the subsequent construction of the safety potential field.

[0073] Step 2: Constructing a Dynamic Safety Potential Field At each discrete moment, a dynamic safety potential field is constructed based on the current motion state of each vehicle and environmental information. This potential field is composed of the superposition of three sub-potential fields: 1) Threat of vehicle movement Each vehicle is modeled as an envelope disk. For CAVs, the envelope radius is determined by the vehicle body geometry parameters; for CMVs, the envelope radius is additionally multiplied by a driving behavior fluctuation coefficient. The threat is magnified. Threat components along the vehicle body direction and perpendicular to the vehicle body direction are calculated separately in the vehicle-fixed coordinate system: the component along the vehicle body direction combines the distance to the vehicle boundary (inversely proportional), the combined effect of current velocity and acceleration (directly proportional), and the asymmetric enhancement factor in the front and rear directions; the component perpendicular to the vehicle body is determined only by the distance to the side of the vehicle. The components in the two directions are synthesized using norms and their reciprocals are taken to obtain a scalar threat value, ensuring that the threat is strongest near the vehicle and decreases with distance. The threat range of the CMV is naturally larger than that of the CAV under the same conditions due to the introduction of the fluctuation coefficient.

[0074] 2) Border Threats The physical boundary is shrunk inward by the envelope radius to obtain the equivalent safety boundary. The area inside the equivalent boundary is the inevitable collision zone, and the threat value is taken as the upper limit. The threat value in the area outside the equivalent boundary decays according to a power function of distance, and a large decay exponent is taken so that the threat decreases rapidly as it moves away from the boundary.

[0075] 3) Terrain threat Local extrema (booms and depressions) are identified from digital elevation data. The threat intensity of each extrema is proportional to its deviation from the average elevation and decreases with distance according to a Gaussian function. The threat contributions of all extrema are spatially superimposed to form the terrain threat potential.

[0076] 4) Potential field superposition By spatially superimposing the motion threat potential, boundary threat potential, and terrain threat potential of all vehicles, the total distribution of the dynamic safety potential field at that moment is obtained.

[0077] Step 3: Safety Indicator Extraction and Comprehensive Scoring At each discrete moment, the intersection space is divided into a uniform grid, and the average potential field value within each grid is calculated to form a potential field matrix. Three instantaneous safety indicators are extracted from this matrix: peak threat (maximum value of the matrix), mean threat (mean value of the matrix), and distribution concentration (matrix information entropy). These three indicators are weighted according to preset weights to form an instantaneous comprehensive safety indicator, with peak threat having the largest weight. The instantaneous indicators are accumulated and summed along the time domain to obtain the comprehensive safety score S of this passage strategy. The lower the S, the safer the strategy.

[0078] Step 4: Policy Security Comparison and Output Steps 1 through 3 are executed for all candidate access strategies to obtain a comprehensive security score S for each strategy. The strategy with the lowest score is identified as the most secure access scheme.

[0079] Furthermore, it can also output the priority ranking of the optimal strategy with the lowest score and the corresponding simulation trajectory, which can be directly used as the initial solution for subsequent trajectory optimization.

[0080] The above-described embodiments are merely some implementation methods of this application. For those skilled in the art, various modifications and improvements can be made without departing from the inventive concept of this application, and these all fall within the protection scope of this application.

Claims

1. A method for assessing the safety of unstructured intersection traffic based on a dynamic safety potential field, characterized in that, Includes the following steps: 1) For a given traffic priority strategy, generate multi-vehicle simulation trajectories for mixed traffic using a priority-aware forward simulator; 2) Construct a dynamic security potential field that integrates multiple sources of threats at each discrete moment; 3) Extract instantaneous safety indicators based on the safety potential field and accumulate them along the time domain to obtain a comprehensive safety score; 4) Use the comprehensive security score as a security metric for the prevailing strategy to compare the security performance of different candidate strategies. After the comparison, select the optimal strategy and output it.

2. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 1, characterized in that, Step 1) The multi-vehicle simulation trajectory includes the position, speed and heading angle sequence of each vehicle at each time moment, which serves as the input for the subsequent construction of the safety potential field.

3. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 2, characterized in that, In step 2), the dynamic safety potential field includes vehicle motion threat potential, boundary threat potential, and terrain threat potential. The motion threat potential, boundary threat potential, and terrain threat potential of all vehicles are superimposed in space to obtain the total distribution of the dynamic safety potential field at that moment.

4. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 2, characterized in that, In step 3), at each discrete moment, the intersection space is divided into a uniform grid. Based on the safety potential field obtained in step 2), the average potential field value in each grid is calculated to form a potential field matrix. Three instantaneous safety indicators are extracted from it: peak threat, mean threat, and distribution concentration. The three are weighted according to preset weights to synthesize an instantaneous comprehensive safety indicator. The instantaneous indicators at each moment are accumulated and summed along the time domain to obtain the comprehensive safety score of the passage strategy.

5. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 4, characterized in that, Steps 1) to 3) are executed for all candidate access strategies to obtain a comprehensive security score for each strategy. The strategy with the lowest score is identified as the safest access strategy.

6. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 1, characterized in that, The priority-aware forward simulator treats each vehicle as a force-bearing point mass, applying path-following guidance force, inter-vehicle repulsion force, boundary repulsion force, and terrain repulsion force to it. It achieves traffic priority by modulating the asymmetry of inter-vehicle repulsion force. Specifically, it assigns a priority identifier to each vehicle. When the priority of vehicle i is higher than that of vehicle j, the repulsion force of vehicle j on vehicle i is set to zero, allowing the high-priority vehicle to drive normally without interference. Conversely, the low-priority vehicle is subjected to the full repulsion force of the high-priority vehicle and slows down to avoid it. By changing the priority allocation scheme, the priority-aware forward simulator generates multi-vehicle simulation trajectories corresponding to different traffic priority strategies.

7. The method for assessing the safety of unstructured intersection traffic based on a dynamic safety potential field as described in claim 6, characterized in that, The design of the repulsion force between vehicles takes into account factors such as the mass of the two vehicles, the relative distance, the relative speed, and the vehicle size: the intensity of the repulsion force is directly proportional to the product of the masses of the two vehicles and inversely proportional to the square of the distance between the boundaries of the two vehicles, and is multiplied by a distance-based Gaussian attenuation term to suppress unreasonable interference at long distances; when the relative speed of the two vehicles is high, the repulsion force is amplified through a speed-sensitive factor to reflect the increased urgency of the collision; when two large vehicles approach each other at close range, the repulsion force is further enhanced through a size coupling factor.

8. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 6, characterized in that, The boundary repulsion force is designed as follows: select several discrete boundary points closest to the vehicle, calculate the exponentially decaying repulsion force for each point, and take the average of the repulsion forces of all boundary points as the boundary repulsion force experienced by the vehicle.

9. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 6, characterized in that, The design of the terrain repulsion force is as follows: based on the digital elevation data of the intersection, local protrusions and depressions are identified as extreme points. Each extreme point is used as a repulsion source. The intensity of the repulsion force is proportional to the deviation of the point from the average road surface height. It decreases with distance according to the Gaussian function. The repulsion forces of all extreme points are superimposed to form the terrain repulsion force.

10. The method for assessing the safety of unstructured intersections based on a dynamic safety potential field as described in claim 6, characterized in that, The resultant force is obtained by vector summation and projected onto the vehicle's current direction of motion to obtain the acceleration command. The steering angle command is obtained by comparing the deviation between the resultant force direction and the vehicle's current heading. The control command is truncated in combination with the vehicle's kinematic constraints and the vehicle state is updated. The process is iterated step by step until all vehicles have completed the passage, and the pose and velocity sequence of each vehicle at each time moment is output.

Images