A delay distribution adaptive closed loop decision coverage evaluation method

Abstract

The application discloses a time delay distribution adaptive closed-loop decision coverage evaluation method, and relates to the technical field of intelligent network connected vehicles and automatic driving decision safety evaluation. The method collects environment observation quantities in a discrete decision cycle and writes them into a time-stamped observation buffer queue, calculates the time delay distribution of the environment observation quantities based on the time stamp, adaptively determines observation delay and outputs delayed observation quantities; the delayed observation quantities are input into a decision module to generate candidate control actions, the candidate control actions are checked, and it is determined whether to cover the candidate control actions according to the checking result, an execution action is obtained and executed, and a closed-loop operation log is recorded; further, the coverage rate of the candidate control actions and corresponding diagnostic quantities are calculated according to the closed-loop operation log, an evaluation report is generated, and parameters are adjusted. The method can quantify the coupling relationship between coverage behavior and risk under real perception / communication uncertainty, and improve the authenticity, traceability and engineering deployment capability of evaluation.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicles and autonomous driving decision-making safety evaluation technology, and in particular to a closed-loop decision coverage evaluation method with adaptive time delay distribution. Background Technology

[0002] With the development of intelligent connected vehicles and vehicle-road cooperative technologies, the vehicle decision-making and control link typically consists of sensing and data acquisition, communication transmission, decision reasoning, and execution control. In real road and network environments, observation data inevitably suffers from random delays, jitter, and even packet loss, causing the observations received by the decision module to be outdated and not synchronized with the current moment. To ensure safety, engineering often introduces safety constraint verification and action coverage (safety shielding) mechanisms. However, the triggering frequency, triggering causes, and risk levels of coverage events change significantly with time delay distribution, necessitating a closed-loop evaluation framework for quantitative diagnosis and interpretable analysis of coverage. Existing methods are mostly based on fixed delays or ideal synchronization assumptions, making it difficult to characterize the coupling relationship between coverage events and safety risks under real-time delay distributions.

[0003] Existing coverage assessments often only count the number of coverage events or the overall coverage rate, lacking diagnostic metrics such as coverage rate curves and coverage amplitudes decomposed by time delay conditions. This makes it difficult to pinpoint time delay segments that lead to frequent coverage. Meanwhile, security verification often relies on a single observation source or ignores missing observations, failing to incorporate the filling / masking mechanism for delayed observations into the closed-loop log. This results in assessment results that are insensitive to missing data and have poor reproducibility and experimental reliability. Furthermore, parameter adjustments usually rely on offline experience settings, making it difficult to achieve online adaptive optimization of buffer length, threshold, and minimum intervention weight based on risk margin, conflict time, and risk weight indicators in the closed-loop log. Summary of the Invention

[0004] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a closed-loop decision-making coverage evaluation method with adaptive time delay distribution. This invention writes observations into a timestamped buffer queue, statistically obtains the delay distribution within a sliding time window, samples and determines the observation delay, outputs the corresponding historical observations as input to the delay observation decision module to generate candidate actions; then, it performs safety constraint verification on the candidate actions and implements coverage according to the principle of minimum intervention, forming a closed-loop operation log, calculating the coverage rate and diagnostic parameters, and generating an evaluation report for parameter adjustment.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] According to the present invention, an adaptive closed-loop decision coverage evaluation method based on time delay distribution includes:

[0007] Step A: Collect current environmental observations within the discrete decision-making cycle and write the environmental observations into the observation buffer queue; at the same time, add a timestamp to the environmental observations;

[0008] Step B: Obtain the delay distribution of environmental observations based on timestamp statistics, determine the observation delay of the discrete decision cycle based on the delay distribution, and output the delay observations based on the observation buffer queue and the observation delay.

[0009] Step C: Input the delayed observations into the decision module to generate candidate control actions;

[0010] Step D: Verify the candidate control actions and obtain the verification results;

[0011] Step E: Based on the verification results, determine whether to cover the candidate control actions to obtain the execution actions; execute the execution actions and record the closed-loop operation log;

[0012] Step F: Calculate the coverage of candidate control actions and their diagnostic parameters based on the closed-loop operation log;

[0013] Step G: Generate an evaluation report based on the coverage and diagnostic quantity of candidate control actions.

[0014] As a further optimization scheme of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step A, the discrete decision cycle index is denoted as... , will the Environmental observations collected during each decision-making cycle are recorded as follows: Establish a first-in-first-out queue for storing environmental observations, denoted as the observation buffer queue. ,and Maximum storage length is ,in Indicates the maximum allowed observation delay steps; and will Write to the observation buffer queue to update Furthermore, a timestamp is recorded for each environmental observation. .

[0015] As a further optimization scheme of the time delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step B, the delay distribution is denoted as... ,in It is obtained from the statistical analysis of actual observation delays within the sliding time window, and supports updates using exponential moving average or histogram normalization; it is obtained by sampling observation delays according to the delay distribution. , , This represents the observation delay steps in the t-th period; based on from Extract historical observations and output delayed observations. ,in ,in This represents environmental observations delayed by several periods from the t-th period backwards; when the observation buffer queue is missing... When outputting, use either padding or masking mechanisms. , Indicates a missing mask and is used to indicate whether delayed observations are obtained through padding; and records... .

[0016] As a further optimization scheme of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step C, the decision module is implemented as a policy function. ,in To reinforce learning strategies, supervise learning strategies, or rule-based controllers; and to generate candidate control actions based on delay observations. ,in and Indicates candidate control actions; candidate actions It is one of the following actions: discrete lane changing action, continuous longitudinal control action, or a combination of both, and the candidate action space is denoted as . .

[0017] As a further optimization of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, step D includes constructing observations for the nearest available observation source or a specified observation source for security shielding calculation. ,and ,in Indicates the source selection function, Indicates the selection of function parameters; from Parse the controlled vehicle index vertical position Longitudinal velocity With lane number And for candidate control actions Perform security constraint verification, where the security constraints include at least one of the following or a combination thereof:

[0018] (1) Lane boundary / lateral feasibility constraints: Let the candidate lateral action be denoted as ,in Includes lane change direction / amplitude; target lane is denoted as ,in Indicate the target lane; and satisfy the following conditions: , Indicates the total number of lanes on the road;

[0019] (2) Front and rear clearance constraints: in the target lane The above is confirmed and ,in Indicates the longitudinal position in the target lane. The license plate number of the vehicle ahead and the nearest vehicle. Indicates the longitudinal position in the target lane. The rearmost and nearest vehicle number; the front clearance is denoted as The back gap is denoted as ,in and Representing vehicles , The vertical coordinate, Indicates the safety clearance correction length; and satisfies and ,in Indicates the lower limit of the anterior gap, Indicates the lower limit of the back gap;

[0020] (3) Conflict time constraint: forward conflict time Backward conflict time ,in and These represent vehicle numbers respectively. and The longitudinal speed of the vehicle, , Represents the zero constant; and satisfies and ,in This represents the lower limit threshold for conflict time. When any safety constraint is not met, the candidate control action is determined to be unsafe, and the final result of the candidate action verification that meets the safety constraints is obtained.

[0021] As a further optimization of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, step E includes: defining a coverage indicator. ,in and This indicates that the candidate control action does not meet the safety constraints. This indicates that the candidate actions satisfy the safety constraints; and is generated according to the principle of minimum intervention. ,in This indicates the execution of an action, and satisfies the following:

[0022] in Indicates a safety action; safety action To maintain lane position, decelerate, or limit maximum lateral movement; and the minimum intervention principle is to cover only when unsafe, otherwise not to change the candidate action; and when At that time, the safe actions are obtained by solving the following constraints: The set of actions that satisfy the safety constraints... Inside, make The smallest action as ,in Indicates the action to be selected. This refers to a weighted matrix or weight coefficients.

[0023] Record for each decision cycle A closed-loop operation log is generated to characterize the coupling relationship between observation latency, coverage events, and action execution; and it supports classifying and recording coverage events by vehicle dimension, time window dimension, or risk level dimension, where the time window length is denoted as... Furthermore, the closed-loop operation log further records the coverage extent. Missing mask and safety margin .

[0024] As a further optimization of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step F, the coverage rate is calculated based on the closed-loop operation log. ,in This represents the coverage rate within the statistical interval, and:

[0025] in This indicates the total number of decision cycles included in the statistical interval; and further outputs the condition coverage rate as a function of delay. ,in This indicates the possible values ​​for the delay steps, and:

[0026] in Indicates indicator functions, Indicates the zero constant; and will As a diagnostic output, or when or Exceeding the threshold Time-triggered alarms / online adjustments, among which This indicates the coverage alarm threshold.

[0027] As a further optimization of the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step G, the evaluation report further outputs at least one of the following: 95% delay quantile or 99% delay quantile, missing rate. Average coverage and risk-weighted coverage ratio ,in The weight is represented by the minimum TTC or minimum safety margin.

[0028] As a further optimization scheme for the time-delay distribution adaptive closed-loop decision coverage evaluation method described in this invention, in step G, after generating the evaluation report, the parameters are adjusted; wherein...

[0029] Parameter adjustments include: updating the delay distribution. Adjust the maximum buffer length Adjust the safety threshold Or adjust the minimum intervention projection / weighting coefficient.

[0030] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0031] This invention constructs and adaptively updates the delay distribution based on timestamps, making the evaluation process consistent with the actual perception / communication latency. In particular, it can output multiple diagnostic quantities that vary with latency, making the cause of coverage triggering traceable. It can be used for safety verification and regression testing of decision modules in intelligent connected vehicles. This invention further integrates safety constraints such as lane boundaries, front and rear gaps, and conflict time, obtains safety actions through minimum intervention projection, and records the coverage range and safety margin. It supports risk-weighted coverage and threshold alarms, thereby allowing online adjustment of buffer length and safety thresholds, improving evaluation efficiency, interpretability, and engineering deployability. Attached Figure Description

[0032] This is a flowchart illustrating the technical process of the present invention. Detailed Implementation

[0033] To further understand this method, preferred embodiments are described below with reference to examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of this method, and not for limiting the scope of the claims. The descriptions in this section pertain only to typical embodiments, and this method is not limited to the scope described in the embodiments. Combinations of different embodiments, substitution of some technical features in different embodiments, and substitution of identical or similar prior art with some technical features in the embodiments are also within the scope of this method's description and protection.

[0034] This invention relates to the field of safety evaluation for intelligent connected vehicles and autonomous driving decisions, and particularly to a closed-loop decision coverage evaluation method adapted to observation delay distribution, applicable to safety shielding, coverage statistics and diagnostic report generation in vehicle-mounted sensing / vehicle-road cooperative scenarios.

[0035] This embodiment is designed for "three-lane highways ( The scenario of "Closed-loop decision-making coverage assessment for connected autonomous vehicles" involves random latency and occasional packet loss in vehicle-to-everything (V2X) communication, which may cause the decision-making module to generate candidate actions based on historical observations. The system performs safety constraint checks on the candidate actions and, when unsafe, generates safe actions using the principle of minimum intervention, forming a closed-loop operation log. Within a statistical interval, it calculates diagnostic metrics such as coverage rate, conditional coverage rate, and latency quantiles / missing rate / average coverage amplitude / risk-weighted coverage rate, and triggers alarms and parameter adjustments accordingly. This mainly includes the following steps: Figure 1 As shown:

[0036] Step A: Collect current environmental observations within the discrete decision-making cycle and write the environmental observations into the observation buffer queue; at the same time, attach a collection timestamp to the environmental observations;

[0037] Step B: Obtain the delay distribution based on timestamp statistics, determine the observation delay of this discrete decision cycle based on the delay distribution, and output the corresponding historical observations as delay observations based on the observation buffer queue, while recording the observation delay.

[0038] Step C: Input the delayed observations into the decision module to generate candidate control actions;

[0039] Step D: Perform safety constraint verification on the candidate control actions based on the observations to obtain the verification results;

[0040] Step E: Based on the verification results, decide whether to overwrite the control action to obtain the execution action; execute the execution action and generate a closed-loop operation log;

[0041] Step F: Calculate the coverage and diagnostic quantities based on the closed-loop operation log;

[0042] Step G: Generate an evaluation report and adjust parameters;

[0043] In step A, firstly, current environmental observations are collected within the discrete decision-making period and written into the observation buffer queue; simultaneously, a collection timestamp is added to the environmental observations; in this embodiment, the discrete decision-making period index is denoted as... ,in Indicates the first The decision-making cycle; will the first The original environmental observations collected in each cycle are denoted as follows: ,in Represent the vehicle and environment state vectors obtained from onboard sensors and / or vehicle-to-infrastructure (V2I) communication; establish and maintain the observation buffer queue. ,in Indicates the period A first-in, first-out queue that does not store historical decision-making cycle observations, and the maximum storage length of the queue is [missing information]. ,in This represents the maximum allowed observation delay steps (in this embodiment, we take...). ); and will Write to the observation buffer queue to update Furthermore, a timestamp is recorded for each observation that joins the team. (In this embodiment, the timestamp precision is 1ms).

[0044] In this embodiment, the control period is taken as (10 Hz); Therefore Timestamp In milliseconds, when At that time, the queue saves ;when Arrival Writing Then, First-In-First-Out (FIFO) is updated to This ensures that the queue length does not exceed .

[0045] In step B, the delay distribution is obtained based on timestamp statistics. The observation delay for the current discrete decision period is determined according to the delay distribution, and the corresponding historical observations are output as delay observations based on the observation buffer queue. Simultaneously, the observation delay is recorded. In this embodiment, the delay sampling rule is to sample the observation delay according to the delay distribution. ,in Indicates the first The number of observation delay steps in each period; the delay distribution is denoted as ,in It is obtained from the statistical analysis of actual observation delays within the sliding time window, and supports updates using exponential moving average or histogram normalization; based on from Extract historical observations and output delayed observations. ,in When the queue is missing When outputting, use either padding or masking mechanisms. ,in This represents a missing mask and is used to indicate whether delayed observations are obtained through padding; it also records the observation delay. .

[0046] In this embodiment, the length of the sliding time window is taken as... One cycle (approximately 20 seconds); the exponential moving average coefficient is taken as... When missing, use "nearest available observation forward fill" and set... In recent times The observed delay step sequence in each period is as follows: Then the delay counts are: Normalization yields: If the sampling in this period follows this distribution, Then output If the queue is missing due to packet loss at this time Then output (Forward Fill) Juxtaposition .

[0047] In step C, the delay observations are input into the decision module to generate candidate control actions; in this embodiment, the decision module is implemented as a policy function. ,in It is one of the reinforcement learning strategy, supervised learning strategy, or rule controller; and generates candidate control actions based on delayed observations. ,in and Indicates a candidate action; the candidate action It is one of the following actions: discrete lane changing action, continuous longitudinal control action, or a combination of both, and the candidate action space is denoted as . .

[0048] This embodiment uses a "combined action" to make ,in These represent changing lanes to the left, keeping in the same lane, and changing lanes to the right, respectively. Represents longitudinal acceleration; policy function The reinforcement learning policy network is selected for offline training. If the input in a certain cycle... Then, the strategy output. (Change lanes to the right) (Slight acceleration) then candidate control action .

[0049] In step D, safety constraint verification is performed on the candidate control action based on the observations to obtain the verification results; in this embodiment, observations for safety verification are constructed. ,in This represents the observations from the nearest available observation source or a specified observation source used for security shielding calculations, and ,in This represents the observation source selection function (in this embodiment, the original observations of the current period are selected first). If unavailable, select );from Parse the controlled vehicle index vertical position Longitudinal velocity With lane number And for candidate actions In this embodiment, security constraint verification is performed. The safety clearance correction length is taken as follows: The front / back gap threshold is taken as follows: ; Take zero constant ; Lower limit of conflict time .

[0050] The safety check calculation method is as follows:

[0051] Security example: Let Candidate satisfy On lane 3, take... The car in front The car behind .but , . , Therefore, the candidate action is deemed safe.

[0052] Unsafe example: Keep If the following car is closer and faster: .but ,and If any constraint is not satisfied, the system is deemed unsafe.

[0053] In step E, based on the verification result, it is determined whether to overwrite the control action to obtain the execution action; the execution action is executed and a closed-loop operation log is generated; in this embodiment, an overwrite indicator is defined. ,in and This indicates that the candidate control action does not meet the safety constraints. This indicates that the candidate actions meet the safety constraints; and the execution actions are generated based on the principle of minimum intervention. ,in This represents the final action to be performed, and satisfies: when hour ,when hour ,in This indicates a safe action; the safe action is at least one of lane keeping, deceleration, and limiting maximum lateral movement; and the minimum intervention principle is "cover only when unsafe, otherwise do not change the candidate action"; and when At that time, the safety action is obtained by solving the following constraints: within the set of possible actions that satisfy the safety constraints Inside, make The smallest action as ,in This is a weighted matrix or weight coefficients; records are made for each decision cycle. A closed-loop operation log is generated, and it supports classifying and recording covered events by vehicle, time window, or risk level, where the time window length is denoted as... (In this embodiment, we take) Furthermore, the closed-loop operation log further records the coverage range. Missing mask and safety margin .

[0054] In this embodiment, the candidate set of safety actions is limited to: ; weighted matrix When covered, prioritize "keep lane + decelerate".

[0055] If the candidate action is And if it is determined to be unsafe in step D, then Assuming in Candidate actions that meet safety constraints are Calculate the weighted distances separately: Therefore, choose Execute actions The coverage area is:

[0056] In step F, coverage and diagnostic metrics are calculated based on the closed-loop operation log; in this embodiment, coverage is calculated based on the closed-loop operation log. ,in This represents the coverage rate within the statistical interval, and ,in This indicates the total number of decision cycles included in the statistical interval; and further outputs the condition coverage rate as a function of delay. ,in This indicates the possible values ​​for the delay steps, and ,in Indicates indicator functions, Represents the zero constant (taken in this embodiment) ); and will As a diagnostic output, or when or Exceeding the threshold Time-triggered alarm / online adjustment (in this embodiment, it is taken as...) ).

[0057] In this embodiment, the statistical interval is taken as... One cycle (20 s); Alarm threshold .exist Internal statistics obtained ,but Not exceeding This does not trigger an overall coverage alarm. Further statistical analysis of the delay distribution count is as follows: (The rest are 0), and the corresponding number of coverages is: .but:

[0058]

[0059]

[0060]

[0061] in and Exceed This triggers a "latency-related coverage" alarm, indicating that the system is more prone to security coverage under higher latency.

[0062] In step G, an evaluation report is generated and parameters are adjusted; in this embodiment, the evaluation report further outputs the following indicators: delay quantiles. Missing rate Average coverage and risk-weighted coverage ratio ,in by minimum Or a minimum safety margin construction; wherein the parameter adjustment includes at least one of the following: updating the delay distribution Adjust the maximum buffer length Adjust the safety threshold Or adjust the minimum intervention projection / weighting coefficient (corresponding to in this embodiment) ).

[0063] In this embodiment by minimum Constructed as ; Missing alarm threshold ;when Prioritize adding (Initial in this embodiment) The relevant indicators are calculated as follows:

[0064] (1) Delayed quantile: From the above... (Total 200) can be accumulated to Just reached Therefore, the first The quantile corresponds to the first Take a sample, ;No. The sample fell interval, take .

[0065] (2) Missing rate: The number of times the missing mask is 1 within the statistical interval. ,but No missing alerts will be triggered.

[0066] (3) Average coverage: If coverage occurs secondary ,but .

[0067] (4) Risk-weighted coverage ratio: Assuming the interval And the weighted sum when covering ,but The coverage rate is too high under the "high-risk weighting" factor, in conjunction with the above. Alarms can indicate that the system triggers security coverage more frequently under conditions of high latency and high risk.

[0068] (5) Parameter adjustment: due to The latency is relatively high, so in this embodiment, the following online adjustment is performed: prioritize updating the latency distribution. Increase support Statistical weights are used to improve adaptability; and the safety threshold is adjusted. from Upgraded to To proactively shield potential conflicts, while using a minimum intervention weighting matrix The horizontal weighting was adjusted from 10 to 15 to reduce the tendency for aggressive lane changing; if subsequent monitoring shows Approaching or exceeding Then Increased to 6 to reduce the fill ratio caused by queue missing items.

[0069] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A time-delay distribution adaptive closed-loop decision coverage evaluation method, characterized in that, include: Step A: Collect current environmental observations within the discrete decision-making cycle and write the environmental observations into the observation buffer queue; At the same time, timestamps are added to environmental observations; Step B: Obtain the delay distribution of environmental observations based on timestamp statistics, determine the observation delay of the discrete decision cycle based on the delay distribution, and output the delay observations based on the observation buffer queue and the observation delay. Step C: Input the delayed observations into the decision module to generate candidate control actions; Step D: Verify the candidate control actions and obtain the verification results; Step E: Based on the verification results, determine whether to cover the candidate control actions to obtain the execution actions; execute the execution actions and record the closed-loop operation log; Step F: Calculate the coverage of candidate control actions and their diagnostic parameters based on the closed-loop operation log; Step G: Generate an evaluation report based on the coverage and diagnostic quantity of candidate control actions.

2. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, In step A, the discrete decision cycle index is denoted as... , will the Environmental observations collected during each decision-making cycle are recorded as follows: Establish a first-in-first-out queue for storing environmental observations, denoted as the observation buffer queue. ,and Maximum storage length is ,in Indicates the maximum allowed observation delay steps; and will Write to the observation buffer queue to update Furthermore, a timestamp is recorded for each environmental observation. .

3. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 2, characterized in that, In step B, the delay distribution is denoted as ,in It is obtained from the statistical analysis of actual observation delays within the sliding time window, and supports updates using exponential moving average or histogram normalization; it is obtained by sampling observation delays according to the delay distribution. , , This represents the observation delay steps in the t-th period; based on from Extract historical observations and output delayed observations. ,in ,in This represents the environmental observations that are delayed by several cycles from the t-th cycle. When the observation buffer queue is missing When outputting, use either padding or masking mechanisms. , Indicates a missing mask and is used to indicate whether delayed observations are obtained through padding; and records... .

4. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, In step C, the decision module is implemented as a policy function. ,in To reinforce learning strategies, supervise learning strategies, or rule-based controllers; and to generate candidate control actions based on delay observations. ,in and Indicates candidate control actions; candidate actions It is one of the following actions: discrete lane changing action, continuous longitudinal control action, or a combination of both, and the candidate action space is denoted as . .

5. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, Step D includes constructing an observation set of the nearest available observation source or a specified observation source for security shielding calculations. ,and ,in Indicates the source selection function, Indicates the selection of function parameters; from Parse the controlled vehicle index vertical position Longitudinal velocity With lane number And for candidate control actions Perform security constraint verification, where the security constraints include at least one of the following or a combination thereof: (1) Lane boundary / lateral feasibility constraints: Let the candidate lateral action be denoted as ,in Includes lane change direction / amplitude; target lane is denoted as ,in Indicate the target lane; and satisfy the following conditions: , Indicates the total number of lanes on the road; (2) Front and rear clearance constraints: in the target lane The above is confirmed and ,in Indicates the longitudinal position in the target lane. The license plate number of the vehicle ahead and the nearest vehicle. Indicates the longitudinal position in the target lane. The rearmost and nearest vehicle number; the front clearance is denoted as The back gap is denoted as ,in and Representing vehicles , The vertical coordinate, Indicates the safety clearance correction length; and satisfies and ,in Indicates the lower limit of the anterior gap, Indicates the lower limit of the back gap; (3) Conflict time constraint: forward conflict time Backward conflict time ,in and These represent vehicle numbers respectively. and The longitudinal speed of the vehicle, , Represents the zero constant; and satisfies and ,in Indicates the lower limit threshold for conflict time; If any safety constraint is not met, the candidate control action is determined to be unsafe, and the final result of the candidate action verification that meets the safety constraints is obtained.

6. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, Step E includes: defining the coverage indicator. ,in and This indicates that the candidate control action does not meet the safety constraints. This indicates that the candidate actions satisfy the safety constraints; and is generated according to the principle of minimum intervention. ,in This indicates the execution of an action, and satisfies the following: in Indicates a safety action; safety action To maintain lane position, decelerate, or limit maximum lateral movement; and the minimum intervention principle is to cover only when unsafe, otherwise not to change the candidate action; and when At that time, the safe actions are obtained by solving the following constraints: The set of actions that satisfy the safety constraints... Inside, make The smallest action as ,in Indicates the action to be selected. This refers to a weighted matrix or weight coefficients. Record for each decision cycle A closed-loop operation log is generated to characterize the coupling relationship between observation latency, coverage events, and action execution; and it supports classifying and recording coverage events by vehicle dimension, time window dimension, or risk level dimension, where the time window length is denoted as... Furthermore, the closed-loop operation log further records the coverage extent. Missing mask and safety margin .

7. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, In step F, the coverage rate is calculated based on the closed-loop operation log. ,in This represents the coverage rate within the statistical interval, and: in This indicates the total number of decision cycles included in the statistical interval; and further outputs the condition coverage rate as a function of delay. ,in This indicates the possible values ​​for the delay steps, and: in Indicates indicator functions, Indicates the zero constant; and will As a diagnostic output, or when or Exceeding the threshold Time-triggered alarms / online adjustments, among which This indicates the coverage alarm threshold.

8. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, In step G, the evaluation report further outputs at least one of the following: 95th or 99th latency quantile, missing rate. Average coverage and risk-weighted coverage ratio ,in The weight is represented by the minimum TTC or minimum safety margin.

9. The time-delay distribution adaptive closed-loop decision coverage evaluation method according to claim 1, characterized in that, In step G, an evaluation report is generated and parameters are adjusted. in Parameter adjustments include: updating the delay distribution. Adjust the maximum buffer length Adjust the safety threshold Or adjust the minimum intervention projection / weighting coefficient.

Images