Definition of the autonomous driving operation design domain and online monitoring methods and systems

Abstract

This invention discloses a definition and online monitoring method and system for the operational design domain (ODD) of autonomous driving. Taking lane keeping function as an example, the method includes: acquiring an observation dataset; establishing a structural causal graph (OCD) between the lane detection accuracy of the autonomous driving system and the elements of the operational design domain based on the observation dataset; fitting the causal transmission path (i.e., structural causal equation) of the ODD based on the observed samples; acquiring various combinations of operating environments based on the operating parameter space of the autonomous vehicle; applying the acquired structural equation to perform counterfactual inference on the autonomous driving performance (taking lane detection accuracy as an example) under various combinations of operating environments; selecting an environment combination whose counterfactual calculation result is greater than a threshold to characterize the designed operational domain; if the lane detection accuracy in the autonomous driving performance indicators is consistently less than a set threshold, activating a function degradation strategy or prompting the driver to take over. This invention effectively improves the safety of autonomous vehicles through ODD online monitoring.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and autonomous driving technology, and in particular to a method, system, terminal, and computer-readable storage medium for defining and online monitoring of autonomous driving operation design domains. Background Technology

[0002] Autonomous vehicles face significant safety challenges in complex and unpredictable environments. To address this, the industry has proposed the concept of Operational Design Domain (ODD), which attempts to reduce traffic accidents by specifying the environment required for the safe operation of autonomous vehicles.

[0003] However, current ODD (Operational Disclosure) definition tasks often rely on subjective experience and lack clear operational guidelines. Furthermore, the driving environment of a vehicle is always dynamic, meaning that deployed autonomous vehicles frequently encounter conditions beyond the defined ODD range. Therefore, the design of a real-time ODD monitoring system is particularly important. This system should be able to assess online whether the current driving environment exceeds safety boundaries and switch to the lowest-risk mode (or alert the human driver to take over) when the autonomous driving system's performance deteriorates.

[0004] Currently, there is no suitable design scheme specifically for online monitoring of ODD (Optical Disk Domain). Only some studies have reported on the online verification problem of autonomous driving subsystems, but these schemes generally suffer from poor adaptability, high computational complexity, and excessive reliance on engineering experience, and cannot effectively handle the online monitoring task of ODD for autonomous vehicles.

[0005] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0006] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for defining and online monitoring of the operation design domain for autonomous driving, aiming to solve the problems of poor adaptability, high computational complexity, and excessive reliance on engineering experience in existing online monitoring schemes for the operation design domain.

[0007] To achieve the above objectives, the present invention provides a method for defining and online monitoring an autonomous driving operation design domain, the method comprising the following steps:

[0008] Obtain the observation dataset, establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fit the structural causal equation of the structural causal graph based on the observation samples.

[0009] Based on the operating parameter space of autonomous vehicles, various combinations of operating environments are obtained. The obtained structural causal equations are applied to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments. The combination of environments with counterfactual calculation results greater than a threshold is selected to characterize the design operating domain.

[0010] In actual deployment, if the lane detection accuracy in the autonomous driving performance indicators is consistently lower than the set threshold, a function degradation strategy will be activated or the driver will be prompted to take over. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories.

[0011] Optionally, the method for defining and online monitoring the autonomous driving operation design domain includes, in part, acquiring an observation dataset, establishing a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fitting a structural causal equation to the structural causal graph based on the observation samples.

[0012] The environmental variables to be examined are combined into conditions, and a certain number of images are captured for each combination;

[0013] The collected images are cleaned and labeled, and no-reference image quality assessment features are extracted from the images to obtain the observation dataset;

[0014] The causal structure discovery algorithm identifies key operational design domain elements from a high-dimensional parameter space.

[0015] Based on the observed dataset, a causal structure discovery method is used to establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain, and a structural causal equation is fitted to the structural causal graph based on the observed samples.

[0016] Optionally, the method for defining and monitoring the autonomous driving operation design domain further includes:

[0017] A directed acyclic graph G(v,ε) represents the causal relationships between a set of variables, where node v represents a random variable and edge ε represents a causal path between the variables.

[0018] The process of constructing a structural cause-effect graph starts from a blank graph G. init Start by gradually adding or deleting edges until the BIC scoring function converges.

[0019] Optionally, in the definition and online monitoring method for the autonomous driving operation design domain, the scoring function used is the Bayesian information criterion:

[0020] BIC=2·ln P(data|θ,G)-c·k·ln(n)

[0021] Where θ represents the parameters of the directed acyclic graph G(v,ε), c is a constant with a value of 1, k is the number of graph parameters, and n represents the sample size of the observed data;

[0022] The structural equation is:

[0023]

[0024] Where X represents the causal variable, H represents the mediating variable, and Y represents the response variable; U X U H U Y Let X, H, and Y be the values ​​of the exogenous variables corresponding to X, H, and Y, respectively; α, β, and γ represent the causal transmission path; for

[0025] Structural equations are fitted using linear or nonlinear functions;

[0026] When an autonomous vehicle faces a new operating environment, performing counterfactual calculations includes:

[0027] Find the closest approximation to the current operating environment X from the known observational data. new Record {X obs H obs To obtain the values ​​of the exogenous variable U, we can use the formulas ,Y}.

[0028]

[0029] Where obs represents the corresponding item in the observation dataset;

[0030] Set the causal variable as a variable in the current runtime environment, and define the new observation environment X. new Down The counterfactual result is expressed as:

[0031]

[0032] in, This indicates that the causal variable X takes the value X. new The value of the response variable Y; M x This represents a counterfactual causal model; u represents the specific value of the exogenous variable;

[0033] New Environment X new The counterfactual estimation results for the lane detection accuracy are as follows:

[0034]

[0035] Structural equations are parameterized representations of structural cause-effect graphs.

[0036] Optionally, the method for defining and online monitoring the autonomous driving operation design domain includes, in part, obtaining various combinations of operating environments based on the operating parameter space of the autonomous vehicle, applying the obtained structural causal equations to perform counterfactual inference on the autonomous driving performance under various combinations of operating environments, and selecting environment combinations whose counterfactual calculation results are greater than a threshold to characterize the design operating domain. Specifically, this includes:

[0037] The operation design domain is defined and modeled as a probability distribution of a safety subspace of the autonomous driving operating environment;

[0038] If φ all :=φ(X) is the joint probability density distribution function of a given parameter space, φ ODD :=φ(X|c>Thr) is the joint probability density distribution function of the subspace of the operation design domain, Thr is the vision-based lane detection accuracy threshold, and φ all and φ ODD These represent the joint probability density distributions of the given parameter space and the safety subspace of the operation design domain, respectively.

[0039] The dependency relationships between various marginal distributions are accurately modeled using the statistical mathematical tool Copulas.

[0040] If F(X) is the joint cumulative distribution function, then F(X) i If the marginal joint cumulative distribution function is given, then:

[0041] F(X) = C(F(X1),F(X2),…,F(X) d ))

[0042] Where C:[0,1] d →[0,1] represents the Copulas function, d represents the dimension, and the corresponding probability density function is:

[0043] φ=c(F(X1),F(X2),…,F(X d ))·f(X1)…f(X d )

[0044] Where: c(·) is the probability density of the Copulas function, and f(Xi) is the marginal joint probability density distribution function.

[0045] Optionally, in the definition and online monitoring method for the autonomous driving operation design domain, during actual deployment, if the lane detection accuracy in the autonomous driving performance indicators is consistently lower than a set threshold, a function degradation strategy is activated or the driver is alerted to take over. The function degradation strategy utilizes radar to infer the relative positions of vehicles ahead in the same lane and obtains a reference path through historical trajectory fitting, specifically including:

[0046] The Extended Kalman Filter (EPF) combines radar information as observations with a prediction model. The prediction and measurement equations of the EPF are expressed as follows:

[0047]

[0048]

[0049] Where, x t Let x represent the state vector. t = [x,y,θ,v] represent the longitudinal distance, lateral distance, heading angle, and velocity of the preceding vehicle relative to the preceding vehicle, respectively, z t Let Q represent the observation vector, and R represent the noise matrices in the prediction model and measurement model, respectively. H is the unit measurement equation matrix, T represents the sampling time, and F represents the state transition matrix.

[0050] Within one time step, the vehicle's motion is approximated as uniform circular motion, and the historical positions x′ and y′ of the vehicle ahead in the main lane are transformed into the current coordinate system of the vehicle:

[0051]

[0052] Where Δx and Δy represent the position changes in the vehicle coordinate system, and α is the heading angle offset, which is updated by the following equation:

[0053]

[0054] Where, ω ego and ω′ ego These are the yaw rates at the current time step and the previous time step, respectively.

[0055] Polynomial curve fitting is used to convert the points into trajectory points in the current master vehicle coordinate system, and the fitted curve is used as the reference path.

[0056] Optionally, in the method for defining and monitoring the autonomous driving operation design domain, the prediction equation is used to estimate the system state at the next time step based on the system's dynamic model; the measurement equation is used to compare sensor measurement data with the predicted state and generate a corrected estimate of the system state; the measurement equation corrects errors in the prediction by comparing measurement information with the predicted state.

[0057] Furthermore, to achieve the above objectives, the present invention also provides a definition and online monitoring system for the autonomous driving operation design domain, wherein the system comprises:

[0058] The causal graph construction module is used to acquire the observation dataset, establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fit the structural causal equation of the structural causal graph based on the observation samples.

[0059] The counterfactual reasoning module is used to obtain various combinations of operating environments based on the operating parameter space of autonomous vehicles, apply the obtained structural causal equations to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments, and select the combination of environments where the counterfactual calculation result is greater than a threshold to characterize the design operating domain.

[0060] The monitoring and degradation module is used in actual deployment to activate a function degradation strategy or remind the driver to take over if the lane detection accuracy in the autonomous driving performance indicators is consistently lower than a set threshold. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories.

[0061] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a definition and online monitoring program for an autonomous driving operation design domain stored in the memory and executable on the processor, wherein when the definition and online monitoring program for an autonomous driving operation design domain is executed by the processor, it implements the steps of the definition and online monitoring method for an autonomous driving operation design domain as described above.

[0062] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a definition and online monitoring program for an autonomous driving operation design domain, and the definition and online monitoring program for an autonomous driving operation design domain, when executed by a processor, implements the steps of the definition and online monitoring method for an autonomous driving operation design domain as described above.

[0063] This invention acquires an observation dataset, establishes a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operational design domain based on the observation dataset, and fits the structural causal equation of the structural causal graph based on the observed samples. Various combinations of operating environments are obtained based on the operating parameter space of the autonomous vehicle. The acquired structural causal equation is used to perform counterfactual inference on the autonomous driving performance under various operating environment combinations. Environment combinations with counterfactual calculation results greater than a threshold are selected to represent the designed operating domain. In actual deployment, if the lane detection accuracy in the autonomous driving performance indicators is consistently less than a set threshold, a function degradation strategy is activated or the driver is alerted to take over. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path through historical trajectory fitting. This invention is the first in the industry to propose the application of causal inference theory to deeply explore the causal relationship between environmental conditions and algorithm performance, thereby enabling a more accurate assessment of the impact of different environmental factors on the algorithm, making it more scientific and reliable than existing technologies. Attached Figure Description

[0064] Figure 1 This is a flowchart of a preferred embodiment of the present invention, which defines and monitors the operation design domain for autonomous driving.

[0065] Figure 2 This is a method framework diagram of a preferred embodiment of the present invention for defining and online monitoring methods for autonomous driving operation design domains;

[0066] Figure 3 This is a schematic diagram of the observation dataset construction process in a preferred embodiment of the definition and online monitoring method for autonomous driving operation design domain of the present invention;

[0067] Figure 4 This is a schematic diagram of the cause-effect graph structure discovery step in a preferred embodiment of the definition and online monitoring method for autonomous driving operation design domain of the present invention;

[0068] Figure 5 This is a schematic diagram of the original causal graph M in a preferred embodiment of the causal counterfactual reasoning schematic diagram of the definition and online monitoring method for the autonomous driving operation design domain of the present invention;

[0069] Figure 6 This is a schematic diagram of a modified causal counterfactual causal graph Mx, which is a preferred embodiment of the causal counterfactual reasoning diagram of the definition and online monitoring method for the autonomous driving operation design domain of the present invention.

[0070] Figure 7 This is a schematic diagram of the ODD definition process in a preferred embodiment of the method for defining and online monitoring the autonomous driving operation design domain according to the present invention;

[0071] Figure 8This is a schematic diagram of the ODD online monitoring algorithm based on causal counterfactual in a preferred embodiment of the definition and online monitoring method for autonomous driving operation design domain of the present invention;

[0072] Figure 9 This is a schematic diagram of a function-degraded lane fitting strategy based on radar information in a preferred embodiment of the definition of the autonomous driving operation design domain and the online monitoring system of the present invention;

[0073] Figure 10 This is a schematic diagram illustrating the updating of the historical position of the preceding vehicle to the current coordinates of the autonomous vehicle in a preferred embodiment of the definition of the autonomous driving operation design domain and the online monitoring system of the present invention.

[0074] Figure 11 This is a schematic diagram illustrating the principle of a preferred embodiment of the present invention, which defines the autonomous driving operation design domain and provides an online monitoring system.

[0075] Figure 12 This is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0076] Currently, the ODD (Operational Design Principle) definition methods for autonomous vehicles are generally vague and lack universal solutions. In addition, existing online monitoring solutions, such as skill maps and diagnostic maps, cannot fully handle dynamic road conditions; and the computational complexity of temporal logic is too high, posing a significant challenge to practical applications in complex scenarios.

[0077] This invention proposes a method for defining and monitoring the ODD (Optical Distributed Domain) of autonomous vehicles based on causal inference. This method can reasonably, stably, and scalably identify key ODD elements that affect the performance of autonomous driving systems and characterize the ODD space in a probability distribution manner. Based on causal counterfactual reasoning, it achieves performance evaluation of autonomous driving systems under random dynamic environments with minimal computational burden, thereby establishing an online ODD monitor. Taking lane keeping as an example, this invention further proposes a novel function degradation strategy based on the historical trajectory of the preceding vehicle.

[0078] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0079] The preferred embodiment of the present invention describes a method for defining and monitoring the operational design domain for autonomous driving, such as... Figure 1 and Figure 2 As shown, the definition and online monitoring method for the autonomous driving operation design domain includes the following steps:

[0080] Step S10: Obtain the observation dataset, establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fit the structural causal equation of the structural causal graph based on the observation samples.

[0081] Specifically, the environmental variables to be examined are combined into conditions, and a certain number of images are captured for each combination; the collected images are cleaned and labeled, and no-reference image quality assessment features are extracted from the images to obtain the observation dataset; key operational design domain elements are identified from the high-dimensional parameter space based on the causal structure discovery algorithm; according to the observation dataset, the causal structure discovery method is used to establish a structural causal graph between the lane line detection accuracy of the autonomous driving system and the operational design domain elements, and the structural causal equation of the structural causal graph is fitted according to the observation samples.

[0082] This invention first identifies critical operation design domain (ODD) elements from a high-dimensional parameter space based on a causal structure discovery algorithm, and then establishes a causal graph between lane detection accuracy and ODD elements. Typically, this requires obtaining randomized controlled trial data, where the environmental variables to be examined are randomly assigned to either the control or analysis group. However, obtaining randomized controlled trials is costly or even impractical.

[0083] Therefore, this invention proposes using the combined testing principle to collect the observation dataset (i.e. Figure 2 (Observational datasets in the dataset). Specifically, such as... Figure 3 As shown, the environmental variables to be examined are combined into conditions, and a certain number of images are captured for each combination. Subsequently, the collected data are cleaned and labeled, and no-reference image quality assessment features are extracted from the images to measure the impact of various environmental factors on image quality.

[0084] Furthermore, based on the observed dataset and incorporating domain knowledge, a causal structure diagram of the impact of environmental factors on perception performance is constructed, and a structural causal equation is fitted to the structural causal diagram based on the observed samples.

[0085] Specifically, based on the collected observation dataset, this invention uses a causal structure discovery method to construct a structural causal graph represented by a Bayesian diagram. This method focuses on describing the causal relationships between a set of variables, represented by a directed acyclic graph G(v,ε), where nodes ν represent random variables and edges ε represent causal paths between variables. The process of constructing the causal graph starts from a blank graph G. init Start by gradually adding or deleting edges until the BIC scoring function converges. The construction process is as follows: Figure 4 As shown, begin: give empty graph G initAdd an edge. Step 1: Check for cycles or disconnections. Step 2: Determine if the BIC score has increased. Step 3: If the score has increased, add or delete the edge and update the causal structure, then loop back to step 1; otherwise, end.

[0086] The scoring function used in this invention is the Bayesian Information Criterion (BIC):

[0087] BIC=2·ln P(data|θ,G)-c·k·ln(n)

[0088] Where θ represents the parameters of the directed acyclic graph G(ν,ε), c is a constant with a value of 1, k is the number of graph parameters, and n represents the sample size of the observed data.

[0089] Generally, a higher BIC value means that the discovered causal graph can provide a more reliable causal explanation for the observed dataset without unnecessarily complicating it. It's important to note that in practical applications, expert verification can still be used to prune the causal graph to obtain even more reliable results.

[0090] Subsequently, the established causal graph provides a valuable analytical foundation for lane detection accuracy. In practical applications, it is always necessary to quantitatively describe the causal relationships between nodes in the causal graph. Structural equation modeling, as a mathematical representation, describes the fundamental mechanisms and dependencies of interactions between endogenous variables in a causal model. These equations embody the original mechanisms and the connections between cause and response. Instantiating structural equations through nonlinear regression models theoretically lays the computational mathematical foundation for counterfactual reasoning.

[0091] Causal counterfactual reasoning aims to infer the potential impact on lane detection accuracy when the values ​​of certain ODD factors change. This enables, for the first time, the quantitative and interpretable calculation of lane detection performance under different operating conditions from a causal perspective, without the need for extensive testing. For ease of explanation, it is assumed that the following exists... Figure 4 The cause-effect graph structure shown.

[0092] like Figure 5 As shown, the structural equation is:

[0093]

[0094] Where X represents the causal variable, H represents the mediating variable, and Y represents the response variable; U X U H U Y Let X, H, and Y be the values ​​of the exogenous variables corresponding to X, H, and Y, respectively; α, β, and γ represent the causal transmission path; for

[0095] Structural equations are fitted using linear or nonlinear functions.

[0096] When an autonomous vehicle faces a new operating environment, performing counterfactual calculations includes:

[0097] Find the closest approximation to the current operating environment X from the known observational data. new Record {X obs H obs To obtain the values ​​of the exogenous variable U, we can use the formulas ,Y}.

[0098]

[0099] Here, obs represents the corresponding item in the observation dataset.

[0100] Set the causal variable to the current runtime environment variable, that is, modify the causal graph, such as... Figure 6 As shown. The new observation environment X new Down The counterfactual result is expressed as:

[0101]

[0102] in, This indicates that the causal variable X takes the value X. new The value of the response variable Y; M x Let represent the counterfactual causal model; u represents the specific value of the exogenous variable. Using the modified counterfactual causal model and the values ​​of the exogenous variables obtained in the first step, we can answer the following counterfactual question: "What will be the lane detection accuracy when an autonomous vehicle encounters a new ODD environment instead of known observational data?"

[0103] according to Figure 6 Cause-and-effect diagram, new environment X new The counterfactual estimation results for the lane detection accuracy are as follows:

[0104]

[0105] In summary, structural equations (i.e., causal transit path functions fitted using linear / nonlinear models, α, β, and γ) are parameterized representations of causal graphs, with each equation explaining the process by which a causal variable acquires its value.

[0106] This invention provides a prediction method based on causal reasoning theory. It compares historical data to infer the algorithm's performance under new operating conditions. Once autonomous vehicles possess this capability, it's equivalent to adding a real-time monitor. If the estimated algorithm performance falls short of expectations, it can promptly alert humans to take over or activate backup plans. This will help further improve the safety of autonomous vehicles.

[0107] Step S20: Obtain various combinations of operating environments based on the operating parameter space of the autonomous vehicle, apply the obtained structural causal equations to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments, and select the combination of environments with counterfactual calculation results greater than a threshold to characterize the design operating domain.

[0108] Specifically, this invention proposes to define and model the ODD (Optical Distribution Domain) as a probability distribution of a safe subspace of the autonomous driving operating environment, meaning that autonomous vehicles can operate safely under various combinations of conditions sampled within the safe subspace. Unlike existing fuzzy specifications, a performance surrogate index is defined. ( To help define the ODD space, we can use the field representing the positive real number field, such as... Figure 7 As shown in the definition of ODD, the various combinations of environmental elements are no longer uniformly sampled, but sampled according to the actual distribution. For example, the annual rainfall is not uniformly distributed, but follows a certain probability distribution.

[0109] Where, assuming φ all :=φ(X) is the joint probability density function (PDF) of a given parameter space, φ ODD :=φ(X|c>Thr) is the joint probability density distribution function of the subspace of the operation design domain, Thr is the vision-based lane detection accuracy threshold, and φ all and φ ODD These represent the joint probability density distributions of the given parameter space and the safety subspace of the operation design domain, respectively.

[0110] In order to obtain the probability distribution φ in the ODD space ODD Using causal counterfactual reasoning techniques, it is inferred that from a given parameter space φ all The accuracy of lane detection under the generated environment combination is assessed, and samples with predicted values ​​greater than a given threshold are used to reconstruct φ. ODD .

[0111] It is worth noting that the marginal distributions of ODD elements are not independent, which means that the marginal probability distributions cannot be simply multiplied together to obtain φ. all and φ ODD To this end, this invention introduces the statistical mathematical tool Copulas to accurately model the dependencies between various marginal distributions.

[0112] If F(X) is the joint cumulative distribution function (CDF), then F(X) i If is the marginal joint cumulative distribution function, then:

[0113] F(X) = C(F(X1),F(X2),…,F(X) d ))

[0114] Where C:[0,1] d →[0,1] represents the Copulas function, d represents the dimension, and the corresponding probability density function is:

[0115] φ=c(F(X1),F(X2),…,F(X d ))·f(X1)…f(X d )

[0116] Where: c(·) is the probability density of the Copulas function, f(X) i ) is the marginal joint probability density distribution function.

[0117] Step S30: In actual deployment, if the lane line detection accuracy in the autonomous driving performance indicators is consistently less than the set threshold, a function degradation strategy is activated or the driver is reminded to take over. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories.

[0118] Specifically, based on the aforementioned causal graph structure and counterfactual calculation method, this invention proposes a novel online monitoring method for ODD (Autonomous Vehicle Descriptor). If the autonomous driving performance consistently falls below a set threshold, a function degradation strategy is activated or the driver is alerted to take over, thereby improving the safety of the autonomous vehicle. Figure 8 As shown, counter and flag represent the counter cache variable and the flag indicating whether the function degradation strategy is enabled, respectively. Initially, the variables counter and flag are set to 0, and the current driving environment X is obtained. new Check if counter is less than or equal to 5. If it is, set flag to True and end. If it is, find the match between counter and X in the observed dataset. new The closest match is used to further infer lane line detection accuracy using causal counterfactual analysis. Then determine whether If so, increment counter by 1, and then re-acquire the current driving environment X. new If not, then reacquire the current driving environment X. new .

[0119] If the autonomous driving performance consistently falls below a set threshold, a function degradation strategy will be activated or the driver will be prompted to take over to improve the safety of the autonomous vehicle. The autonomous driving performance refers to the use of multi-dicescore scores as an indicator to measure the accuracy of visual lane line detection.

[0120] Causal inference theory provides a formal framework for modeling the causal relationships between the performance of autonomous driving systems and different ODD (Optical Distribution Components). This study utilizes counterfactual reasoning techniques to derive the performance of autonomous vehicles in unknown and stochastic dynamic environments, thereby addressing the problem of existing research over-reliance on engineering experience in ODD definition and online monitoring.

[0121] Specifically, millimeter-wave radar is highly effective in various weather conditions due to its ability to penetrate fog, smoke, and dust. Therefore, this invention proposes a lane-keeping strategy for performance degradation to address the problem of continuously declining performance of visual perception algorithms when the operating environment exceeds the ODD (Operational Distance Detection) range. This strategy utilizes radar to infer the relative positions of vehicles ahead in the same lane and derives a reference path through historical trajectory fitting. The detailed processing flow is as follows... Figure 9 As shown, the input radar point cloud information is mapped to the pixel coordinate system, the target region of interest is obtained based on the DBSCAN clustering algorithm, the position of the cluster center is then extracted, the trajectory points are stored in the queue through extended Kalman filtering and coordinate transformation is performed, and finally the output is subjected to polynomial fitting to obtain the reference path.

[0122] The Extended Kalman Filter (EKF) combines radar information as observations (which can be understood as the location of cluster centers) with the prediction model. The prediction and measurement equations of the Extended Kalman Filter are expressed as follows:

[0123]

[0124]

[0125] Where, x t Let x represent the state vector. t = [x,y,θ,v] represent the longitudinal distance, lateral distance, heading angle, and velocity of the preceding vehicle relative to the preceding vehicle, respectively, z t Let Q represent the observation vector, and R represent the noise matrices in the prediction and measurement models, respectively. H is the unit measurement equation matrix, T represents the sampling time, and F represents the state transition matrix. Considering that the vehicle is in continuous motion, it is necessary to convert the historical trajectory coordinates of the preceding vehicle to the current master vehicle coordinate system in real time. Figure 10 As shown, X′O′Y′ and YOY are the coordinate systems of the previous and current timestamps of the vehicle, respectively.

[0126] The prediction equation is used to estimate the system state at the next time step based on the system's dynamic model. It generates a predicted state based on previous state estimates and the system's control inputs. By considering the system's dynamics, the prediction equation allows filters to update the system state over time to predict future states. The observation equation compares sensor measurement data with the predicted state and generates a corrected estimate of the system state. It considers measurement noise and uncertainties in the system state to generate a corrected state estimate. The observation equation improves the accuracy of the state estimate by comparing measurement information with the predicted state and correcting errors in the prediction.

[0127] Within one time step, the vehicle's motion is approximated as uniform circular motion, and the historical positions x′ and y′ of the vehicle ahead in the main lane are transformed into the current coordinate system of the vehicle:

[0128]

[0129] Where Δx and Δy represent the changes in the vehicle's position in the coordinate system, and α is the heading angle offset, which is updated using the following equation:

[0130]

[0131] Where, ω ego and ω′ ego yaw rate at the current time step and the previous time step, respectively, and T represents the sampling time.

[0132] The curve is converted into a trajectory point in the current master vehicle coordinate system using polynomial curve fitting. The fitted curve is then used as a reference path for the downstream lane keeping control algorithm.

[0133] The operating environment of autonomous driving is dynamic and random, and it is impossible to cover all driving conditions simply through testing. When encountering new conditions never before seen by autonomous vehicles, the key point of this invention is to propose a method for defining and monitoring the ODD (Operational Distributed Domain) of autonomous vehicles based on causal reasoning. This includes qualitative mining of causal relationships, instantiation of causal transmission paths (represented by structural equations), defining the probability distribution of the ODD space from the global parameter space, designing an online ODD monitor based on counterfactual computation, and a lane-keeping function degradation strategy based on radar information. In summary, this invention proposes a novel method for defining and monitoring ODD, which can effectively improve the safety of autonomous vehicles.

[0134] Compared with the best existing technology, the present invention has the following advantages (i.e. beneficial effects):

[0135] (1) Deeply explore causal relationships: This invention is the first in the industry to propose the application of causal inference theory to deeply explore the causal relationship between environmental conditions and algorithm performance, so as to more accurately evaluate the impact of different environmental factors on the algorithm, which is more scientific and reliable than existing technologies.

[0136] (2) Reasonable ODD definition method: This invention is the first in the industry to propose a clear and quantitative definition method for ODD, which solves the shortcomings of existing methods that mainly rely on fuzzy specifications and engineering experience.

[0137] (3) Explainable ODD real-time online monitoring method: This invention is the first in the industry to propose an explainable, easily scalable, and low computational complexity ODD online monitoring method. This method uses the aforementioned causal structure and instantiates the structural equation, and provides a method for predicting the performance of autonomous driving in a novel environment using causal counterfactuals.

[0138] (4) A novel lane-keeping function degradation strategy: This invention utilizes all-weather radar to record the historical trajectory information of the preceding vehicle and converts it to the current coordinate system of the current vehicle in real time. Subsequently, a reference trajectory for lane keeping is obtained by multinomial fitting. It has high accuracy and good real-time performance, and has good potential as a function degradation strategy when the visual algorithm declines.

[0139] In summary, this invention has significant advantages in causal relationship mining, ODD definition methods, interpretable ODD real-time online monitoring methods, and lane keeping function degradation strategies, and can more accurately evaluate and verify the performance of visual algorithms in complex environments.

[0140] In addition, the method described in this invention has been subjected to a large number of experiments. Compared with the scheme without ODD definition and monitoring, the accident rate of autonomous vehicles under this scheme has decreased by three orders of magnitude.

[0141] Furthermore, such as Figure 11 As shown, based on the above-described definition and online monitoring method for autonomous driving operation design domains, this invention also provides a corresponding definition and online monitoring system for autonomous driving operation design domains, wherein the system includes:

[0142] The cause-effect graph construction module 51 is used to acquire the observation dataset, establish a structural cause-effect graph between the lane line detection accuracy of the autonomous driving system and the operation design domain elements based on the observation dataset, and fit the structural cause-effect equation of the structural cause-effect graph based on the observation samples.

[0143] The counterfactual reasoning module 52 is used to obtain various combinations of operating environments based on the operating parameter space of the autonomous vehicle, apply the obtained structural causal equations to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments, and select the combination of environments where the counterfactual calculation result is greater than the threshold to characterize the design operating domain.

[0144] The monitoring degradation module 53 is used to, during actual deployment, activate a function degradation strategy or remind the driver to take over if the lane detection accuracy in the autonomous driving performance indicators is continuously less than a set threshold. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories.

[0145] Furthermore, such as Figure 12 As shown, based on the above definition of the autonomous driving operation design domain and the online monitoring method and system, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 12 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0146] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a definition and online monitoring program 40 for the autonomous driving operation design domain. This definition and online monitoring program 40 for the autonomous driving operation design domain can be executed by the processor 10 to implement the definition and online monitoring method for the autonomous driving operation design domain in this application.

[0147] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the definition of the autonomous driving operation design domain and the online monitoring method.

[0148] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0149] In one embodiment, when the processor 10 executes the definition and online monitoring program 40 for the autonomous driving operation design domain in the memory 20, it implements the steps of the definition and online monitoring method for the autonomous driving operation design domain as described above.

[0150] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a definition and online monitoring program for an autonomous driving operation design domain, and the definition and online monitoring program for an autonomous driving operation design domain, when executed by a processor, implements the steps of the definition and online monitoring method for an autonomous driving operation design domain as described above.

[0151] In summary, this invention provides a method and related equipment for defining and online monitoring the operational design domain (ODD) of autonomous driving. The method includes: acquiring an observation dataset; establishing a structural causal graph (SCR) between the lane detection accuracy of the autonomous driving system and the elements of the operational design domain based on the observation dataset; fitting a SCR equation to the SCR based on observed samples; acquiring various combinations of operating environments based on the autonomous vehicle's operating parameter space; applying the acquired SCR equation to perform counterfactual inference on the autonomous driving performance under various operating environment combinations; selecting an environment combination where the counterfactual calculation result is greater than a threshold to characterize the designed operational domain; and in actual deployment, if the lane detection accuracy in the autonomous driving performance indicators is consistently less than a set threshold, activating a function degradation strategy or prompting the driver to take over. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path through historical trajectory fitting. This invention obtains a reference trajectory for lane keeping through online monitoring of the ODD, achieving high accuracy and good real-time performance, which can effectively improve the safety of autonomous vehicles.

[0152] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0153] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0154] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for defining and online monitoring of a design domain for autonomous driving operation, characterized in that, The definition and online monitoring method for the autonomous driving operation design domain include: Obtain the observation dataset, establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fit the structural causal equation of the structural causal graph based on the observation samples. Based on the operating parameter space of autonomous vehicles, various combinations of operating environments are obtained. The obtained structural causal equations are applied to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments. The combination of environments with counterfactual calculation results greater than a threshold is selected to characterize the design operating domain. In actual deployment, if the lane detection accuracy in the autonomous driving performance indicators is consistently less than the set threshold, a function degradation strategy will be activated or the driver will be prompted to take over. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories. The process of acquiring the observation dataset, establishing a structural causal graph between the lane detection accuracy of the autonomous driving system and the operational design domain elements based on the observation dataset, and fitting the structural causal equation of the structural causal graph based on the observation samples specifically includes: The environmental variables to be examined are combined into conditions, and a certain number of images are captured for each combination; The collected images are cleaned and labeled, and no-reference image quality assessment features are extracted from the images to obtain the observation dataset; The causal structure discovery algorithm identifies key operational design domain elements from a high-dimensional parameter space. Based on the observed dataset, a causal structure discovery method is used to establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain, and a structural causal equation is fitted to the structural causal graph based on the observed samples. The definition and online monitoring method for the autonomous driving operation design domain also include: Through a directed acyclic graph Represents a causal relationship between a set of variables, where nodes Represents a random variable, edge Indicates the causal path between variables; The process of constructing a structural cause-effect graph starts from a blank graph. Begin by adding or deleting edges gradually until... The scoring function has converged; The scoring function used is the Bayesian information criterion: in, Represents a directed acyclic graph The parameters, It is a constant that takes the value 1. It refers to the number of graphics parameters. n Indicates the sample size of the observed data; The structural equation is: in, Indicates causal variables, Indicates the mediating variable. Represents the response variable; , , They are respectively , , The corresponding values ​​of exogenous variables; , and Indicates the causal transmission path; for =0; Structural equations are fitted using linear or nonlinear functions; When an autonomous vehicle faces a new operating environment, performing counterfactual calculations includes: Find the closest approximation to the current operating environment from known observational data. Records To obtain exogenous variables The possible values ​​of: in, Represents the corresponding item in the observed dataset; Set the causal variable as a variable in the current runtime environment, and the new observation environment. Down The counterfactual result is expressed as: in, Representing causal variables Values Time response variable The value of ; This represents a counterfactual causal model; Indicates the specific values ​​that the exogenous variable can take; New environment The counterfactual estimation results for the lane detection accuracy are as follows: Structural equations are parameterized representations of structural cause-effect graphs.

2. The method for defining and online monitoring the autonomous driving operation design domain according to claim 1, characterized in that, The process involves obtaining various combinations of operating environments based on the operating parameter space of the autonomous vehicle, applying the obtained structural causal equations to perform counterfactual inference on the autonomous driving performance under various combinations of operating environments, and selecting environment combinations whose counterfactual calculation results are greater than a threshold to characterize the design operating domain. Specifically, this includes: The operation design domain is defined and modeled as a probability distribution of a safety subspace of the autonomous driving operating environment; like Let be the joint probability density distribution function of the given parameter space. To design the joint probability density distribution function of the subspace of the operation domain, It is a vision-based lane line detection accuracy threshold. as well as These represent the joint probability density distributions of the given parameter space and the safety subspace of the operation design domain, respectively. Using statistical mathematical tools Copulas Accurately model the dependencies between various edge distributions; like For the joint cumulative distribution function, Let be the marginal joint cumulative distribution function, then we have: in, for Copulas function, The dimension is represented by the following probability density function: in: yes Copulas The probability density of the function, It is the marginal joint probability density distribution function.

3. The method for defining and online monitoring the autonomous driving operation design domain according to claim 2, characterized in that, In actual deployment, if the lane detection accuracy in the autonomous driving performance indicators remains below a set threshold, a function degradation strategy is activated or the driver is prompted to take over. The function degradation strategy uses radar to infer the relative positions of vehicles ahead in the same lane and obtains a reference path through historical trajectory fitting, specifically including: The Extended Kalman Filter (EPF) combines radar information as observations with a prediction model. The prediction and measurement equations of the EPF are expressed as follows: in, Represents the state vector, state vector = These represent the longitudinal distance, lateral distance, heading angle, and speed of the vehicle in front relative to the vehicle behind. Represents the observation vector. and These represent the noise matrices in the prediction model and the measurement model, respectively. It is the unit measurement equation matrix. Indicates the sampling time. Represents the state transition matrix; Within a time step, approximating the vehicle's motion as uniform circular motion, the historical position of the vehicle ahead in the main lane... 'and Transform to the current vehicle's coordinate system: in, and These represent the changes in the vehicle's position on the coordinate system. The heading angle offset is updated using the following equation: in, as well as These are the yaw rates at the current time step and the previous time step, respectively. Polynomial curve fitting is used to convert the points into trajectory points in the current master vehicle coordinate system, and the fitted curve is used as the reference path.

4. The method for defining and online monitoring the autonomous driving operation design domain according to claim 3, characterized in that, The prediction equation is used to estimate the system state at the next time step based on the system's dynamic model; the measurement equation is used to compare sensor measurement data with the predicted state and generate a corrected estimate of the system state; the measurement equation corrects errors in the prediction by comparing measurement information with the predicted state.

5. A system for defining and online monitoring the operational design domain of autonomous driving, characterized in that, The system for defining and monitoring the autonomous driving operation design domain is used to implement the method for defining and monitoring the autonomous driving operation design domain as described in any one of claims 1-4, wherein the system comprises: The causal graph construction module is used to acquire the observation dataset, establish a structural causal graph between the lane detection accuracy of the autonomous driving system and the elements of the operation design domain based on the observation dataset, and fit the structural causal equation of the structural causal graph based on the observation samples. The counterfactual reasoning module is used to obtain various combinations of operating environments based on the operating parameter space of autonomous vehicles, apply the obtained structural causal equations to perform counterfactual reasoning on the autonomous driving performance under various combinations of operating environments, and select the combination of environments where the counterfactual calculation result is greater than a threshold to characterize the design operating domain. The monitoring and degradation module is used in actual deployment to activate a function degradation strategy or remind the driver to take over if the lane detection accuracy in the autonomous driving performance indicators is consistently lower than a set threshold. The function degradation strategy uses radar to infer the relative position of vehicles ahead in the same lane and obtains a reference path by fitting historical trajectories.

6. A terminal, characterized in that, The terminal includes: a memory, a processor, and a definition and online monitoring program for autonomous driving operation design domains stored in the memory and executable on the processor. When the definition and online monitoring program for autonomous driving operation design domains is executed by the processor, it implements the steps of the definition and online monitoring method for autonomous driving operation design domains as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a definition and online monitoring program for an autonomous driving operation design domain, which, when executed by a processor, implements the steps of the method for defining and monitoring an autonomous driving operation design domain as described in any one of claims 1-4.

Images