Automatic driving control method, device and equipment

Abstract

The application discloses an automatic driving control method, device and equipment, relates to the technical field of automatic driving, and discloses an automatic driving control method, which comprises the following steps: acquiring multi-source perception information and running state information of a vehicle-mounted sensor; determining confidence degree information of the vehicle-mounted sensor in a sub-region according to environment state information and the running state information; determining a fusion weight of the vehicle-mounted sensor in a corresponding sub-region according to the environment state information and the confidence degree information; performing fusion processing on the multi-source perception information according to the fusion weight to obtain regional fusion perception features; and performing automatic driving control according to the regional fusion perception features. The scheme can improve the accuracy of perception recognition in the automatic driving control process.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to autonomous driving control methods, devices and equipment. Background Technology

[0002] In autonomous driving scenarios, some solutions involve deploying multiple sensing devices throughout the vehicle and using fixed weights to perform weighted data fusion to achieve comprehensive vehicle perception. However, this approach cannot match sensor perception with the characteristics of the vehicle's driving environment, leading to inaccurate perception and recognition.

[0003] Therefore, how to improve the accuracy of perception and recognition during autonomous driving control has become an issue that needs to be addressed.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide an autonomous driving control method, device, and equipment, which aims to solve the technical problem of how to improve the accuracy of perception and recognition during autonomous driving control.

[0006] To achieve the above objectives, this application proposes an autonomous driving control method, which includes: Acquire multi-source sensing information and operational status information from vehicle-mounted sensors; The confidence information of the vehicle-mounted sensor in the sub-region is determined based on the environmental status information and the operational status information. The fusion weights of the vehicle-mounted sensors in the corresponding sub-regions are determined based on the environmental state information and the confidence information. The multi-source sensing information is fused according to the fusion weights to obtain regional fused sensing features; Autonomous driving control is performed based on the region fusion perception features.

[0007] In one embodiment, the step of determining the confidence information of the vehicle-mounted sensor in a sub-region based on environmental state information and operational state information includes: Obtain sub-region characteristic information; Determine the environmental state influence vector based on environmental state information; Determine sensor status assessment information based on the aforementioned operating status information; The environmental state influence vector, the sensor state evaluation information, and the sub-region characteristic information are concatenated to obtain a combined feature vector. The confidence information of the vehicle sensor in the sub-region is determined based on the combined feature vector and the preset weight parameters.

[0008] In one embodiment, the step of determining the fusion weights of the vehicle-mounted sensors in the corresponding sub-region based on the environmental state information and the confidence information includes: Determine the current environmental state vector based on the environmental state information; The fusion strategy selection factor is determined based on the confidence information and the current environment state vector; The fusion weights of the vehicle sensors in the corresponding sub-regions are determined based on the confidence information and the fusion strategy selection factor.

[0009] In one embodiment, the step of fusing the multi-source sensing information according to the fusion weights to obtain regional fused sensing features includes: The multi-source sensing information is weighted and fused according to the fusion weights to obtain weighted fusion features; The dynamic coupling term is determined based on the confidence information, the fusion weight, and the multi-source sensing information. The region fusion sensing features are determined based on the weighted fusion features and the dynamic coupling term.

[0010] In one embodiment, the method further includes: Extract a global feature library based on the multi-source sensing information; Conflict detection is performed on the global feature library to obtain conflict detection results; When the conflict detection result indicates the existence of a systematic conflict, the target sensor subset is determined based on the global sensor set; Based on the global feature library and security task information, determine the task allocation information corresponding to the target sensor subset; Autonomous driving control is performed based on the target sensor subset and the task allocation information.

[0011] In one embodiment, the step of determining the task allocation information corresponding to the target sensor subset based on the global feature library and security task information includes: Obtain the feature matrix and feature complementarity scores corresponding to the target sensor subset; Based on the security task information, obtain the task importance weight corresponding to the security task; The task performance score corresponding to the target sensor subset is determined based on the feature matrix and the global feature library. The task allocation information corresponding to the target sensor subset is determined based on the feature complementarity score, the task importance weight, and the task performance score.

[0012] In one embodiment, the step of performing autonomous driving control based on the region fusion perception features includes: Obtain current vehicle operation scenario information and local baseline policies; The scenario compensation increment is determined based on the cloud-based herd immunity network and the current vehicle operation scenario information. The target action vector is determined based on the local baseline strategy and the scene compensation increment. Automated driving control is performed based on the target motion vector.

[0013] In one embodiment, the step of determining the scene compensation increment based on the current vehicle operating scene information further includes: Common failure modes are obtained from the cloud-based herd immunity network to obtain a set of common failure modes; The plugin complexity penalty term is determined based on the preset regularization coefficient and the plugin complexity regularization term; The scenario compensation increment is determined based on the current vehicle operation scenario information, the cloud-based herd immunity network, the common failure mode set, and the plugin complexity penalty term.

[0014] Furthermore, to achieve the above objectives, this application also proposes an automatic driving control device, which includes: The data acquisition module is used to acquire multi-source sensing information and operating status information from vehicle-mounted sensors; The data processing module is used to determine the confidence information of the vehicle-mounted sensor in the sub-region based on the environmental state information and the operating state information; The data processing module is further configured to determine the fusion weight of the vehicle-mounted sensor in the corresponding sub-region based on the environmental state information and the confidence information; The data fusion module is used to fuse the multi-source sensing information according to the fusion weight to obtain regional fusion sensing features; The driving control module is used to perform autonomous driving control based on the region fusion perception features.

[0015] In addition, to achieve the above objectives, this application also proposes an autonomous driving control device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the autonomous driving control method as described above.

[0016] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the autonomous driving control method described above.

[0017] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the autonomous driving control method described above.

[0018] One or more technical solutions proposed in this application have at least the following technical effects: By collecting multi-source perception information and operational status information from vehicle sensors, and combining this with the external environmental status to calculate the confidence level of each sensor in different regions, and then dynamically generating differentiated fusion weights based on the confidence levels, the weighted fusion is used to obtain regional fusion perception features. Based on these regional fusion perception features, autonomous driving control is executed. This approach can match driving environments under different conditions, adapt to the real-time working status of different vehicle sensors, fully leverage the complementary advantages of different types of vehicle sensors, improve the accuracy of perception and recognition during autonomous driving control, and ensure that the vehicle's environmental perception results closely match real driving scenarios. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating an embodiment of the autonomous driving control method of this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the autonomous driving control method of this application; Figure 3 A simplified flowchart illustrating the autonomous driving control method provided in Embodiment 2 of this application; Figure 4 This is a schematic diagram of the module structure of the autonomous driving control device according to an embodiment of this application; Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the autonomous driving control method in this application embodiment.

[0022] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0024] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0025] The main solution of this application embodiment is: to acquire multi-source perception information and operating status information of vehicle sensors; to determine the confidence information of the vehicle sensors in a sub-region based on the environmental status information and the operating status information; to determine the fusion weight of the vehicle sensors in the corresponding sub-region based on the environmental status information and the confidence information; to perform fusion processing on the multi-source perception information based on the fusion weight to obtain regional fusion perception features; and to perform autonomous driving control based on the regional fusion perception features.

[0026] This application provides a solution that collects multi-source perception information and operational status information from vehicle sensors, calculates the confidence level of each sensor in different regions based on the external environment status, dynamically generates differentiated fusion weights based on the confidence levels, and obtains regional fusion perception features through weighted fusion. Autonomous driving control is then executed based on these regional fusion perception features. This solution can match driving environments under different conditions, adapt to the real-time working status of different vehicle sensors, fully leverage the complementary advantages of different types of vehicle sensors, improve the accuracy of perception and recognition during autonomous driving control, and ensure that the vehicle's environmental perception results closely match real driving scenarios.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or autonomous driving control device capable of performing the above functions. The following description uses an autonomous driving control device as an example to illustrate this embodiment and the subsequent embodiments.

[0028] Based on this, embodiments of this application provide an autonomous driving control method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the autonomous driving control method of this application.

[0029] In this embodiment, the autonomous driving control method includes steps S10 to S50: Step S10: Acquire multi-source sensing information and operating status information from the vehicle-mounted sensors; It should be noted that the vehicle-mounted sensors are data acquisition and data transmission devices deployed on the vehicle, and may include perception sensors and communication sensors. Perception sensors include LiDAR, millimeter-wave radar, perception cameras, surround-view cameras, etc., while communication sensors can be Vehicle to Everything (V2X) communication sensors. The vehicle may also be equipped with auxiliary systems such as a combined inertial navigation system, an onboard unit (OBU), and a control and scheduling unit (CSU) for vehicle positioning and information exchange with the dispatching platform. The vehicle in this application can be an unmanned vehicle, specifically an unmanned truck; this embodiment does not impose specific limitations on this.

[0030] In addition, multi-source sensing information consists of heterogeneous environmental raw data collected by different types of vehicle-mounted sensors, including 3D point clouds, image semantics, ranging, and communication interaction data. Operational status information reflects the real-time operating conditions of the vehicle-mounted sensors, including signal-to-noise ratio, fault indicators, and noise levels, which can indicate the current stability of the sensor's perception.

[0031] It should be understood that during vehicle operation, multi-source perception information collected by all on-board sensors is continuously acquired, along with the corresponding operating status information of each sensor.

[0032] Step S20: Determine the confidence information of the vehicle-mounted sensor in the sub-region based on the environmental status information and the operating status information; It should be noted that environmental state information is a set of parameters characterizing the vehicle's external driving conditions, including external environmental variables such as illumination, rainfall, and object reflectivity, which can change the perception accuracy of various sensors.

[0033] Furthermore, a sub-region is the smallest independent spatial unit obtained by dividing the vehicle's surrounding perception space. The distribution of obstacles and the degree of environmental interference to the sensors differ in different sub-regions, and differentiated environmental sensitivities are matched to different sub-regions. The confidence information includes evaluation scores of the perception reliability of multiple onboard sensors in the corresponding sub-regions. The evaluation scores are limited to a fixed range, with higher scores indicating higher accuracy in obstacle recognition in that area.

[0034] It should be understood that by reading the real-time collected environmental status information and the stored vehicle sensor operating status information, and combining the environmental sensitivity characteristics corresponding to each sub-region, the confidence information of each vehicle sensor in each sub-region is quantitatively calculated.

[0035] In one feasible implementation, step S20 may include steps S21 to S25: Step S21: Obtain sub-region characteristic information; It should be noted that the sub-region characteristic information is standardized data that characterizes the inherent sensing attributes of the corresponding sub-region. It is generated based on data collected from the vehicle's historical driving. Sub-regions with different orientations and distances from the vehicle will be matched with different values ​​of sub-region characteristic information, which can characterize the sensitivity of different sub-regions to external environmental factors such as light and rainfall. Each divided sub-region stores a set of corresponding sub-region characteristic information, which does not dynamically change with real-time weather or sensor hardware operating conditions.

[0036] It should be understood that the perception space surrounding the vehicle can be divided into several sub-regions according to a preset partitioning rule. Pre-trained environmental sensitivity mask data is read, and the characteristic information of each sub-region is extracted and cached in the intelligent driving computing platform. This platform, installed inside the vehicle, is used for real-time perception of environmental information, path planning, and vehicle control. Specifically, the preset partitioning rule can be to use a bird's-eye view rasterization method to uniformly divide the space around the vehicle into rectangular grid cells of fixed size, with each cell treated as an independent sub-region for confidence assessment and weight allocation.

[0037] Step S22: Determine the environmental state influence vector based on the environmental state information; It should be noted that the environmental state information is a set of parameters describing the real-time driving conditions outside the vehicle, including external variables such as light intensity, rainfall, and road surface reflectivity that can change the sensor's perception accuracy.

[0038] In addition, the environmental state influence vector is the local influence vector of the environmental state in the corresponding sub-region, which can quantify the degree of interference of the external environment on the perception effect of a single sub-region.

[0039] It should be understood that by reading the environmental state information around the current vehicle operation, extracting the current environmental state vector based on the environmental state information, and multiplying it with the environmental sensitivity mask of the corresponding sub-region, the environmental state influence vector can be obtained.

[0040] For example, the formula for calculating the environmental state influence vector is as follows:

[0041] In the formula, Indicates the system's running time; This represents the corresponding sub-region number; For environmental factor functions, Subregion At any moment The vector of influence of the sensor's environmental state; express The time-state vector is used to describe external environmental variables such as illumination, rainfall, and reflectance. Subregion The environmental sensitivity mask, learned through training on historical data, is used to characterize the sensitivity coefficients of different regions to environmental changes.

[0042] Step S23: Determine sensor status evaluation information based on the operating status information; It should be noted that the sensor status assessment information is an assessment vector generated by mapping the operating status information to the weight matrix obtained through actual vehicle calibration. Operating status information with different fault levels and signal-to-noise ratio values ​​will generate sensor status assessment information with numerical differences, which can reflect the working reliability of a single sensor in the current period.

[0043] It should be understood that the system retrieves the operating status information corresponding to each vehicle sensor, calls the pre-calibrated weight matrix to complete the linear mapping operation, and outputs the sensor status evaluation information corresponding to the sensor.

[0044] Step S24: The environmental state influence vector, the sensor state evaluation information, and the sub-region characteristic information are concatenated to obtain a combined feature vector; It should be noted that vector concatenation is a data processing operation that sequentially merges multiple sets of feature vectors with different dimensions into a single long vector, which can simultaneously retain all feature data from the three dimensions of environment, sensor, and region. Furthermore, the combined feature vector is a unified feature vector that integrates environmental state, sensor operating conditions, and sub-region attributes.

[0045] It should be understood that the environmental state influence vector corresponding to the current sub-region, the sensor state evaluation information of the corresponding vehicle sensor, and the sub-region characteristic information are extracted. The vector concatenation operation is then performed to integrate the three types of data into a combined feature vector. A set of combined feature vectors corresponds to the input features of a single vehicle sensor in a single sub-region.

[0046] Step S25: Determine the confidence information of the vehicle sensor in the sub-region based on the combined feature vector and preset weight parameters.

[0047] It should be noted that the preset weight parameters are fixed vector parameters that are learned and adjusted in advance through real vehicle calibration and real vehicle road test data. The preset weight parameters are used to measure the influence of environmental dimension, sensor dimension, and sub-region attribute dimension features on perception credibility respectively.

[0048] It should be understood that by weighting the combined feature vector with preset weight parameters and superimposing preset bias terms, and by constraining the calculation results to a fixed standardized numerical range through mapping processing, the confidence score of the corresponding vehicle sensor in the current sub-region can be obtained. The confidence information is obtained by organizing the confidence scores of all vehicle sensors on the vehicle.

[0049] For example, the formula for calculating confidence level is as follows:

[0050] In the formula, Indicates the sensor number; Indicates sensor In sub-region ,time The confidence score, with a range of values. It is used to quantify the reliability of sensors in perceiving environmental targets in a local area; This refers to the Sigmoid function, used to map the model's computational results to... Confidence interval; superscript Indicates vector transpose; The weight vector, which is learned and adjusted in real time through real vehicle calibration, is used to characterize the weights of different environmental factors on the confidence level. This represents the vector concatenation operator; The output of the environmental factor function represents the sub-region. At any moment The influence vector of the sensor's environmental state; The output of the sensor state function represents... For sensors The operational status assessment results, i.e., sensor status assessment information; Indicates the region The inherent characteristic parameters of the region are used to characterize the attributes of the local region itself, that is, the characteristic information of the sub-region; This represents the bias term used in confidence score calculations to adjust the confidence score baseline.

[0051] The calculation formula for the sensor operating status evaluation result is as follows:

[0052] In the formula, Represents the sensor state function. Indicates at time sensor The operational status assessment results; The sensor state weight matrix is ​​obtained from the actual vehicle sensor calibration and is used to weight and calculate various sensor operating indicators. This represents the sensor's own operating parameters at time t, including signal-to-noise ratio, fault indicators, noise level, etc., and is used to characterize the sensor's real-time working status.

[0053] In this embodiment, by extracting the environmental state influence vector and sensor state assessment information of sub-regions step by step, and then splicing them to generate a unified combined feature vector to calculate confidence information, it is possible to achieve refined perception reliability quantification by region, providing a reliable data foundation for subsequent dynamic allocation of fusion weights.

[0054] Step S30: Determine the fusion weight of the vehicle-mounted sensor in the corresponding sub-region based on the environmental state information and the confidence information; It should be noted that the fusion weight is the percentage of information contribution allocated to a single vehicle sensor in the corresponding sub-region, and the weight value will be dynamically adjusted according to the driving scenario.

[0055] It should be understood that by extracting the confidence information of each vehicle sensor in each sub-region, and combining the real-time adjusted temperature parameters and fusion strategy selection factors, the fusion weight of each vehicle sensor in the corresponding sub-region is calculated in a normalized manner, thereby achieving differentiated allocation of weights in different spatial regions.

[0056] In one feasible implementation, step S30 may include steps S31 to S33: Step S31: Determine the current environmental state vector based on the environmental state information; It should be noted that the current environmental state vector is a unified vector extracted based on real-time collected environmental state information. Each dimension corresponds to a quantitative value of a type of external environmental condition. The number and order of the vector dimensions are determined in advance through actual vehicle calibration and can include dimensions such as illumination, rainfall, and reflectance coefficient.

[0057] It should be understood that by reading real-time collected environmental status information, numerical standardization of environmental parameters with different dimensions such as light intensity, rainfall, and road surface object reflectance coefficient is performed, and various quantified environmental parameters are integrated into a vector of a unified dimension according to a preset dimensional arrangement order to generate the current environmental status vector, so as to uniformly quantify the strength of environmental interference.

[0058] Step S32: Determine the fusion strategy selection factor based on the confidence information and the current environment state vector; It should be noted that the fusion strategy selection factor is the identifier of the corresponding fusion weight allocation strategy. Different fusion strategy selection factors are adapted to different scenarios. The fusion weight allocation strategies include fusion weight allocation strategies for high-confidence stable environments, low-confidence extreme environments, and intermediate transitional environments.

[0059] It should be understood that the confidence information obtains the confidence scores of all sub-regions, calculates the mean of the maximum confidence scores of sensors in all sub-regions to obtain the global average confidence, compares the global average confidence with the preset confidence threshold, and compares the magnitude of the current environmental state vector with the preset environmental state threshold, and matches the corresponding strategy parameters according to the preset judgment rules.

[0060] Specifically, the preset confidence thresholds include a first confidence threshold and a second confidence threshold. The first confidence threshold is greater than the second confidence threshold. The first confidence threshold is a preset numerical standard used to distinguish the overall high confidence level of the sensor, and the second confidence threshold is a preset numerical standard used to distinguish the overall low confidence level of the sensor. The preset environmental state threshold is a preset numerical standard used to distinguish between normal environments and extremely harsh environments.

[0061] In specific implementation, when the global average confidence level is greater than or equal to the first confidence threshold and the magnitude of the current environment state vector is less than or equal to the preset environment state threshold, the fusion strategy selection factor is determined as the first factor. The first factor is used to match the fusion weight allocation strategy for high-confidence stable environments. When the global average confidence level is greater than or equal to the second confidence threshold, or the magnitude of the current environment state vector is greater than the preset environment state threshold, the fusion strategy selection factor is determined as the second factor. The second factor is used to match the fusion weight allocation strategy for intermediate transition environments. In other cases, the fusion strategy selection factor is determined as the third factor. The third factor is used to match the fusion weight allocation strategy for low-confidence extreme environments.

[0062] For example, the formula for calculating the global average confidence score is as follows:

[0063] In the formula, express Time-based global average confidence level; This represents the total number of sub-regions perceived by the vehicle. A set of regions representing all subregions; This indicates the maximum value of the confidence scores for all sensors. Indicates sensor In sub-region ,time The confidence score.

[0064] For example, the formula for calculating the fusion strategy selection factor is as follows:

[0065] In the formula, express Moment-based fusion strategy selection factors; express Time-based global average confidence level; This represents the high confidence threshold, also known as the first confidence threshold, which is a pre-set critical parameter used to classify highly reliable and stable scenarios. This represents the low confidence threshold, also known as the second confidence threshold, which is a pre-set critical parameter used to classify severe and extreme scenarios. express The environmental state vector of the vehicle at any given moment; This represents the vector norm operation, used to calculate the magnitude of the environment state vector; This represents the preset environmental state threshold, which is a pre-set critical parameter used to determine extreme environments; This represents the standard fusion strategy parameters, i.e., the first factor. The representative standard weighted average corresponds to the fusion weight allocation strategy under a high-confidence stable environment; This represents the robust fusion strategy parameter, i.e., the second factor. This represents the choice with the lowest confidence level, corresponding to the fusion weight allocation strategy under extreme low-confidence conditions; This represents the adaptive strategy parameter, also known as the third factor, which corresponds to the fusion weight allocation strategy that dynamically adjusts according to the environment under intermediate operating conditions.

[0066] Step S33: Determine the fusion weight of the vehicle sensor in the corresponding sub-region based on the confidence information and the fusion strategy selection factor.

[0067] It should be noted that the preset temperature parameter is an adjustment parameter obtained in advance through calibration and adjustment in a real vehicle operation scenario. It is used to control the sensitivity of the numerical distribution between the fusion weights and can participate in the weight calculation in real time. The higher the temperature parameter value, the more significant the impact of the difference in confidence level on the weight allocation; the lower the temperature parameter value, the more the weight allocation tends to be averaged.

[0068] It should be understood that the confidence information of the vehicle sensors in the current sub-region is retrieved, and the normalization operation is performed by combining the fusion strategy selection factor and the preset temperature parameter to output the fusion weights of each vehicle sensor.

[0069] For example, the formula for calculating the fusion weights of vehicle-mounted sensors is as follows:

[0070] In the formula, This indicates the sequence number of the onboard sensor being summed and iterated. Indicates the total number of onboard sensors; Indicates sensor In the region ,time The fusion weights are used to allocate the proportion of data from each sensor when fusing multi-source sensing data; This represents the temperature parameter, which is calibrated and adjusted in real time according to the actual vehicle operation scenario to control the distribution sensitivity of the fusion weights; express Moment-based fusion strategy selection factors; Indicates sensor In the region ,time Confidence score; Indicates sensor In the region ,time The confidence score.

[0071] In this embodiment, environmental state assessment information is obtained by first quantifying the interference level of the vehicle's external environment, and then the confidence information is combined with the selection factor of the fusion strategy adapted to the current scenario. The fusion weight of each sensor sub-region is dynamically calculated based on the two types of information, which allows sensors with higher perception reliability to obtain a higher information contribution ratio, realizes the on-demand activation of redundant perception units, adapts to high interference long-tail scenarios such as rain, fog and strong light, and provides an accurate weight allocation basis for subsequent multi-source perception information fusion processing.

[0072] Step S40: Perform fusion processing on the multi-source sensing information according to the fusion weight to obtain regional fusion sensing features; It should be noted that the fusion processing is a calculation process that integrates multi-source perception information from all vehicle sensors by dynamically allocating fusion weights for each sub-region.

[0073] In addition, the dynamic coupling term is a compensation parameter that is automatically activated in low-confidence scenarios. It relies on adaptive gain to enhance the contribution of redundant sensor units and make up for the perception and recognition deviation caused by sudden environmental changes.

[0074] It should be understood that by calling the fusion weights of each vehicle sensor in the current sub-region, weighted calculations are performed on all multi-source perception information in the region, and dynamic coupling terms are superimposed to complete perception deviation compensation, thereby obtaining the regional fusion perception features of the corresponding sub-region.

[0075] In one feasible implementation, step S40 may include steps S41 to S43: Step S41: Perform weighted fusion of the multi-source sensing information according to the fusion weight to obtain weighted fusion features; It should be noted that the weighted fusion feature is the basic environmental feature obtained after weighting the data based on the fusion weights. The fusion weights and multimodal perception information corresponding to each vehicle-mounted multimodal sensor in the current sub-region are obtained. The multimodal perception information is multiplied by the matched fusion weights to obtain the multiplication result. Then, the multiplication results of all sensors in the same sub-region are accumulated to obtain the weighted fusion feature.

[0076] Step S42: Determine the dynamic coupling term based on the confidence information, the fusion weight, and the multi-source sensing information; It should be noted that the dynamic coupling term is the sensing compensation data enabled in scenarios of sudden environmental changes, which can compensate for the perception feature deviation caused by the failure of a single sensor or environmental interference.

[0077] It should be understood that the global average confidence level can be calculated based on the confidence level information, and then mapped to the corresponding adaptive gain coefficient through a mapping function. By iterating through all fusion weights and multi-source sensing information in the current sub-region, the maximum value of the single-sensor weighted output is calculated. The difference between the maximum value and the weighted fusion feature is then obtained, and this difference is multiplied by the adaptive gain to obtain the dynamic coupling term.

[0078] For example, the formula for calculating the dynamic coupling term is as follows:

[0079] In the formula, Represents the total number of vehicle-mounted sensors; Subregion ,time Dynamic coupling terms; Indicates the adaptive gain coefficient; Indicates sensor In sub-region ,time The fusion weights; Indicates sensor In sub-region ,time The raw sensory data, i.e., multi-source sensory information; This indicates the weighted fusion feature.

[0080] The mapping formula for the adaptive gain coefficient is as follows:

[0081] In the formula, This represents the adaptive gain mapping function, which is used to map the global average confidence score to the corresponding adaptive gain coefficient. Indicates the adaptive gain coefficient; This represents the global average confidence level.

[0082] Step S43: Determine the region fusion sensing features based on the weighted fusion features and the dynamic coupling term.

[0083] It should be understood that the dynamic coupling term is superimposed on the weighted fusion feature to obtain the regional fusion perception feature.

[0084] For example, the formula for expressing the region fusion sensing features is as follows:

[0085] In the formula, Represents the total number of vehicle-mounted sensors; Indicates in Sub-regions obtained by time-mapping Regional fusion perception features; Indicates in Time sub-region The weighted fusion characteristics; Indicates in Time sub-region Dynamic coupling terms.

[0086] In this embodiment, the multimodal sensing information is first weighted by dynamically allocated fusion weights to obtain weighted fusion features. Then, the dynamic coupling term is solved by combining confidence information, weights, and sensing data. Finally, the two types of data are fused to generate the final fused sensing features. The compensation logic can be automatically activated under low-confidence extreme conditions such as sudden changes in illumination and rain / fog obstruction, actively amplifying the sensing contribution of redundant sensors.

[0087] Step S50: Perform autonomous driving control based on the region fusion perception features.

[0088] It should be understood that key environmental parameters such as obstacle distance, lane position, and surrounding vehicle speed are extracted from the fusion perception features of each group of regions. Combined with the vehicle's current real-time speed and vehicle positioning, driving action commands adapted to the current driving scenario are generated. The generated action commands are then sent to the intelligent driving controller so that the intelligent driving controller can perform steering, braking, acceleration, or deceleration operations according to the received commands, thereby performing autonomous driving control.

[0089] This embodiment provides an autonomous driving control method. By collecting multi-source perception information and operating status information from vehicle sensors, and combining the external environment status to calculate the confidence level of each sensor in different regions, a differentiated fusion weight is dynamically generated based on the confidence level. The weighted fusion is then used to obtain regional fusion perception features. Autonomous driving control is executed based on these regional fusion perception features. This method can match driving environments under different conditions, adapt to the real-time working status of different vehicle sensors, fully leverage the complementary advantages of different types of vehicle sensors, improve the accuracy of perception and recognition during autonomous driving control, and make the vehicle's environmental perception results closely match real driving scenarios.

[0090] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 This may include steps A10 to A50: Step A10: Extract a global feature library based on the multi-source sensing information; It should be noted that the global feature library is a standardized feature storage set that integrates heterogeneous features extracted from all vehicle sensors. It includes 3D point cloud features, image semantic features, ranging features, communication interaction features, etc., output by different sensors in sub-regions.

[0091] It should be understood that, based on the extraction of initial features from multi-source sensing information, the heterogeneous features initially extracted by all sensors are obtained, forming a global feature library.

[0092] Step A20: Perform conflict detection on the global feature library to obtain conflict detection results; It should be noted that conflict detection is a data verification process that compares multiple sets of heterogeneous features of similar environmental targets within a global feature library to determine whether multi-sensor perception data contradict each other over a large scale. Verification is performed sequentially according to pre-defined spatial sub-regions. This process can include two types of data anomalies: one is local feature deviation caused by short-term noise from a single sensor; the other is anomalies where feature values ​​deviate significantly across multiple sub-regions and multiple heterogeneous sensors simultaneously. Conflict detection is periodically performed before each round of multi-source perception information processing to identify data anomalies that could affect the reliability of the overall vehicle environmental perception results.

[0093] In addition, the conflict detection results are the labeling data after determining the consistency of heterogeneous sensing data in the global feature library, including two categories: labels that do not have conflicts and labels that have systematic conflicts.

[0094] It should be understood that, based on the full-domain feature library, feature consistency comparison is performed sequentially according to the division order of spatial sub-regions. Within a single sub-region, the heterogeneous feature values ​​output by different vehicle sensors for the same environmental targets such as obstacles, lane markings, and traffic signs are compared to determine whether multiple sets of feature data exceed the preset reasonable error range. After completing the feature verification work of all sub-regions within the vehicle's perception coverage area, the corresponding conflict detection results are generated based on the coverage range of abnormal data in the full domain.

[0095] Step A30: When the conflict detection result indicates the existence of a systematic conflict, determine the target sensor subset based on the global sensor set; It should be noted that the full-domain sensor set includes parameter data corresponding to all on-board sensors in the vehicle, including the installation location, data output type, hardware acquisition range, and other offline calibration parameters of sensors such as LiDAR, millimeter-wave radar, perception cameras, surround view cameras, and V2X communication sensors deployed throughout the vehicle.

[0096] In addition, systemic conflict is an abnormal operating condition in which the perception data output by multiple types of vehicle-mounted sensors for the same environmental target in the entire domain show large-scale contradictions. This condition will affect the accuracy of the overall environmental judgment.

[0097] In addition, the target sensor subset is the minimum combination of sensors required to complete a single security task. When splitting the target sensor subset, the constraint of minimizing the number of hardware is followed, so as to reduce the computing power consumption generated by the independent operation of a single set of sensor subsets. Each set of target sensor subsets contains only the minimum number of heterogeneous sensor hardware necessary to complete a single security task, and the sensor hardware combinations selected between different target sensor subsets do not overlap.

[0098] It should be understood that if the conflict detection result determines that there is no conflict in the global feature library, then global fusion is performed based on the regional fusion perception features, and autonomous driving control is executed on the global fusion perception features. If the conflict detection result determines that there is a systematic conflict in the global feature library, then all on-board sensors included in the global sensor set are traversed, and multiple independent target sensor subsets are obtained by following the constraint of minimizing the number of subset devices.

[0099] Step A40: Determine the task allocation information corresponding to the target sensor subset based on the global feature library and security task information; It should be noted that safety task information refers to the task identifiers and supporting perception requirements data corresponding to various core functions that ensure the overall driving safety of the unmanned truck. This includes target recognition standards and perception data input constraints corresponding to different safety functions such as emergency braking, lane keeping, obstacle avoidance and overtaking. All safety tasks need to be fully covered after splitting the sensor subsets. The risk priority of each safety task is differentiated under different driving conditions and traffic density conditions.

[0100] In addition, task allocation information is standardized association data that establishes a one-to-one correspondence between a single target sensor subset and a single security task. It records the specific types of security tasks that each target sensor subset needs to independently perform, limiting each target sensor subset to perform only one security task and preventing the spread of perception conflicts between different security functions.

[0101] It should be understood that by extracting all heterogeneous sensing features from the global feature library and performing feature cross-referencing, and then combining this with the functional execution requirements of each security task, a single security task is matched to a set of independent target sensor subsets, generating task allocation information corresponding to each subset. Feature cross-referencing is a data processing method that extracts heterogeneous features from the global feature library to supplement the feature matrix of the target sensor subset, which can enhance the complementarity of data from different sensors within the subset.

[0102] In one feasible implementation, step A40 may include steps A41 to A44: Step A41: Obtain the feature matrix and feature complementarity score corresponding to the target sensor subset; It should be noted that the feature matrix is ​​a standardized set of features generated by integrating the raw sensing data collected by a subset of target sensors through a feature deep cross-referencing mechanism. The feature matrix contains heterogeneous features such as point clouds, images, and radar ranging extracted by all sensing hardware in the subset. Feature deep cross-referencing can strengthen the correlation between different sensing data, making the feature matrix carry richer environmental information. Each subset of target sensors will generate a corresponding feature matrix.

[0103] In addition, the feature complementarity score is a quantitative indicator that quantifies the degree of complementarity of heterogeneous sensing data within a subset of target sensors. When the feature complementarity score is high enough, it means that the various sensors in the subset can cancel each other out the acquisition noise and perception bias generated by a single sensor, and that the other hardware data can make up for the information when a sensor is disturbed.

[0104] It should be understood that by traversing all the target sensor subsets that have been split, extracting the feature matrix generated by each subset, and then calculating the feature complementarity score of the corresponding subset based on the diversity of heterogeneous features of the subsets.

[0105] Step A42: Obtain the task importance weight corresponding to the security task based on the security task information; It should be noted that the task importance weight is a value obtained based on the safety risk assessment of the current driving scenario, used to characterize the proportion of the contribution of the corresponding safety task to the overall vehicle driving safety.

[0106] It should be understood that the process involves reading safety task information, extracting all safety tasks to be executed, and combining the surrounding traffic flow, obstacle distribution, and vehicle speed perceived by the vehicle to determine the risk level of the scenario. Then, according to the preset weight matching rules, each safety task is assigned a task importance weight that is appropriate for the current working condition.

[0107] Step A43: Determine the task performance score corresponding to the target sensor subset based on the feature matrix and the global feature library; It should be noted that the task performance score is a quantitative score that measures the effectiveness of the target sensor subset in completing the specified safety task. The higher the score, the more suitable the subset's perception output is for the judgment requirements of the corresponding task.

[0108] It should be understood that the feature matrix and the global feature library are subjected to feature cross-reference operation to fuse global heterogeneous features, which are then substituted into the preset task evaluation logic to calculate the comprehensive ability of the target sensor subset to perform safety tasks, and output the task performance score corresponding to the subset.

[0109] Step A44: Determine the task allocation information corresponding to the target sensor subset based on the feature complementarity score, the task importance weight, and the task performance score.

[0110] It should be understood that when a systemic conflict occurs in the global feature library, the feature complementarity score of a single set of target sensor subsets, the task importance weight of the corresponding security task, and the performance score of the subset task are substituted into the robustness optimization objective formula. By optimizing the combination of sensor subsets and security tasks, the maximum expected robustness score is solved, the optimal pairing relationship is selected, and the task allocation information corresponding to each set of target sensor subsets is generated.

[0111] For example, the robustness optimization objective formula is as follows:

[0112]

[0113]

[0114] In the formula, This represents the maximum expected robustness score that the vehicle perception system can achieve. This represents the total number of sub-guarantee units after decoupling; Representing the The smallest set of sensors for each sub-protection unit; Representing the Each sub-protection unit is bound to a single security task; The subset of sensors selected by the sub-protection unit; Each represents a single security task handled by a sub-protection unit. This corresponds to a key safety function, including emergency braking, lane keeping assist, obstacle avoidance and overtaking; This represents solving for the optimal configuration within the sensor subset and security task combination space; The mathematical expectation operator represents the expression in a real dataset. The expected average is used to ensure the optimization objective's ability to generalize to the real environment; Representing the The importance weight of each safety task is used to characterize the proportion of the corresponding task's contribution to the overall vehicle safety; subscript Represents complementary attribute identifiers; Represents a subset of sensors Extracted heterogeneous feature matrix; Represents a subset of sensors The feature complementarity score is used to measure the feature complementarity capability of multiple sensors, where , Indicates feature diversity; The feature cross-reference operator is used to fuse sub-unit features with information from the global feature library. Represents the full-domain feature library, used to store heterogeneous features extracted from all sensors; The representative task performance function is used to evaluate the completion of a security task under the current combination of features. The actual effect.

[0115] In the constraints This represents the conflict detection function. It is assigned a value of 1 when a systematic conflict occurs in the global feature library, which is used to trigger automatic decoupling optimization of the system. A global sensor set representing all onboard sensors; Represents a set of constraints, indicating A comprehensive sensor suite, including LiDAR, cameras, millimeter-wave radar, and high-precision positioning. This means minimizing the number of sensors selected for each subunit; This is a collection of tasks that cover all critical security requirements. This represents the set of constraints, indicating that all critical tasks are fully covered after decoupling. Additionally, Use global integration at times.

[0116] In this embodiment, by first collecting the feature matrices and feature complementarity scores of all target sensor subsets, then matching the task importance weights corresponding to each security task, and calculating the subset task performance score based on the feature matrices and the global feature library, the pairing of subsets and security tasks is completed to generate task allocation information. This can maximize the feature complementarity effect of heterogeneous sensors within each sub-protection unit, allowing high-security-priority tasks to be matched with the sensor subsets with the best perception performance. This effectively reduces the chain perception failure problem caused by global systemic conflicts and improves the stability and anti-interference performance of each security task under conflict conditions.

[0117] Step A50: Perform autonomous driving control based on the target sensor subset and the task allocation information.

[0118] It should be noted that the sub-support unit is the target sensor subset operation unit after being bound with the corresponding task allocation information. Each sub-support unit independently completes the sensing and calculation operations, and there is no data coupling interference between the units.

[0119] It should be understood that each subset of target sensors is constructed as an independent sub-support unit based on the task allocation information. Each unit relies solely on the data collected by its own onboard sensors to complete the judgment of matching safety tasks and output the corresponding vehicle driving control commands to complete autonomous driving control.

[0120] In this embodiment, when a conflict occurs, the target sensor subset is split and bound to the corresponding safety task, and autonomous driving-related safety control is executed separately. This can split the highly coupled perception operation into independent operation units that do not interfere with each other, so as to avoid the chain failure problem caused by the perception conflict in the whole domain. At the same time, the robustness of single-task perception is improved by relying on the complementary heterogeneous features of the target sensor subset, so as to realize the vehicle performance safety degradation in the conflict scenario and improve the anti-interference ability and driving safety of the vehicle in the process of driving in complex factory areas.

[0121] In one feasible implementation, step S50 may include steps B10 to B40: Step B10: Obtain current vehicle operation scenario information, local baseline strategy, and cloud-based herd immunity network information; It should be noted that the current vehicle operating scenario information is a state vector representing the real-time driving condition of the vehicle, which includes real-time perceived environmental and vehicle driving parameters such as obstacle distance, vehicle speed, and weather conditions, and is used to identify the current extreme or normal driving scenario.

[0122] Additionally, the local baseline strategy is a pre-deployed native autonomous driving decision-making logic. It outputs the corresponding vehicle action vector by inputting the state vector of the current vehicle operating scenario. This action vector includes underlying control variables such as steering angle, acceleration, and braking force. The local baseline strategy covers basic driving control logic under normal road conditions, such as following other vehicles, lane keeping, and basic obstacle avoidance. However, in extreme long-tail interference scenarios such as heavy rain, dense fog, backlighting, strong sunlight, and highly reflective road surfaces, the decision results output by the local baseline strategy may have certain recognition biases, requiring the addition of scenario compensation increments for correction.

[0123] It should be understood that during vehicle operation, perception data is collected in real time to obtain environmental parameter data and driving parameter data. Based on the environmental parameter data and driving parameter data, the current vehicle operation scenario information is generated, and the local baseline strategy stored on the vehicle is read.

[0124] Step B20: Determine the scenario compensation increment based on the cloud-based herd immunity network and the current vehicle operation scenario information; It should be noted that the scene compensation increment is a driving action correction vector generated by the lightweight plugin, used to compensate for decision-making biases caused by the local baseline strategy under extreme common failure scenarios, and to optimize driving safety judgment results. The cloud-based herd immunity network is a shared knowledge base that is continuously and dynamically updated on the cloud server side. It stores the scene compensation plugins corresponding to different driving areas and the metadata of various common failure modes. It is continuously updated by the fleet operation group data, and the fleet contains multiple unmanned vehicles.

[0125] It should be understood that a communication link is established to retrieve the cloud-based herd immunity network information updated in real time on the cloud side. Based on the current vehicle operation scenario information, the common failure modes stored in the cloud-based herd immunity network are retrieved, the corresponding lightweight compensation plugin is retrieved to complete the lightweight calculation, and the scenario compensation increment adapted to the current driving conditions is output.

[0126] In one feasible implementation, step B20 may include steps B21 to B23: Step B21: Obtain common failure modes from the cloud-based herd immunity network to obtain a set of common failure modes; It should be noted that common failure modes are extreme conditions of sensor perception failure that occur repeatedly during the operation of multiple vehicles in the fleet. These include perception failure cases under various long-tail interference scenarios such as heavy rain, dense fog, backlight, strong light, and highly reflective road surfaces. Each type of common failure mode corresponds to a typical extreme driving scenario. Under the same type of failure mode, the vehicle's perception and decision-making will have similar deviation problems.

[0127] Additionally, the common failure mode set is a standardized collection of historical samples corresponding to all common failure modes. It contains multiple sets of one-to-one corresponding driving scenario state vectors and actual scenario results. The driving scenario state vector corresponds to the current vehicle operating scenario information, which may include driving parameters such as vehicle speed, obstacle distance, and road attributes, as well as environmental parameters such as road attributes and weather. The actual scenario result is a binary safety indicator, with a first identifier value (e.g., 0) representing driving safety and a second identifier value (e.g., 1) representing perception failure.

[0128] It should be understood that all common failure modes are extracted from the cloud-based herd immunity network, and the historical sample data corresponding to each type of failure mode are summarized and organized. Data format is standardized and invalid and abnormal samples are removed to ultimately form a collection of common failure modes.

[0129] Step B22: Determine the plugin complexity penalty term based on the preset regularization coefficient and plugin complexity regularization term; It should be noted that the preset regularization coefficient is an adjustment parameter determined in advance through real-vehicle calibration. It is used to balance the compensation and correction effect with the lightweight attributes of the plugin, and to adjust the weight of complexity constraints in the overall optimization goal. The larger the value of the preset regularization coefficient, the stronger the constraint on the complexity of the plugin during the optimization process, and the lower the computational load of the generated compensation plugin, while sacrificing the corresponding compensation and correction effect; the smaller the value, the more the optimization process focuses on improving the compensation effect, and the complexity of the generated plugin will increase accordingly.

[0130] In addition, the plugin complexity regularization term is a quantitative term used to measure the computational complexity of the compensation plugin. Its value is positively correlated with the number of parameters and computational steps of the compensation plugin. The higher the value, the higher the computational load of the plugin and the longer the loading and running time of the vehicle terminal.

[0131] In addition, the plugin complexity penalty term is a quantitative constraint term used to constrain the computational complexity of the compensation plugin. It is obtained by multiplying the preset regularization coefficient with the plugin complexity regularization term. It is used to apply lightweight constraints during the compensation incremental optimization process to ensure that the generated compensation plugin does not excessively occupy the vehicle's computing power resources. It supports the loading efficiency and real-time fusion computation of the plugin on the vehicle side, so as to simultaneously take into account the effect and running efficiency and adapt to the actual operating environment of the vehicle side.

[0132] It should be understood that the plugin complexity regularization term is determined based on the parameter size constraints and computational logic requirements of the compensation plugin, and the plugin complexity penalty term is obtained by multiplying the preset regularization coefficient with the plugin complexity regularization term.

[0133] Step B23: Determine the scenario compensation increment based on the current vehicle operation scenario information, the cloud-based herd immunity network, the common failure mode set, and the plugin complexity penalty term.

[0134] It should be noted that the scene compensation increment is the action correction vector output by the lightweight compensation plugin, which is used to correct the driving decision deviation caused by the local baseline strategy under the corresponding extreme failure scenario.

[0135] It should be understood that, based on the optimization logic built into the cloud-based herd immunity network, the common failure mode set is used as the training sample, and a plugin complexity penalty term is introduced as a constraint to perform parameter optimization calculations for the compensation plugin. The overall optimization objective is to minimize the sum of the expected average value of the loss function and the plugin complexity penalty term. The loss function is used to quantify the difference between the fused action and the ideal result. The input of the loss function includes two types of data: the fused action and the actual scene result corresponding to the scene. The fused action consists of the vehicle action vector output by the local baseline policy plus the action correction vector. The expected average value of the loss function on the common failure mode set is used to measure the overall correction effect of the compensation plugin. Through parameter optimization calculations, the optimal plugin parameters that minimize the overall optimization objective are found, generating a scene compensation increment adapted to the current vehicle operating scenario.

[0136] For example, the formula for solving the scene compensation increment is as follows:

[0137] In the formula, This represents the optimal scenario compensation increment; This represents finding the plugin parameters that minimize the objective function within the plugin parameter space. ; n represents the scene number; Representing the A state vector for each driving scenario; It represents the actual outcome of a scenario and is a binary safety indicator. Represents safety, The representation is invalid; Representative common failure modes; This indicates the common failure mode dataset. The mathematical expectation is calculated, and the calculation process will traverse all scenario samples and corresponding real labeled data in the common failure mode set, and statistically analyze the average deviation between the compensation decision result and the real result. This indicates a plugin parameter optimization operation. The goal of the operation is to find a set of optimal compensation plugin parameters P that minimizes the overall optimization objective. Represents the state vector of the corresponding scene The vehicle motion vector can be steering angle, acceleration, braking force, etc. Represents the state vector of the corresponding scene Action correction vector, This represents the original action being superimposed with scene-specific modifications to obtain the final merged output; This represents the loss function, used to quantify the compensation effect of the plugin; This represents the preset regularization coefficient, which can be preset to a value of 0.01; This represents the plugin complexity regular expression.

[0138] In this embodiment, common failure modes are extracted from the cloud-based herd immunity network and organized into a common failure mode set. A plugin complexity penalty term is generated by combining a preset regularization coefficient and a plugin complexity regularization term. By integrating multiple types of input data, a scenario compensation increment adapted to the current operating conditions is obtained. Based on the accumulated extreme failure data, targeted decision correction quantities can be generated to ensure the matching degree and correction effect between the compensation increment and the current driving conditions, thereby improving the driving safety level and operating efficiency of autonomous vehicles in complex scenarios.

[0139] Step B30: Determine the target action vector based on the local baseline strategy and the scene compensation increment; It should be noted that the target motion vector is the final decision result of dynamically integrating local policies and cloud plugins. The performance indicators include safety risk reduction and efficiency improvement, which are obtained by superimposing the vehicle motion vector output by the local baseline policy and the scene compensation increment dimension by dimension.

[0140] It should be understood that the state vector of the current vehicle operation scenario is input into the local baseline strategy to calculate the initial vehicle action vector, obtain the scene compensation increment adapted to the current scenario, and then perform bit-by-bit vector superposition of the vehicle action vector and the scene compensation increment according to the corresponding control dimensions such as steering, acceleration, and braking to obtain the target action vector.

[0141] For example, the formula for synthesizing enhanced security policies is as follows:

[0142] In the formula, This represents the output action vector of the enhanced safety strategy, i.e., the target action vector, which is used to output the final driving control decision of the autonomous vehicle. Represents the current operating state vector of the vehicle; Represents a cloud-based herd immunity network; This indicates that the local baseline policy is based on the scene state vector. The output is the initial vehicle motion vector. This represents the local baseline autonomous driving policy function, used to output the vehicle's native basic control actions; The scene compensation increment function is based on the scene state vector. Cloud-based herd immunity network Common failure modes Calculate the generated scene compensation increment.

[0143] Step B40: Perform automatic driving control based on the target motion vector.

[0144] It should be understood that parsing the output action vector involves protocol conversion of the action quantities contained within the vector, such as steering angle, acceleration, and braking force. This is broken down into steering angle control commands for the steering actuator, torque and speed control commands for the power actuator, and pressure and deceleration control commands for the braking actuator. The converted control commands are then synchronously sent to the corresponding underlying actuators via the vehicle bus. Each hardware unit performs the corresponding acceleration, deceleration, steering, and braking operations according to the received command parameters, driving the vehicle to travel according to the decision target of the output action vector. During vehicle operation, onboard sensors continuously collect data on the surrounding environment and vehicle status, updating vehicle operating scenario information in real time. This process verifies the matching degree between the actual vehicle driving state and the output action vector, continuously and dynamically adjusting execution parameters to achieve continuous and stable autonomous driving control. After executing the autonomous driving control output vehicle driving commands, new extreme failure condition data encountered during the current driving process are simultaneously collected and transmitted back to the cloud via the communication link, updating the common failure mode set stored in the cloud.

[0145] This embodiment provides an autonomous driving control method that simultaneously acquires real-time local basic strategies and cloud-shared knowledge base information, generates lightweight scenario compensation increments based on common failure conditions in the cloud, and superimposes and corrects these increments to obtain enhanced safety strategies before executing autonomous driving control. This method retains the stable control capability of the local baseline strategy in normal scenarios while supplementing the cloud plugin with specialized optimizations for extreme long-tail scenarios, thus simultaneously taking into account both operational efficiency in normal conditions and driving safety in extreme conditions.

[0146] For example, to help understand the implementation flow of the autonomous driving control method obtained by combining this embodiment with the above embodiment one, please refer to... Figure 3 , Figure 3 A simplified flowchart of an autonomous driving control method is provided, specifically: Multiple autonomous vehicles are equipped with onboard sensors. These sensors collect environmental and operational data and output sensor status. This sensor status is then input into a multimodal perception fusion framework to complete basic data fusion. The framework outputs fused perception features based on the operating environment and passes these features to a heterogeneous sensor fusion strategy for conflict verification. The heterogeneous sensor fusion strategy has two processing branches: conflict-free and conflict-prone. In conflict-free scenarios, a full-domain fusion operation mode is used to obtain perception features. In conflict-prone scenarios, the system is split into independent sub-guarantee units to compute perception features separately. The autonomous driving module executes vehicle control based on these perception features. Perception and operational data generated during a single vehicle's operation are transmitted to the cloud via a data upload channel. Operational data from multiple autonomous vehicles converge into a group scenario immunity network. This network summarizes common failure scenarios and optimization strategies for the fleet and dynamically injects reinforcement strategies via Over-The-Air (OTA) technology, sending lightweight compensation logic to the vehicle-side perception fusion model. The vehicle-side perception fusion model simultaneously connects to the perception fusion link on the individual vehicle side, forming an overall operational architecture of real-time vehicle operation and iterative optimization in the cloud.

[0147] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the autonomous driving control method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0148] This application also provides an automatic driving control device, please refer to... Figure 4 The automatic driving control device includes: Data acquisition module 10 is used to acquire multi-source sensing information and operating status information from vehicle-mounted sensors; Data processing module 20 is used to determine the confidence information of the vehicle-mounted sensor in the sub-region based on the environmental state information and the operating state information; The data processing module 20 is further configured to determine the fusion weight of the vehicle-mounted sensor in the corresponding sub-region based on the environmental state information and the confidence information; Data fusion module 30 is used to perform fusion processing on the multi-source sensing information according to the fusion weight to obtain regional fusion sensing features; The driving control module 40 is used to perform autonomous driving control based on the regional fusion perception features.

[0149] In one embodiment, the data processing module 20 is further configured to acquire sub-region characteristic information; Determine the environmental state influence vector based on environmental state information; Determine sensor status assessment information based on the aforementioned operating status information; The environmental state influence vector, the sensor state evaluation information, and the sub-region characteristic information are concatenated to obtain a combined feature vector. The confidence information of the vehicle sensor in the sub-region is determined based on the combined feature vector and the preset weight parameters.

[0150] In one embodiment, the data processing module 20 is further configured to determine the current environmental state vector based on the environmental state information; The fusion strategy selection factor is determined based on the confidence information and the current environment state vector; The fusion weights of the vehicle sensors in the corresponding sub-regions are determined based on the confidence information and the fusion strategy selection factor.

[0151] In one embodiment, the data fusion module 30 is further configured to perform weighted fusion of the multi-source sensing information according to the fusion weight to obtain weighted fusion features; The dynamic coupling term is determined based on the confidence information, the fusion weight, and the multi-source sensing information. The region fusion sensing features are determined based on the weighted fusion features and the dynamic coupling term.

[0152] In one embodiment, the driving control module 40 is further configured to extract a global feature library based on the multi-source perception information; Conflict detection is performed on the global feature library to obtain conflict detection results; When the conflict detection result indicates the existence of a systematic conflict, the target sensor subset is determined based on the global sensor set; Based on the global feature library and security task information, determine the task allocation information corresponding to the target sensor subset; Autonomous driving control is performed based on the target sensor subset and the task allocation information.

[0153] In one embodiment, the driving control module 40 is further configured to acquire the feature matrix and feature complementarity score corresponding to the target sensor subset; Based on the security task information, obtain the task importance weight corresponding to the security task; The task performance score corresponding to the target sensor subset is determined based on the feature matrix and the global feature library. The task allocation information corresponding to the target sensor subset is determined based on the feature complementarity score, the task importance weight, and the task performance score.

[0154] In one embodiment, the driving control module 40 is further configured to acquire current vehicle operating scenario information and local baseline strategy; The scenario compensation increment is determined based on the cloud-based herd immunity network and the current vehicle operation scenario information. The target action vector is determined based on the local baseline strategy and the scene compensation increment. Automated driving control is performed based on the target motion vector.

[0155] In one embodiment, the driving control module 40 is further configured to obtain common failure modes from the cloud-based herd immunity network to obtain a set of common failure modes; The plugin complexity penalty term is determined based on the preset regularization coefficient and the plugin complexity regularization term; The scenario compensation increment is determined based on the current vehicle operation scenario information, the cloud-based herd immunity network, the common failure mode set, and the plugin complexity penalty term.

[0156] The autonomous driving control device provided in this application, employing the autonomous driving control method in the above embodiments, can solve the technical problem of how to improve the accuracy of perception and recognition during autonomous driving control. Compared with the prior art, the beneficial effects of the autonomous driving control device provided in this application are the same as those of the autonomous driving control method provided in the above embodiments, and other technical features in the autonomous driving control device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0157] This application provides an autonomous driving control device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the autonomous driving control method in Embodiment 1 above.

[0158] The following is for reference. Figure 5 The diagram illustrates a structural schematic suitable for implementing the autonomous driving control device of the embodiments of this application. The autonomous driving control device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The autonomous driving control device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0159] like Figure 5 As shown, the autonomous driving control device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the autonomous driving control device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touch screens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the autonomous driving control device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows an autonomous driving control device with various systems, it should be understood that it is not required to implement or possess all of the systems shown. More or fewer systems may be implemented alternatively.

[0160] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0161] The autonomous driving control device provided in this application, employing the autonomous driving control method in the above embodiments, can solve the technical problem of how to improve the accuracy of perception and recognition during autonomous driving control. Compared with the prior art, the beneficial effects of the autonomous driving control device provided in this application are the same as those of the autonomous driving control method provided in the above embodiments, and other technical features in this autonomous driving control device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0162] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0163] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0164] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the autonomous driving control method in the above embodiments.

[0165] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory), or flash memory, optical fiber, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0166] The aforementioned computer-readable storage medium may be included in the autonomous driving control device; or it may exist independently and not be assembled into the autonomous driving control device.

[0167] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by an autonomous driving control device, the autonomous driving control device causes the following: it acquires multi-source perception information and operational status information from onboard sensors; determines the confidence level information of the onboard sensors in a sub-region based on the environmental status information and the operational status information; determines the fusion weight of the onboard sensors in the corresponding sub-region based on the environmental status information and the confidence information; performs fusion processing on the multi-source perception information based on the fusion weight to obtain regional fusion perception features; and performs autonomous driving control based on the regional fusion perception features.

[0168] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0169] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that individual blocks in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0170] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0171] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described autonomous driving control method, and can solve the technical problem of how to improve the accuracy of perception and recognition during autonomous driving control. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the autonomous driving control method provided in the above embodiments, and will not be repeated here.

[0172] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described automatic driving control method.

[0173] The computer program product provided in this application can solve the technical problem of how to improve the accuracy of perception and recognition during autonomous driving control. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the autonomous driving control method provided in the above embodiments, and will not be repeated here.

[0174] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. An automatic driving control method, characterized in that, The autonomous driving control method includes: Acquire multi-source sensing information and operational status information from vehicle-mounted sensors; The confidence information of the vehicle-mounted sensor in the sub-region is determined based on the environmental status information and the operational status information. The fusion weights of the vehicle-mounted sensors in the corresponding sub-regions are determined based on the environmental state information and the confidence information. The multi-source sensing information is fused according to the fusion weights to obtain regional fused sensing features; Autonomous driving control is performed based on the region fusion perception features.

2. The method as described in claim 1, characterized in that, The step of determining the confidence information of the vehicle-mounted sensor in the sub-region based on the environmental state information and the operating state information includes: Obtain sub-region characteristic information; Determine the environmental state influence vector based on environmental state information; Determine sensor status assessment information based on the aforementioned operating status information; The environmental state influence vector, the sensor state evaluation information, and the sub-region characteristic information are concatenated to obtain a combined feature vector. The confidence information of the vehicle sensor in the sub-region is determined based on the combined feature vector and the preset weight parameters.

3. The method as described in claim 1, characterized in that, The step of determining the fusion weight of the vehicle-mounted sensor in the corresponding sub-region based on the environmental state information and the confidence information includes: Determine the current environmental state vector based on the environmental state information; The fusion strategy selection factor is determined based on the confidence information and the current environment state vector; The fusion weights of the vehicle sensors in the corresponding sub-regions are determined based on the confidence information and the fusion strategy selection factor.

4. The method as described in claim 1, characterized in that, The step of fusing the multi-source sensing information according to the fusion weights to obtain regional fused sensing features includes: The multi-source sensing information is weighted and fused according to the fusion weights to obtain weighted fusion features; The dynamic coupling term is determined based on the confidence information, the fusion weight, and the multi-source sensing information. The region fusion sensing features are determined based on the weighted fusion features and the dynamic coupling term.

5. The method as described in claim 1, characterized in that, The method further includes: Extract a global feature library based on the multi-source sensing information; Conflict detection is performed on the global feature library to obtain conflict detection results; When the conflict detection result indicates the existence of a systematic conflict, the target sensor subset is determined based on the global sensor set; Based on the global feature library and security task information, determine the task allocation information corresponding to the target sensor subset; Autonomous driving control is performed based on the target sensor subset and the task allocation information.

6. The method as described in claim 5, characterized in that, The step of determining the task allocation information corresponding to the target sensor subset based on the global feature library and security task information includes: Obtain the feature matrix and feature complementarity scores corresponding to the target sensor subset; Based on the security task information, obtain the task importance weight corresponding to the security task; The task performance score corresponding to the target sensor subset is determined based on the feature matrix and the global feature library. The task allocation information corresponding to the target sensor subset is determined based on the feature complementarity score, the task importance weight, and the task performance score.

7. The method as described in claim 1, characterized in that, The step of performing autonomous driving control based on the region fusion perception features includes: Obtain current vehicle operation scenario information and local baseline policies; The scenario compensation increment is determined based on the cloud-based herd immunity network and the current vehicle operation scenario information. The target action vector is determined based on the local baseline strategy and the scene compensation increment. Automated driving control is performed based on the target motion vector.

8. The method as described in claim 7, characterized in that, The step of determining the scene compensation increment based on the current vehicle operation scene information further includes: Common failure modes are obtained from the cloud-based herd immunity network to obtain a set of common failure modes; The plugin complexity penalty term is determined based on the preset regularization coefficient and the plugin complexity regularization term; The scenario compensation increment is determined based on the current vehicle operation scenario information, the cloud-based herd immunity network, the common failure mode set, and the plugin complexity penalty term.

9. An automatic driving control device, characterized in that, The device includes: The data acquisition module is used to acquire multi-source sensing information and operating status information from vehicle-mounted sensors; The data processing module is used to determine the confidence information of the vehicle-mounted sensor in the sub-region based on the environmental state information and the operating state information; The data processing module is further configured to determine the fusion weight of the vehicle-mounted sensor in the corresponding sub-region based on the environmental state information and the confidence information; The data fusion module is used to fuse the multi-source sensing information according to the fusion weight to obtain regional fusion sensing features; The driving control module is used to perform autonomous driving control based on the region fusion perception features.

10. An automatic driving control device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the autonomous driving control method as described in any one of claims 1 to 8.

Images