Chassis control method and device based on risk field construction, electronic equipment and system

Abstract

The application relates to a chassis control method and device based on a risk field construction, electronic equipment and system, and belongs to the technical field of vehicle active safety and chassis cooperative control. The method comprises the following steps: performing probability fusion on multi-source sensing data based on a Bayesian network to obtain fused sensing information; obtaining a state estimation value and a corresponding covariance matrix by adopting a Bayesian filtering method based on the fused sensing information; constructing a comprehensive risk field based on environmental obstacle information in the fused sensing information and the state estimation value; performing probability prediction by adopting a Monte Carlo sampling method based on the comprehensive risk field and the covariance matrix to obtain a risk probability distribution, and performing cooperative control on a chassis actuator based on the risk probability distribution. The application can realize accurate quantification and dynamic prediction of vehicle driving risk, and realize adaptive cooperative control of a chassis multi-subsystem based on uncertainty quantification results, so that the steering stability and active safety of a vehicle under complex working conditions are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle active safety and chassis cooperative control technology, and in particular to a chassis control method, device, electronic device and system based on risk field construction. Background Technology

[0002] With the development of vehicle electrification and intelligence, vehicle active safety technology is gradually shifting from passive protection to active warning and intervention control. Active safety systems such as Electronic Stability Control (ESC), Active Front Steering (AFS), Direct Yaw Control (DYC), and Active Suspension System (ASS) have been gradually applied in mass-produced vehicles, and the coordinated control of multiple chassis subsystems has become a research hotspot in the industry.

[0003] In terms of chassis cooperative control, existing technologies acquire real-time vehicle operating parameters through a drive-by-wire chassis domain controller, calculate vehicle stability margin, construct an instability risk field model, dynamically correct actuator performance boundaries based on instability risk values, and solve for control commands through optimization algorithms. Regarding vehicle state estimation, existing technologies employ filtering methods such as Kalman filtering to estimate vehicle driving states. For multi-source perception fusion, existing technologies collect environmental data through multimodal sensors and utilize methods such as Bayesian deep neural networks for target recognition and state estimation. In AFS / DYC cooperative control, existing technologies coordinate active front wheel steering and direct yaw moment control based on extension theory or game theory. Regarding risk prediction, existing technologies use Monte Carlo methods to extract trajectory sampling points for risk field intensity calculation.

[0004] However, existing technologies still have the following shortcomings: First, the construction of the risk field mainly relies on deterministic indicators such as vehicle stability margin, without considering the uncertainty inherent in the state estimation itself, which makes it difficult to guarantee the accuracy of risk judgment under high uncertainty conditions; Second, there is a lack of effective linkage between risk assessment and chassis control decisions, and the risk quantification results are not fully transmitted to the control link, making it difficult to achieve coordinated optimization of risk assessment and control decisions. Summary of the Invention

[0005] In view of this, it is necessary to provide a chassis control method, device, electronic device and system based on risk field construction to solve the technical problems in the prior art where the accuracy of risk judgment is difficult to guarantee and there is a lack of effective coordinated control linkage between risk assessment and chassis control decision.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a chassis control method based on risk field construction, comprising: Collect multi-source perception data of the vehicle, and perform spatiotemporal synchronization and preprocessing on the multi-source perception data; Based on Bayesian networks, probabilistic fusion is performed on preprocessed multi-source sensing data to obtain fused sensing information. Based on the fused perception information, a Bayesian filtering method is used to estimate the vehicle's driving state and key parameters, obtaining the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. Environmental obstacle information is extracted from the fused perception information, and a comprehensive risk field is constructed based on the environmental obstacle information and the state estimate. Using the comprehensive risk field and the covariance matrix as prior information, the Monte Carlo sampling method is used to predict the probability of risk evolution and obtain the risk probability distribution. The vehicle's multiple chassis actuators are controlled collaboratively based on the aforementioned risk probability distribution.

[0007] In one possible implementation, the spatiotemporal synchronization and preprocessing of the multi-source sensing data includes: The global satellite navigation system timing system is used to synchronize the clocks of all sensors so that all sensor data use the same timestamp as a unified reference. The sensing data acquired by each sensor is uniformly transformed to the vehicle coordinate system through a calibrated extrinsic parameter matrix, and the original sensing data is subjected to median filtering for noise reduction, bad value removal, and outlier detection.

[0008] In one possible implementation, the probabilistic fusion of preprocessed multi-source sensing data based on a Bayesian network includes: The measured values ​​of each sensor are used as observed variables, environmental conditions are used as latent variables, and the historical accuracy statistics of the sensors are used as prior probabilities. The conditional probability tables of each sensor are updated by combining the influence factors of environmental conditions on the reliability of sensor measurements. Based on the conditional probability table, the posterior probability of each sensor measurement under the current environmental conditions is calculated using Bayes' theorem. The normalized posterior probabilities are used as the fusion weights for each sensor, and the probabilistic fusion of multi-source sensing data is completed in a weighted manner.

[0009] In one possible implementation, constructing the integrated risk field based on the environmental obstacle information and the state estimate includes: A collision risk field is constructed based on the environmental obstacle information; An instability risk field is constructed based on the state estimate; The collision risk field and the instability risk field are fused to obtain the comprehensive risk field.

[0010] In one possible implementation, the use of the Monte Carlo sampling method to probabilistically predict risk evolution includes: Several vehicle state sampling particles are generated based on the covariance matrix, and the size of the covariance matrix is ​​positively correlated with the dispersion of the vehicle state sampling particles. Using each of the vehicle state sampling particles as the initial state, state recursion is performed based on the vehicle dynamics model within a preset prediction time domain to obtain the predicted trajectory of each vehicle state sampling particle. Based on the comprehensive risk field, the risk field value of the predicted trajectory of each vehicle state sampling particle at each time in the prediction time domain is calculated, and the risk field evolution trajectory of each vehicle state sampling particle is obtained. The risk probability distribution is obtained by probabilistically predicting the risk field evolution trajectory of all vehicle state sampled particles.

[0011] In one possible implementation, the coordinated control of multiple chassis actuators of the vehicle based on the risk probability distribution includes: The risk level of the vehicle is determined based on the risk probability distribution. The combination of active safety subsystems to be activated under the current operating condition is determined based on the risk level. The risk level and the combination of subsystems are then sent down to the middle model prediction and control layer through the upper-layer strategy scheduling layer. The middle-level model predictive control layer receives the risk level and the subsystem combination, constructs a multi-objective optimization problem with the objectives of minimizing the vehicle yaw rate tracking error, suppressing the center of gravity sideslip angle, and resisting rollover, and solves it using model predictive control. The optimal control command obtained from the solution is then sent down to the lower-level execution allocation layer through the middle-level model predictive control layer. The optimal control command is received by the lower execution allocation layer. Based on the optimal control command, a non-cooperative game theory method is used to coordinate the active front wheel steering subsystem and the direct yaw moment control subsystem. Nash equilibrium is used as the optimal coordination strategy, and the game control is driven by the risk probability distribution. The coordinated control command is then distributed to each actuator to complete the collaborative control.

[0012] In one possible implementation, the method further includes: Monitor the health status of each chassis actuator; Determine the adjustment strategy based on the described health status; Based on the aforementioned adjustment strategy, the collaborative control tasks corresponding to each chassis actuator are adjusted.

[0013] Secondly, the present invention also provides a chassis control device based on a risk field, comprising: The multi-source sensing data acquisition and preprocessing module is used to acquire multi-source sensing data of the vehicle and perform spatiotemporal synchronization and preprocessing on the multi-source sensing data. The multi-source sensing data fusion module is used to perform probabilistic fusion of preprocessed multi-source sensing data based on a Bayesian network and output the fused sensing information. The vehicle state Bayesian estimation module is used to estimate the vehicle's driving state and key parameters based on the fused perception information using a Bayesian filtering method, and outputs the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. The integrated risk field construction module is used to extract environmental obstacle information from the fused perception information and construct an integrated risk field based on the environmental obstacle information and the state estimate. The Monte Carlo risk propagation prediction module is used to predict the probability of risk evolution using the integrated risk field and the covariance matrix as prior information and the Monte Carlo sampling method to obtain the risk probability distribution. The chassis multi-subsystem collaborative control module is used to collaboratively control multiple chassis actuators of the vehicle based on the risk probability distribution.

[0014] Thirdly, the present invention also provides an electronic device, including a data collector, a memory, and a processor, wherein the data collector is used to collect multi-source perception data of a vehicle; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the chassis control method based on risk field construction as described in any of the above implementations.

[0015] Fourthly, the present invention also provides a chassis control system based on a risk field, the chassis control system based on a risk field includes an electronic device and a result output device as described above, the result output device is connected to the electronic device, and the result output device is used to output and execute chassis cooperative control.

[0016] The beneficial effects of this invention are as follows: The chassis control method based on risk field construction provided by this invention explicitly introduces the uncertainty of state estimation into the instability risk field, enabling the risk field value to adaptively reflect the credibility of the estimation result, thus solving the problem of misjudgment or omission under high uncertainty in traditional risk fields; by using the covariance matrix to directly drive the dynamic adjustment of the control command boundary, the confidence perception of the control strategy is realized, that is, the more inaccurate the estimation, the more conservative the control; by establishing a full-link probabilistic closed-loop architecture from perception fusion to state estimation, and from risk field construction to control boundary adjustment, a reverse-driven closed loop from uncertainty quantification to control decision-making is realized; by applying non-cooperative Nash equilibrium game to AFS and DYC coordination, and with game control driven by the predicted risk distribution, deep collaboration between risk assessment and control decision-making is achieved. Therefore, this invention successfully solves the technical problems in existing technologies, such as the failure to consider the uncertainty of state estimation in the instability risk field leading to misjudgment or omission, the failure to propagate the uncertainty of perception fusion downstream, and the lack of a closed-loop collaborative mechanism based on a unified uncertainty metric between risk assessment and control decision-making, achieving high-precision risk quantification and adaptive collaborative control of vehicles under complex operating conditions. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an embodiment of the chassis control method based on risk field construction provided by the present invention; Figure 2 A schematic diagram of the overall process of an embodiment of the chassis control method based on risk field construction provided by the present invention; Figure 3 for Figure 1 A flowchart of a method according to an embodiment of step S101; Figure 4 A flowchart illustrating the spatiotemporal synchronous acquisition and preprocessing of multi-source sensing data in an embodiment of the chassis control method based on risk field construction provided by the present invention. Figure 5 A schematic diagram of the multi-source data fusion principle based on dynamic Bayesian network, which is an embodiment of the chassis control method based on risk field construction provided by the present invention. Figure 6 for Figure 1 A flowchart of a method according to an embodiment of step S102; Figure 7 A flowchart of vehicle state estimation using Bayesian uncertainty quantification, as an embodiment of the chassis control method based on risk field construction provided by the present invention. Figure 8 A functional block diagram of the chassis control device based on risk field construction provided by the present invention; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] This invention provides a chassis control method, device, electronic device, and system based on risk field construction, which will be described below.

[0023] Figure 1 This is a flowchart illustrating an embodiment of the chassis control method based on risk field construction provided by the present invention. The executing entity of the chassis control method based on risk field construction can be a computer, a portable smart device, or a cloud server, etc. The specific executing entity is the storage medium in the computer, but this embodiment does not limit it.

[0024] like Figure 1 As shown, the chassis control method based on risk field construction includes: S101. Collect multi-source perception data of the vehicle, and perform spatiotemporal synchronization and preprocessing on the multi-source perception data.

[0025] In the specific implementation, this embodiment can refer to Figure 2As shown, the overall process includes the following steps: spatiotemporal synchronous acquisition and preprocessing of multi-source sensing data; fusion of multi-source sensing data based on dynamic Bayesian networks; Bayesian uncertainty quantification estimation of vehicle driving status and key parameters; construction of a spatiotemporal risk field that integrates collision risk and instability risk; risk propagation prediction based on Monte Carlo sampling; dynamic collaborative control of chassis multi-subsystems based on predicted risk distribution; control allocation based on actuator health status; and adjustment of control command allocation strategy according to the health status of each actuator when any actuator is detected to have a fault or performance degradation.

[0026] In this embodiment, multi-source perception data is collected in real time through an onboard sensor array, which includes an inertial measurement unit (IMU), a global navigation satellite system (GNSS) receiver, wheel speed sensors, a steering wheel angle sensor, a lidar, a millimeter-wave radar, and a vision camera. The IMU measures the vehicle's three-axis acceleration and three-axis angular velocity; the GNSS receiver measures the vehicle's position and speed; the wheel speed sensors collect the rotational speeds of each wheel; the steering wheel angle sensor measures the steering wheel angle; the lidar acquires three-dimensional point cloud data of the surrounding environment; the millimeter-wave radar detects the distance and relative speed of obstacles in front and to the sides; and the vision camera collects image information of lane lines, traffic signs, and obstacles. All of this information and these parameters constitute the multi-source perception data.

[0027] In some embodiments of the present invention, reference is made to... Figure 3 As shown, the spatiotemporal synchronization and preprocessing process in step S101 specifically includes: S301. The global satellite navigation system timing system is used to synchronize the clocks of all sensors so that all sensor data are based on the same timestamp as a unified reference.

[0028] S302. Transform the perception data acquired by each sensor into the vehicle coordinate system through the calibrated extrinsic parameter matrix, and perform median filtering for noise reduction, bad value removal, and outlier detection on the original perception data.

[0029] Specifically, the collected multi-source sensing data undergoes spatiotemporal synchronization preprocessing. Time synchronization uses a GNSS timing system to synchronize the clocks of each sensor, ensuring that all sensor data uses the same timestamp as a unified reference, guaranteeing alignment accuracy across the time dimension. Spatial synchronization transforms the sensing data acquired by each sensor into the vehicle coordinate system using a calibrated extrinsic parameter matrix, eliminating spatial deviations caused by different sensor installation locations. Preprocessing also includes median filtering for noise reduction, outlier removal, and outlier detection to improve data quality. The above process can be further referenced... Figure 4 As shown, Figure 4This is a flowchart of the spatiotemporal synchronous acquisition and preprocessing of multi-source sensing data. Through the above processing, the data from various sensors are synchronized under a unified spatiotemporal reference, providing accurate and reliable input for subsequent multi-source sensing data fusion.

[0030] S102. Based on Bayesian network, perform probabilistic fusion on the preprocessed multi-source sensing data to obtain fused sensing information.

[0031] In this embodiment, a Bayesian network model is constructed using the preprocessed multi-source sensing data as input to perform probabilistic fusion of the multi-source sensing data.

[0032] Reference Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the principle of multi-source data fusion based on dynamic Bayesian networks. Figure 5 As shown, the measured values ​​of each sensor node (obstacle distance detected by lidar, relative speed detected by millimeter-wave radar, lane line information detected by visual camera, etc.) are used as observed variables, while environmental conditions (including weather conditions, light intensity, and road type) are used as latent variables. Using historical accuracy statistics of the sensors as prior probabilities, and combining these with factors influencing the reliability of sensor measurements under environmental conditions (e.g., the reliability reduction factor for lidar under rainy conditions and the reliability reduction factor for visual cameras under backlight conditions), a conditional probability table for each sensor under various environmental conditions is determined. The conditional probability table reflects the probability distribution of each sensor's measured values ​​under given environmental conditions and is the core of Bayesian inference.

[0033] In some embodiments of the present invention, reference is made to... Figure 6 As shown, step S102 specifically includes: S601. Take the measured values ​​of each sensor as the observed variable, the environmental conditions as the latent variable, and the historical accuracy statistics of the sensors as the prior probability. Combine the influence factors of environmental conditions on the reliability of sensor measurement to update the conditional probability table of each sensor.

[0034] S602. Based on the conditional probability table, calculate the posterior probability of each sensor measurement under the current environmental conditions using Bayes' theorem.

[0035] S603. The posterior probability is normalized and used as the fusion weight of each sensor. The probabilistic fusion of multi-source sensing data is completed in a weighted manner.

[0036] Specifically, based on the conditional probability table, the posterior probability of each sensor measurement under the current environmental conditions is calculated using Bayes' theorem:

[0037] in, This indicates the measurement reliability status of the i-th sensor. Represents the current environmental conditions. Posterior probability. This reflects the observation of current environmental conditions. Then, the sensor Updated beliefs about measurement reliability.

[0038] The posterior probabilities are normalized and used as fusion weights for each sensor. Probabilistic fusion of multi-source sensing data is completed in a weighted manner, and the uncertainty measure of the fusion result is output. By transmitting the fusion uncertainty as an independent information flow downstream, this approach differs from existing dynamic Bayesian fusion schemes that only apply Bayesian methods to the sensing layer and do not transmit the fusion uncertainty downstream.

[0039] S103. Based on the fused perception information, the vehicle's driving state and key parameters are estimated using a Bayesian filtering method to obtain the state estimate and its corresponding covariance matrix.

[0040] In this embodiment, the fused perception information is a comprehensive perception result output after probabilistic fusion of multi-source perception data based on a Bayesian network. Specifically, it includes two parts: first, environmental obstacle information, including the distance, relative speed, and category of obstacles, which is used to construct a collision risk field in subsequent steps; second, vehicle motion observation information, including position and velocity obtained from a GNSS receiver, acceleration and angular velocity obtained from an IMU, wheel speed obtained from a wheel speed sensor, and steering wheel angle obtained from a steering wheel angle sensor, which is used for vehicle state estimation in this step.

[0041] It should be noted that during probabilistic fusion, the posterior probability of each sensor's measurement is calculated based on Bayes' theorem, and this normalized posterior probability is used as the fusion weight for each sensor. This fusion weight reflects the reliability of each sensor's measurement under the current environmental conditions; a higher fusion weight indicates a more reliable sensor measurement. The fusion weight is used to adjust the observation noise covariance of each observation in the Bayesian filter. Specifically, the fusion weight of each sensor is mapped to the adjustment coefficient of its corresponding observation noise covariance. A higher fusion weight corresponds to a smaller observation noise covariance, meaning the Bayesian filter has a higher degree of confidence in the sensor's observation information; conversely, a lower fusion weight corresponds to a larger observation noise covariance, meaning the Bayesian filter has a lower degree of confidence in the sensor's observation information. Therefore, the fusion result (fusion weight) directly participates in the Bayesian filter's observation update process, realizing the transfer of uncertainty from the perception layer to the downstream state estimation stage.

[0042] Based on the vehicle motion observation information from the fused perception information, a Bayesian filtering method is used to estimate the vehicle's driving state and key parameters in real time. Bayesian filtering recursively fuses prediction and observation information, outputting the posterior probability distribution of the state estimate at each time step, thus providing both the state estimate and its uncertainty measure. The covariance matrix is ​​the state estimate covariance matrix output during the Bayesian filtering recursion process; its diagonal elements represent the variance of each state estimate, used to quantitatively characterize the uncertainty of the state estimate.

[0043] It should be noted that the driving state includes longitudinal speed, lateral speed, yaw rate, four-wheel speed, and front wheel steering angle, and the key parameter includes the road adhesion coefficient. Among them, longitudinal speed, lateral speed, and yaw rate are the core state variables characterizing the vehicle's motion state; four-wheel speed and front wheel steering angle reflect the operating state of each wheel and steering system; and the road adhesion coefficient characterizes the adhesion characteristics between the tires and the road surface, and is a key parameter affecting vehicle dynamics and stability control. The road adhesion coefficient and the vehicle motion state are jointly estimated using Bayesian filtering.

[0044] Reference Figure 7 As shown, Figure 7 A flowchart for vehicle state estimation using Bayesian uncertainty quantification. Figure 7 As shown, a vehicle dynamics model is established as a priori for state prediction. This embodiment uses a seven-degree-of-freedom vehicle dynamics model, including longitudinal motion, lateral motion, yaw motion, and rotational motion of the four wheels. The state vector includes longitudinal velocity, lateral velocity, yaw rate, four-wheel speed, front wheel steering angle, and road adhesion coefficient.

[0045] Specifically, this embodiment employs Extended Kalman Filter (EKF) for recursive state estimation. EKF linearizes the nonlinear system by performing a first-order Taylor expansion of the nonlinear function (compiling the Jacobian matrix), and is a commonly used method for vehicle state estimation. Under strongly nonlinear conditions, Unscented Kalman Filter (UKF) can be used as an alternative. It handles the nonlinear transformation through a deterministic sampling method (unscented transformation), eliminating the need to compute the Jacobian matrix and typically offering better estimation performance.

[0046] The state update equation for the extended Kalman filter is as follows:

[0047]

[0048] in, This is the estimated state value at the current moment. This is the state estimate from the previous time step. Estimate the covariance matrix for the current state. The covariance matrix is ​​estimated for the state at the previous time step. For Kalman gain, For the observation vector, Let be the Jacobian matrix of the observation equation. In the prediction phase, the dynamic model is used to propagate the state from the previous time step to the current time step, while considering the influence of process noise. In the update phase, the Kalman gain is used to perform optimal weighted fusion between the predicted and observed values, so as to minimize the covariance of the posterior estimate.

[0049] Through recursive iteration, the estimated values ​​and their covariance matrices of the centroid sideslip angle, yaw rate, vertical load of each wheel, and road adhesion coefficient are output, quantitatively characterizing the uncertainty of each state estimate. The diagonal elements of the covariance matrix represent the variance of each state estimate, and the standard deviation is obtained by taking the square root.

[0050] S104. Extract environmental obstacle information from the fused perception information, and construct a comprehensive risk field based on the environmental obstacle information and the state estimate.

[0051] In this embodiment, environmental obstacle information is extracted from the fused perception information, and two subfields, collision risk field and instability risk field, are constructed simultaneously based on the environmental obstacle information and the vehicle driving state estimated in the above steps. The two are then fused to obtain a comprehensive risk field.

[0052] In some embodiments of the present invention, the process of constructing a comprehensive risk field specifically includes constructing a collision risk field based on the environmental obstacle information; constructing an instability risk field based on the state estimate; and fusing the collision risk field and the instability risk field to obtain the comprehensive risk field.

[0053] Specifically, the process of constructing the collision risk field is as follows: The fused obstacle and lane line information is transformed into obstacle potential energy fields and lane line potential energy fields, respectively. The obstacle potential energy field characterizes the risk of collision between the vehicle and surrounding static and dynamic obstacles, while the lane line potential energy field characterizes the risk of the vehicle deviating from the lane boundary. The obstacle potential energy field is established using a modified Gaussian potential function, and the potential energy amplitude is related to the type, size, and relative speed of the obstacle. The scale parameter reflects the influence range of the obstacle. The fused obstacle and lane line information are then superimposed to form the collision risk field.

[0054] The process of constructing the potential energy field of the obstacle is as follows: The obstacle potential energy field characterizes the collision risk to a vehicle posed by static and dynamic obstacles in a road environment. Obstacles include static obstacles (such as stopped vehicles, medians, roadblocks, etc.) and dynamic obstacles (such as moving vehicles, pedestrians, non-motorized vehicles, etc.).

[0055] The obstacle potential energy field is established using a Gaussian potential function, and its field strength increases as the distance between the vehicle and the obstacle decreases.

[0056] in, Let be the potential energy field strength of the obstacle at position (x, y). The total number of obstacles within the sensing range. Let i be the center position of the i-th obstacle. Let be the potential energy amplitude of the i-th obstacle. , These are scale parameters related to the size and motion state of the obstacle.

[0057] Among them, potential energy amplitude The method for determining it is as follows: Static obstacles: It is related to the type and size of the obstacle; for example, the potential energy amplitude of a large obstacle (such as a stopped vehicle or a median strip) is greater than that of a small obstacle (such as a roadblock or cones). Dynamic obstacles: It is related to the relative speed of the obstacle; for example, the greater the relative speed, the greater the potential energy amplitude, reflecting the emergency risk of approaching the obstacle at high speed.

[0058] Scale parameters , The method for determining the scale parameter is as follows: , It is related to the physical size of the obstacle; for example, the larger the obstacle, the wider the area affected. The larger the value, the higher the scale parameter. , It is related to the motion state of the obstacle. For example, the influence range of a dynamic obstacle along the direction of motion is greater than the lateral range (i.e., the influence range perpendicular to the direction of motion). (Along the direction of motion) greater than (Perpendicular to the direction of movement).

[0059] Furthermore, the process of constructing the potential energy field of the lane line is as follows: The lane line potential energy field characterizes the collision risk that arises when a vehicle deviates from the lane centerline or approaches the lane boundary. When a vehicle deviates from the lane centerline, the risk of collision with oncoming vehicles or roadside obstacles increases.

[0060] Lane line potential energy field The construction method is as follows:

[0061] in, The lateral position of the vehicle within the current lane. The position of the lane center line. The potential energy amplitude of the lane. These are the scale parameters related to lane width. Let be the potential energy term at the lane boundary, and its expression is:

[0062] in, and These are the left and right lane boundary positions, respectively. Let the boundary potential energy amplitude be... This is the scale parameter for the boundary influence range.

[0063] The physical meaning of the lane line potential energy field is that the lane line potential energy field value is the lowest when the vehicle is driving near the center line of the lane; when the vehicle deviates from the center line and approaches the lane boundary, the potential energy field value increases sharply, indicating that the risk of the vehicle leaving the lane or colliding with vehicles in adjacent lanes increases.

[0064] Furthermore, the collision risk field synthesis process is as follows: By superimposing the potential energy fields of the obstacle and the lane lines, a comprehensive collision risk field is obtained:

[0065] This collision risk field comprehensively considers the direct collision risk from static / dynamic obstacles in the road environment and the indirect collision risk caused by lane departure, providing input for subsequent weighted fusion with the instability risk field.

[0066] It should be noted that the collision risk field has the following characteristics: Anisotropy: The influence range of an obstacle along the direction of motion is greater than that along the lateral direction, reflecting the movement trend of a dynamic obstacle; Distance decay: The field strength decreases monotonically with increasing distance, which is consistent with the basic characteristics of an artificial potential energy field; Multi-source superposition: The potential energy fields of multiple obstacles can be linearly superimposed, which is suitable for complex traffic environments.

[0067] Furthermore, the process of constructing the instability risk field is as follows: With yaw rate error And centroid side slip angle error The weighted combination is used as a characteristic quantity of instability risk. Among them, This is the reference value for yaw rate. The reference value for the center of gravity sideslip angle is calculated by a linear two-degree-of-freedom vehicle model based on the current vehicle speed and steering wheel angle.

[0068] Existing research has shown that the centroid sideslip angle and yaw rate z These are core state variables characterizing vehicle stability. The deviation between the actual and nominal values ​​of yaw rate and sideslip angle can be used to effectively determine the vehicle's stability. The difference between the actual and nominal values ​​of sideslip angle and yaw rate is used to determine the vehicle's stability state. Phase plane analysis divides the phase plane into stable and unstable regions, establishing a stable boundary model. Within the unstable region, the vehicle's instability is used to characterize the degree to which the vehicle deviates from a stable state in real time.

[0069] Based on this, the present invention introduces the corresponding estimated standard deviation. , As a moderating factor, an instability risk field incorporating uncertainty information is constructed:

[0070] in, For the field strength value of instability risk, These are weighting coefficients. This refers to the diagonal element in the 2nd row and 2nd column. This represents the diagonal element in the 3rd row and 3rd column. The determination method is as follows: Based on vehicle parameters (wheelbase, center of gravity height, track width, etc.), the weight ratio is initially determined through a vehicle dynamics model. Then, it is calibrated and optimized using real vehicle or simulation data from typical operating conditions (double lane change test, sinusoidal sweep steering test). The typical value range is... ,and .

[0071] The key feature of this formula is that it introduces the standard deviation of the state estimate as the denominator into the instability risk field (i.e., (Weighted).

[0072] The physical meaning of this expression is: When the uncertainty of the state estimate is high ( When (large), Automatic reduction avoids excessive intervention based on unreliable information. This is because, under conditions of high uncertainty, the state estimate... and The reliability of the system is low. If it is still used to make risk judgments with high weight, it is easy to cause false triggering—that is, to trigger stability control incorrectly when the vehicle is actually stable, or to fail to trigger it in time when the vehicle is actually unstable.

[0073] When uncertainty is low ( Hour, It exhibits a sensitive response, quickly identifying even minute state deviations. This is because, under low uncertainty conditions, the state estimate... and The reliability of the estimation is high, and the control system can rely on the estimation results to make accurate risk judgments, thereby achieving "confidence-adaptive" risk field.

[0074] This feature is fundamentally different from existing methods for constructing instability risk fields that do not incorporate variance estimation. Existing phase plane methods are based on... Phase planes divide stable and unstable regions, and instability degrees are used to characterize the degree to which a vehicle deviates from a stable state within the unstable region. However, neither approach incorporates the uncertainty of state estimation into risk quantification. This invention addresses this issue by... The weighting mechanism enables the instability risk field to simultaneously reflect the degree to which the vehicle deviates from a stable state and the reliability of the state estimate.

[0075] Furthermore, the process of constructing the comprehensive risk field is as follows: The collision risk field and the instability risk field are weighted and fused to generate a comprehensive spatiotemporal risk field:

[0076] in, This represents the collision risk field value. For the field value of instability risk, and For the corresponding weight coefficients, and satisfying , .

[0077] Dynamic adjustment method for weighting coefficients: and The system is dynamically adjusted in real time based on vehicle speed, steering wheel angle, and road surface adhesion coefficient, with the following adjustment rules: (1) Adjustment based on vehicle speed: The higher the vehicle speed, the greater the risk of vehicle dynamic instability, and the greater the instability risk field weight. The corresponding increase. Define the vehicle speed factor:

[0078] in, Set a preset vehicle speed threshold (e.g., 80 km / h). hour, ;when hour, .

[0079] (2) Based on steering wheel angle adjustment: The larger the steering wheel angle or the greater the rate of change of the angle, the more likely the driver is to make an emergency steering intention, increasing the risk of vehicle instability. The instability risk field weight The corresponding increase. Define the steering factor:

[0080] in, Maximum steering wheel angle, The maximum rate of change of steering wheel angle. This is the weighting coefficient (typical value 0.5).

[0081] (3) Adjustment based on road surface adhesion coefficient: The lower the road surface adhesion coefficient, the smaller the vehicle stability margin, and the greater the instability risk field weight. The corresponding increase. Define the adhesion coefficient factor:

[0082] in, This is the nominal road adhesion coefficient (for dry asphalt pavement, a typical value of 0.85). When hour, .

[0083] (4) Comprehensive determination of weighting coefficients: Combining the above three factors, the instability risk field weights are determined. Determine by the following formula:

[0084] Collision risk field weight for:

[0085] in, and These are the minimum and maximum values ​​of the instability risk field weights, respectively (typically ranging from...). , ).

[0086] The physical significance of this dynamic weight determination method lies in: When a vehicle is traveling at high speed, making an emergency turn, or on a low-friction surface, the risk of vehicle instability is more prominent. The weight of the instability risk field should be appropriately increased so that the control system pays more attention to the vehicle's own dynamic stability. When a vehicle is traveling at low speed, making small turns, or on a high-friction surface, the risk of collision is relatively more prominent. The weight of the collision risk field should be appropriately increased so that the control system pays more attention to the collision risk of environmental obstacles.

[0087] (5) Weighting in extreme cases: When no obstacles are detected around the vehicle, the collision risk field Approaching zero, even at this point The overall risk field is relatively large, and it is still mainly determined by the instability risk field. Similarly, when the vehicle is in a completely stable state (both the sideslip angle and yaw rate errors are zero), the instability risk field is relatively small. Approaching zero, the overall risk field is primarily determined by the collision risk field. This design ensures that the overall risk field effectively characterizes the dominant risk sources under any operating condition.

[0088] S105. Using the comprehensive risk field and the covariance matrix as prior information, the Monte Carlo sampling method is used to predict the probability of risk evolution and obtain the risk probability distribution.

[0089] In this embodiment, the obtained comprehensive risk field and covariance matrix are used as prior information, and the Monte Carlo sampling method is used to sample and predict the risk evolution within the preset prediction time domain.

[0090] In some embodiments, the probability prediction process specifically includes: generating a plurality of vehicle state sampling particles based on the covariance matrix, wherein the size of the covariance matrix is ​​positively correlated with the dispersion of the vehicle state sampling particles; using each of the vehicle state sampling particles as an initial state, performing state recursion based on the vehicle dynamics model within a preset prediction time domain to obtain the predicted trajectory of each vehicle state sampling particle; calculating the risk field value of each vehicle state sampling particle's predicted trajectory at each moment in the prediction time domain based on the comprehensive risk field to obtain the risk field evolution trajectory of each vehicle state sampling particle; and performing probability prediction based on the risk field evolution trajectories of all vehicle state sampling particles to obtain the risk probability distribution.

[0091] It should be understood that the sampled particles of this invention are directly derived from the output state estimation covariance matrix, directly coupling the uncertainty quantification result of Bayesian filtering with risk propagation prediction. Through the propagation of the covariance matrix, the known error covariance is used to assess the uncertainty of the predicted trajectory, realizing the end-to-end transmission from state estimation uncertainty to future risk probability distribution.

[0092] S106. Based on the risk probability distribution, perform coordinated control of multiple chassis actuators of the vehicle.

[0093] In this embodiment, the predicted comprehensive risk probability distribution is used as the control input, and the active front steering subsystem (AFS), direct yaw moment control subsystem (DYC), and active suspension subsystem (ASS) installed on the vehicle chassis are used as control actuators, adopting a hierarchical collaborative control architecture.

[0094] In some embodiments, the specific process of cooperative control includes: determining the risk level of the vehicle based on the risk probability distribution; determining the combination of active safety subsystems to be activated under the current operating condition based on the risk level; and sending the risk level and the subsystem combination to the middle-level model predictive control layer through the upper-level strategy scheduling layer; receiving the risk level and the subsystem combination through the middle-level model predictive control layer; constructing a multi-objective optimization problem with the objectives of minimizing the vehicle yaw rate tracking error, suppressing the center of gravity sideslip angle, and resisting rollover as objectives; solving the problem using model predictive control; and sending the optimal control command obtained from the solution to the lower-level execution allocation layer through the middle-level model predictive control layer; receiving the optimal control command through the lower-level execution allocation layer; coordinating the active front wheel steering subsystem and the direct yaw moment control subsystem based on the optimal control command using a non-cooperative game theory method, with Nash equilibrium as the optimal coordination strategy, and the game control driven by the risk probability distribution; and allocating the coordinated control command to each actuator to complete the cooperative control.

[0095] Specifically, the upper-level policy scheduling layer includes: Based on the predicted future risk probability distribution, the vehicle's risk level is determined, and the appropriate combination of active safety subsystems to be activated under the current operating conditions is identified. Risk level classification is based on the predicted risk probability exceeding a preset threshold, and is specifically divided into four levels: Low-risk zone: The predicted risk probability is below the first threshold, the vehicle is in a stable state, and only AFS is activated for minor stability intervention; Medium-risk area: When the predicted risk probability is between the first and second thresholds, AFS+DYC joint intervention is activated; High-risk areas: When the predicted risk probability is between the second and third thresholds, AFS+DYC+ASS is used for comprehensive intervention. Critical instability zone: When the predicted risk probability is higher than the third threshold, the vehicle faces instability risk. Strong intervention is mainly carried out with DYC+ASS, and AFS is used as an auxiliary.

[0096] The mid-level model predictive control layer is: With the objectives of minimizing the vehicle's yaw rate tracking error, suppressing the center of gravity sideslip angle, and preventing rollover, a coordinated optimization problem comprising three subsystems is constructed:

[0097] in, For yaw rate tracking error, This is the error in the centroid sideslip angle. The lateral load transfer rate (used to characterize rollover risk). These are the control inputs for each subsystem. and These are the corresponding weighting coefficients. The first three terms of the objective function are performance cost terms (corresponding to yaw stability, centroid sideslip angle suppression, and rollover resistance, respectively), and the last three terms are control energy consumption cost terms.

[0098] Model predictive control (MPC) is used to solve the above optimization problem online. MPC solves a finite-time open-loop optimal control problem at each sampling time by rolling time-domain optimization, applies the first control variable to the system, and repeats the process at the next time-domain.

[0099] The covariance matrix is ​​estimated using the output of the above method. The control command boundaries are dynamically adjusted, and the contraction factor adopts an exponential function form: ) in, It is a contractile factor. The trace of the covariance matrix represents the total uncertainty; As a preset positive constant, its value is determined through offline calibration based on the system's sensitivity to state estimation uncertainty (typical value range is 0.01~0.5, and the specific value needs to be determined through simulation and real vehicle calibration based on vehicle parameters and sensor characteristics).

[0100] This exponential contraction function has the following mathematical properties: Monotonically decreasing property: Follow The value increases and then monotonically decreases, ensuring that the control boundary contraction is more significant when the uncertainty of the state estimate is higher. Boundedness: This ensures that the boundaries of control commands are always within physical constraints; Asymptoticity: when When the state estimate is completely determined, The control boundary is restored to the nominal limit; when When the state estimate is completely unreliable, When the control boundary approaches zero, the system enters the most conservative control mode.

[0101] Multiply the upper limit of each control variable in MPC by the contraction factor. When the uncertainty of state estimation is high, the control command boundaries of AFS and DYC are automatically narrowed:

[0102] in, The nominal control boundary (determined based on actuator physical limits and design parameters). This is the actual control boundary after adjusting for state estimation uncertainty.

[0103] The core idea of ​​this design is: when the trace of the state estimation covariance matrix... A large error indicates that the current state estimate is unreliable, and the control system should actively reduce the control gain and narrow the control boundary to avoid vehicle instability caused by control overshoot based on erroneous estimates. When the state estimate accuracy is high, the control system can make full use of the actuator capabilities to achieve better tracking performance. This mechanism enables the control strategy to have "confidence-awareness" capabilities, that is, the more inaccurate the estimate, the more conservative the control.

[0104] contraction factor The calibration needs to consider factors such as sensor noise characteristics, actuator response capability, and vehicle stability margin. Specifically, it is determined through a combination of offline simulation and real-vehicle calibration: under typical operating conditions (such as double lane change and sinusoidal sweep steering), through simulation scanning... The optimal value is determined by taking a value in the range of 0.01 to 0.5, with the goal of optimizing the root mean square value of the yaw rate tracking error and the weighted comprehensive index of control energy consumption.

[0105] Simultaneously, the tire force constraint conditions are dynamically adjusted based on the estimated probability distribution of the road adhesion coefficient. The tire force constraint adopts an adhesion ellipse model, the boundary of which is dynamically adjusted according to the estimated value and uncertainty of the road adhesion coefficient. That is, when the uncertainty of the road adhesion coefficient is high, the boundary of the adhesion ellipse is appropriately shrunk to ensure that the control command is within the feasible region of tire force.

[0106] The lower-level execution allocation layer specifically includes: using a non-cooperative game theory (Nash equilibrium) method to coordinate AFS and DYC to avoid functional conflicts. Existing research shows that AFS and DYC systems can be used as game objects, and a game coordination controller with variable subsystem weights can be designed based on the non-cooperative feedback Nash equilibrium game theory. The ideal output at the Nash equilibrium point can be obtained by coupling the Riccati equation and its iterative transformation.

[0107] When both AFS and DYC are active, a non-cooperative game model is established. The players in the game are the AFS subsystem and the DYC subsystem, and each player's strategy is its control input, such as the additional front turn angle of the AFS. and the additional yaw moment of DYC When two participants are unable to reach a binding cooperation agreement, they each pursue their own profit maximization while minimizing interference with the other subsystem.

[0108] Each participant's payoff function consists of two parts: the effect of self-control (positive term) and the disturbance to another subsystem (negative term).

[0109]

[0110] in, The profit function value of the AFS subsystem. The profit function value of the DYC subsystem. and These are the self-control effects measures for AFS and DYC, respectively. and Each subsystem measures the degree of interference it causes to another subsystem. , , , These are the weighting coefficients. This reflects the degree of importance the subsystem places on its own control effectiveness. It reflects the degree of importance that a subsystem places on avoiding interference with another subsystem.

[0111] Specifically, Defined as the reduction in vehicle yaw rate tracking error under AFS control:

[0112] in, This refers to the yaw rate tracking error without AFS control. This represents the yaw rate tracking error after applying AFS control. The larger this value, the more significant the control effect of AFS.

[0113] Defined as the reduction in the tracking error of the vehicle's center of gravity sideslip angle under DYC control:

[0114] in, This refers to the tracking error of the centroid sideslip angle without DYC control. The tracking error of the centroid side slip angle after applying DYC control.

[0115] Defined as the degree of interference of AFS control on the DYC control target, it is measured by the change in additional yaw moment caused by the additional front wheel steering angle of AFS:

[0116] in, Add front wheel steering angle to AFS The resulting additional yaw moment (calculated using tire lateral slip characteristics). The larger this value, the more severe the interference of AFS control behavior with DYC.

[0117] Defined as the degree of interference of DYC control on the AFS control target, it is measured by the equivalent change in front wheel steering angle caused by the additional yaw moment of DYC:

[0118] in, Add yaw moment to DYC The resulting equivalent front wheel steering angle (calculated by inverse kinematics of the vehicle dynamics model).

[0119] Weighting coefficient , , , The method for determining it is as follows: Based on control priority: The control priorities of AFS and DYC are determined according to the current risk level and vehicle status. Generally, AFS has a higher priority in low-risk areas. In high-risk areas, DYC has a higher priority. ).

[0120] Game theory-based calibration: Through offline simulation, the range of weighting coefficients is scanned under typical operating conditions (such as double lane change and sinusoidal sweep steering). The final values ​​of each weighting coefficient are determined with the goal of optimizing the weighted comprehensive index of yaw rate tracking error and centroid sideslip angle suppression effect. The typical value range is as follows: , .

[0121] The optimal coordination strategy for the additional rotation angle of AFS and the additional yaw moment of DYC is obtained by solving the Nash equilibrium. Nash equilibrium point. The following conditions must be met:

[0122]

[0123] That is, given the DYC strategy In this case, AFS chooses Maximize its benefits Given an AFS strategy In this case, DYC chooses Maximize its benefits Neither party can gain greater benefits by changing its strategy alone.

[0124] The Nash equilibrium point can be solved using one of the following methods: Iterative algorithm: In each control cycle, one policy is fixed and the other policy is optimized alternately until convergence to the Nash equilibrium point; Coupled Riccati Equations: The game problem is transformed into a problem of solving coupled Riccati equations, and an analytical solution is obtained through iterative transformation.

[0125] Game control is driven by the aforementioned predicted risk distribution. For example, when the predicted risk is high, the weight of the control effect term in the game's payoff function increases, while the weight of the disturbance term decreases, making AFS and DYC more actively involved in stability control. When the predicted risk is low, the weight of the disturbance term increases to avoid unnecessary subsystem interference.

[0126] The lower execution distribution layer also distributes the generalized force commands solved in the middle layer to each actuator according to the current vertical load and adhesion status of each wheel, so as to achieve precise control of four-wheel independent drive / braking, steering and active suspension.

[0127] In some embodiments, fault-tolerant control is also included, specifically by monitoring the health status of each chassis actuator; determining an adjustment strategy based on the health status; and performing a collaborative control task for each chassis actuator based on the adjustment strategy.

[0128] Specifically, control allocation is based on the health status of the actuators. When a fault or performance degradation is detected in any actuator, the control command allocation strategy is adjusted according to the health status of each actuator.

[0129] The health status of each actuator is quantified into health factors. , This indicates that the actuator is completely healthy. This indicates that the executor has completely failed. Construct a diagonal weighted matrix. The control allocation problem is transformed into a weighted pseudo-inverse problem:

[0130] in, This refers to the actual control command vector for each actuator. It is a diagonal weighted matrix. The control efficiency matrix (reflecting the contribution of each actuator to the generalized force, derived from the vehicle dynamics model) is used. for The false rebellion, This is for the generalized force requirements output from the lower layer. Through this allocation method, actuators in good health undertake more control tasks, while the functionality of failed actuators is compensated by other healthy actuators, achieving functional fault-tolerant control of the distributed chassis. Health Factor The online estimation method is as follows: real-time monitoring is performed based on the actuator current-torque response deviation, and the health factor value is reduced accordingly when the response deviation exceeds the preset threshold.

[0131] This implementation explicitly introduces the uncertainty of state estimation into the instability risk field, enabling the risk field value to adaptively reflect the reliability of the estimation results. In other words, the risk field automatically becomes conservative when uncertainty is high, solving the problem of misjudgment or omission in traditional risk fields under high uncertainty. By directly driving the dynamic adjustment of the control command boundary using the covariance matrix, it achieves confidence perception of the control strategy; the more inaccurate the estimation, the more conservative the control. By establishing a closed-loop probability transfer across the entire chain from perception fusion, state estimation, risk field, risk prediction, MPC, and game coordination, it achieves a reverse-driven closed loop from uncertainty quantification to control decision-making. By applying non-cooperative Nash equilibrium game theory to AFS and DYC coordination, and with game control driven by the predicted risk distribution, the game result is directly regulated by the comprehensive risk field, achieving deep synergy between risk assessment and control decision-making. By dividing the risk field into low-risk, medium-risk, high-risk, and critical instability zones, and coordinating different controllers for stability intervention according to different risk zones, it maximizes the effect of dynamic coordination across all scenarios.

[0132] Based on the same idea as the chassis control method based on risk field construction in the above embodiments, this application also provides a chassis control device 800 based on risk field construction, which can be used to execute the chassis control method based on risk field construction described above. Figure 8 As shown, the chassis control device 800 based on the risk field construction specifically includes a multi-source perception data acquisition and preprocessing module 801, a multi-source perception data fusion module 802, a vehicle state Bayesian estimation module 803, a comprehensive risk field construction module 804, a Monte Carlo risk propagation prediction module 805, and a chassis multi-subsystem collaborative control module 806. In some embodiments, the above modules can be programmable software instructions stored in memory and executable by a processor. It is understood that in other embodiments, the above modules can also be program instructions or firmware embedded in the processor.

[0133] The multi-source sensing data acquisition and preprocessing module 801 is used to acquire multi-source sensing data of the vehicle and perform spatiotemporal synchronization and preprocessing on the multi-source sensing data. The multi-source sensing data fusion module 802 is used to perform probabilistic fusion of preprocessed multi-source sensing data based on a Bayesian network and output the fused sensing information. The vehicle state Bayesian estimation module 803 is used to estimate the vehicle's driving state and key parameters based on the fused perception information using a Bayesian filtering method, and output the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. The integrated risk field construction module 804 is used to extract environmental obstacle information from the fused perception information and construct an integrated risk field based on the environmental obstacle information and the state estimate. The Monte Carlo risk propagation prediction module 805 is used to predict the probability of risk evolution using the integrated risk field and the covariance matrix as prior information and the Monte Carlo sampling method to obtain the risk probability distribution. The chassis multi-subsystem collaborative control module 806 is used to collaboratively control multiple chassis actuators of the vehicle based on the risk probability distribution.

[0134] Please refer to Figure 9 , Figure 9 This is a schematic diagram of the structure of an electronic device 900 provided in an embodiment of this application. In this embodiment of the invention, the electronic device 900 includes a processor 901, a memory 902, a display 903, and a data acquisition device 904. Figure 9 Only some components of the electronic device 900 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0135] In some embodiments, processor 901 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 902 or process data, such as the chassis control method based on risk field construction in this invention.

[0136] In some embodiments, display 903 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 903 is used to display information from electronic device 900 and to display visual user applications. Components 901-904 of electronic device 900 communicate with each other via a system bus.

[0137] The data acquisition unit 904 is used to collect multi-source perception data of the vehicle, including inertial measurement unit data, global satellite navigation system receiver data, wheel speed sensor data, steering wheel angle sensor data, lidar data, millimeter-wave radar data, and vision camera data.

[0138] In one embodiment, when processor 901 executes a chassis control program based on a risk field constructed in memory 902, the following steps can be implemented: Collect multi-source perception data of the vehicle, and perform spatiotemporal synchronization and preprocessing on the multi-source perception data; Based on Bayesian networks, probabilistic fusion is performed on preprocessed multi-source sensing data to obtain fused sensing information. Based on the fused perception information, a Bayesian filtering method is used to estimate the vehicle's driving state and key parameters, obtaining the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. Environmental obstacle information is extracted from the fused perception information, and a comprehensive risk field is constructed based on the environmental obstacle information and the state estimate. Using the comprehensive risk field and the covariance matrix as prior information, the Monte Carlo sampling method is used to predict the probability of risk evolution and obtain the risk probability distribution. The vehicle's multiple chassis actuators are controlled collaboratively based on the aforementioned risk probability distribution.

[0139] It should be understood that when the processor 901 executes the chassis control program based on the risk field in the memory 902, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0140] Furthermore, the embodiments of the present invention do not specifically limit the type of the electronic device 900 mentioned. The electronic device 900 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 900 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

Claims

1. A chassis control method based on risk field construction, characterized in that, include: Collect multi-source perception data of the vehicle, and perform spatiotemporal synchronization and preprocessing on the multi-source perception data; Based on Bayesian networks, probabilistic fusion is performed on preprocessed multi-source sensing data to obtain fused sensing information. Based on the fused perception information, a Bayesian filtering method is used to estimate the vehicle's driving state and key parameters, obtaining the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. Environmental obstacle information is extracted from the fused perception information, and a comprehensive risk field is constructed based on the environmental obstacle information and the state estimate. Using the comprehensive risk field and the covariance matrix as prior information, the Monte Carlo sampling method is used to predict the probability of risk evolution and obtain the risk probability distribution. The vehicle's multiple chassis actuators are controlled collaboratively based on the aforementioned risk probability distribution.

2. The chassis control method based on risk field construction according to claim 1, characterized in that, The spatiotemporal synchronization and preprocessing of the multi-source sensing data includes: The global satellite navigation system timing system is used to synchronize the clocks of all sensors so that all sensor data use the same timestamp as a unified reference. The sensing data acquired by each sensor is uniformly transformed to the vehicle coordinate system through a calibrated extrinsic parameter matrix, and the original sensing data is subjected to median filtering for noise reduction, bad value removal, and outlier detection.

3. The chassis control method based on risk field construction according to claim 1, characterized in that, The probabilistic fusion of preprocessed multi-source sensing data based on Bayesian networks includes: The measured values ​​of each sensor are used as observed variables, environmental conditions are used as latent variables, and the historical accuracy statistics of the sensors are used as prior probabilities. The conditional probability tables of each sensor are updated by combining the influence factors of environmental conditions on the reliability of sensor measurements. Based on the conditional probability table, the posterior probability of each sensor measurement under the current environmental conditions is calculated using Bayes' theorem. The normalized posterior probabilities are used as the fusion weights for each sensor, and the probabilistic fusion of multi-source sensing data is completed in a weighted manner.

4. The chassis control method based on risk field construction according to claim 1, characterized in that, The construction of a comprehensive risk field based on the environmental obstacle information and the state estimate includes: A collision risk field is constructed based on the environmental obstacle information; An instability risk field is constructed based on the state estimate; The collision risk field and the instability risk field are fused to obtain the comprehensive risk field.

5. The chassis control method based on risk field construction according to claim 1, characterized in that, The method of using Monte Carlo sampling to predict the probability evolution of risk includes: Several vehicle state sampling particles are generated based on the covariance matrix, and the size of the covariance matrix is ​​positively correlated with the dispersion of the vehicle state sampling particles. Using each of the vehicle state sampling particles as the initial state, state recursion is performed based on the vehicle dynamics model within a preset prediction time domain to obtain the predicted trajectory of each vehicle state sampling particle. Based on the comprehensive risk field, the risk field value of the predicted trajectory of each vehicle state sampling particle at each time in the prediction time domain is calculated, and the risk field evolution trajectory of each vehicle state sampling particle is obtained. The risk probability distribution is obtained by probabilistically predicting the risk field evolution trajectory of all vehicle state sampled particles.

6. The chassis control method based on risk field construction according to claim 1, characterized in that, The coordinated control of multiple chassis actuators of the vehicle based on the risk probability distribution includes: The risk level of the vehicle is determined based on the risk probability distribution. The combination of active safety subsystems to be activated under the current operating condition is determined based on the risk level. The risk level and the combination of subsystems are then sent down to the middle model prediction and control layer through the upper-layer strategy scheduling layer. The middle-level model predictive control layer receives the risk level and the subsystem combination, constructs a multi-objective optimization problem with the objectives of minimizing the vehicle yaw rate tracking error, suppressing the center of gravity sideslip angle, and resisting rollover, and solves it using model predictive control. The optimal control command obtained from the solution is then sent down to the lower-level execution allocation layer through the middle-level model predictive control layer. The optimal control command is received by the lower execution allocation layer. Based on the optimal control command, a non-cooperative game theory method is used to coordinate the active front wheel steering subsystem and the direct yaw moment control subsystem. Nash equilibrium is used as the optimal coordination strategy, and the game control is driven by the risk probability distribution. The coordinated control command is then distributed to each actuator to complete the collaborative control.

7. The chassis control method based on risk field construction according to claim 1, characterized in that, The method further includes: Monitor the health status of each chassis actuator; The adjustment strategy is determined based on the described health status; Based on the aforementioned adjustment strategy, the collaborative control tasks corresponding to each chassis actuator are adjusted.

8. A chassis control device based on a risk field, characterized in that, include: The multi-source sensing data acquisition and preprocessing module is used to acquire multi-source sensing data of the vehicle and perform spatiotemporal synchronization and preprocessing on the multi-source sensing data. The multi-source sensing data fusion module is used to perform probabilistic fusion of preprocessed multi-source sensing data based on a Bayesian network and output the fused sensing information. The vehicle state Bayesian estimation module is used to estimate the vehicle's driving state and key parameters based on the fused perception information using a Bayesian filtering method, and outputs the state estimate and its corresponding covariance matrix. The covariance matrix is ​​used to characterize the uncertainty of the state estimate. The integrated risk field construction module is used to extract environmental obstacle information from the fused perception information and construct an integrated risk field based on the environmental obstacle information and the state estimate. The Monte Carlo risk propagation prediction module is used to predict the probability of risk evolution using the integrated risk field and the covariance matrix as prior information and the Monte Carlo sampling method to obtain the risk probability distribution. The chassis multi-subsystem collaborative control module is used to collaboratively control multiple chassis actuators of the vehicle based on the risk probability distribution.

9. An electronic device, characterized in that, It includes a data acquisition unit, a memory, and a processor, wherein the data acquisition unit is used to acquire multi-source perception data of the vehicle; The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the chassis control method based on risk field construction as described in any one of claims 1 to 7.

10. A chassis control system based on a risk field, characterized in that, The chassis control system based on the risk field includes the electronic device and the result output device as described in claim 9, wherein the result output device is connected to the electronic device and is used to output and execute chassis cooperative control.

Images