Model predictive control based vehicle collision detection and avoidance method, device and medium

Abstract

The application relates to the field of intelligent traffic and vehicle active safety technology, and discloses a vehicle collision detection and avoidance method and device based on model predictive control and a medium, which is aimed at the scene that a vehicle is chased into by a rear vehicle or is sandwiched by front and rear vehicles in an expressway or a high-density traffic environment, and the avoidance path is optimized online by kinematics modeling, Kalman filtering trajectory prediction and a model predictive control algorithm. The system integrates path planning and tracking control, solves the problems of path rigidity and low real-time performance of traditional collision avoidance methods in a dynamic environment, and significantly improves the safety, comfort and decision-making efficiency of a vehicle in a complex traffic environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and active vehicle safety technology, specifically to a model-based predictive control method, device, and medium for vehicle collision detection and avoidance. Background Technology

[0002] Existing technical solutions mainly rely on a combination of sensor perception, trajectory prediction, and path planning, but several key bottlenecks still need to be addressed: 1. Path planning methods based on geometric rules have poor adaptability. Early research and some existing systems used pre-defined geometric curves (such as circular arcs, polynomial curves, or S-curves) as collision avoidance paths. While these methods are computationally simple, the fixed path shape lacks adaptability to dynamic environments. Especially when dealing with vehicles in front and behind experiencing sudden changes in motion at high speeds, the fixed geometric path cannot be adjusted smoothly in real time, easily leading to increased path tracking errors and even triggering new risks of loss of control. Essentially, it relies on an open-loop path generation approach, failing to optimize online based on execution feedback.

[0003] 2. The "planning-tracking" decoupled architecture has inherent flaws. Traditional CDAS typically employs a serial architecture: first, a complete reference path is planned, and then the underlying tracking controller (such as PID control or sliding mode control) drives the vehicle to follow that path. This separation of planning and tracking has inherent drawbacks: the planner cannot predict the dynamic performance errors of the tracking controller, and the tracking controller struggles to handle dynamically infeasible paths generated by the planner. This decoupling leads to problems such as increased system response delay and inconsistent control commands, which are particularly pronounced when avoiding high-speed dynamic obstacles.

[0004] 3. Real-time performance and security are difficult to balance in complex environments. To improve safety, some studies have introduced complex optimization algorithms (such as genetic algorithms and particle swarm optimization) for global path search. However, these algorithms have high computational loads, and their real-time performance often fails to meet the stringent requirements of high-speed driving conditions. On the other hand, while methods based on simplified models ensure real-time performance, they struggle to effectively handle vehicle dynamics constraints, road boundary constraints, and multiple obstacle constraints, thus sacrificing the system's safety boundaries.

[0005] 4. Insufficient utilization of sensor information, resulting in limited forward-looking perception of the system. Most existing systems rely on a single sensor (such as a vision camera or radar) for environmental perception, lacking an effective multi-source information fusion mechanism. In terms of trajectory prediction, simple uniform speed or uniform acceleration models are often used, failing to fully exploit the temporal characteristics of sensor data. This results in insufficient prediction accuracy and look-ahead time for the future motion state of the target vehicle, preventing the system from developing early avoidance strategies with foresight.

[0006] In summary, existing vehicle collision avoidance technologies have significant shortcomings when facing highly dynamic and uncertain real-world traffic scenarios, particularly in terms of path planning adaptability, system architecture integration, balance between computational efficiency and safety, and predictive capabilities of environmental perception. Summary of the Invention

[0007] Firstly, the purpose of this invention is to provide a model-based predictive control method for vehicle collision detection and avoidance, which solves the problems of inaccurate path planning and low collision avoidance efficiency in complex environments, and focuses on improving the reliability of trajectory prediction and path generation.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A model-based predictive control method for vehicle collision detection and avoidance includes the following steps: The current state vector of the main vehicle and the motion parameters of the target vehicle are obtained at the current moment. The current state vector includes at least the longitudinal position, lateral position, heading angle and longitudinal velocity of the main vehicle. The target vehicle includes the vehicle in front and / or the vehicle behind. The motion parameters include at least the target velocity of the target vehicle and the actual relative distance between the main vehicle and the target vehicle. The actual relative distance is compared with a preset distance threshold. When the collision risk triggering condition is met, the obtained motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, as well as the predicted position uncertainty parameter corresponding to the predicted motion trajectory. Based on the predicted motion trajectory of the target vehicle in the prediction time domain and the predicted position uncertainty parameter corresponding to the predicted motion trajectory, the updated safe distance constraint is obtained. The current state vector of the main vehicle is used as the initial state input into the preset model predictive control (MPC) optimization model. The updated safety distance constraint, the preset vehicle kinematics constraint, and the preset control quantity constraint are used as the constraint boundaries. The MPC optimization model is solved to generate the optimal control quantity that satisfies all the constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain. The first optimal front wheel steering angle and the first optimal vehicle acceleration value from the optimal control quantities in the current control cycle are applied to the actuator of the master vehicle to control the vehicle's movement and enter the next control cycle.

[0009] Furthermore, acquiring the motion parameters of the target vehicle includes: The formula for calculating motion parameters is as follows: Where c is the speed of the signal emitted by the sensor. X and Y represent the target locations of the main vehicles. For sensor coordinates, For the target speed, This refers to relative velocity.

[0010] Further, the acquired motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, and the prediction position uncertainty parameters corresponding to the predicted motion trajectory, including: Obtain the target vehicle in Historical state estimation vector at time step and the corresponding historical error covariance matrix ; Obtain the preset state transition matrix and the system noise covariance matrix Calculate the output Predicted state vector at time step ,in ; Calculation output Prediction error covariance matrix at time step ,in ,in The state transition matrix The conjugate transpose of; The predicted state vector The predicted motion trajectory data points are output, and the prediction error covariance matrix is ​​used as the output. The predicted location uncertainty parameter is output as the output.

[0011] Furthermore, by using the predicted motion trajectory of the target vehicle in the prediction time domain and the prediction position uncertainty parameters corresponding to the predicted motion trajectory, the updated safety distance constraint is obtained, including: First, a comprehensive risk index is constructed based on the relative motion state of the vehicles. : in, The collision time between the primary vehicle and the target vehicle. The relative speed between the main vehicle and the target vehicle. Main vehicle speed, To account for the uncertainty in trajectory prediction using Kalman filtering, This refers to the risk weighting coefficient. According to risk indicators Dynamically adjust the safety distance constraints in model predictive control: in, Based on the safe distance, To predict the uncertainty adjustment coefficient, As a risk adjustment model; According to risk indicators Adjust the state weights in the objective function: in, Based on the weight matrix, As a risk adjustment model; The prediction time domain length is dynamically adjusted based on risk indicators: in, Based on the prediction time domain, This is the adjustment coefficient.

[0012] Furthermore, the preset risk adjustment mode includes: Obtain a language mapping table pre-stored in the memory. The language mapping table records the correspondence between continuous motion parameter ranges, discrete risk levels, and risk adjustment modes. Use the currently obtained actual relative distance and the target speed as search indexes to input into the language mapping table for numerical range comparison, and output the corresponding discrete risk level. Based on the discrete risk level obtained from the matching, the corresponding risk adjustment mode is extracted from the language mapping table.

[0013] Furthermore, using the updated safety distance constraint, the preset vehicle kinematics constraint, and the preset control quantity constraint as constraint boundaries, the MPC optimization model is solved to generate the optimal control quantity that satisfies all the constraint boundaries, including: In response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, a finite number of random samples are performed based on the predicted motion trajectory and the predicted position uncertainty parameter to obtain the collision probability; according to the collision probability, the safety distance constraint or related penalty parameter in the MPC optimization model is adjusted and re-solved to update the optimal control quantity.

[0014] Further, in response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, a finite number of random samples are performed based on the predicted motion trajectory and the predicted position uncertainty parameter to obtain a collision probability; according to the collision probability, the safety distance constraint or related penalty parameter in the MPC optimization model is adjusted and re-solved to update the optimal control quantity, including: In response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, the Monte Carlo simulation algorithm is invoked to perform a finite number of random samplings based on the predicted motion trajectory and the predicted position uncertainty parameter, and the collision probability is calculated. The collision probability is used as an index to input a preset probability compensation mapping table for numerical comparison, and the corresponding distance compensation increment and / or the corresponding penalty amplification coefficient are extracted. The dynamic safety distance threshold in the updated safety distance constraint is added to the distance compensation increment to generate the safety distance constraint after secondary expansion; and / or, the basic penalty parameter in the objective function of the MPC optimization model is multiplied with the penalty amplification coefficient to generate the updated objective function penalty parameter. Using the safety distance constraint after the secondary expansion, the preset vehicle kinematics constraint, and the preset control quantity constraint as constraint boundaries, and / or using the MPC optimization model containing the updated objective function penalty parameters as the solution objective, the solution is re-solved to generate the optimal control quantity that satisfies the current constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain.

[0015] Furthermore, the method further includes: in response to the actual relative distance in the motion parameters being greater than a preset forward critical distance threshold and no forward target speed being detected, obtaining the historical duty cycle of the master vehicle in the previous control cycle. and the preset maximum main vehicle speed ; The historical duty cycle The target speed of the vehicle behind the target vehicle. and the maximum main vehicle speed enter Calculate and output the target duty cycle of the current control cycle. ; Based on the target duty cycle Determine the target coordinated acceleration used to keep the acceleration of the lead vehicle consistent with that of the following vehicle; The target cooperative acceleration is used as an additional longitudinal control constraint to update the MPC optimization model so that the optimal vehicle acceleration value solved and output in the current control cycle approximates the target cooperative acceleration.

[0016] Secondly, the purpose of this invention is to provide a model-based predictive control vehicle collision detection and avoidance device to solve the problems of inaccurate path planning and low collision avoidance efficiency in complex environments, with a focus on improving the reliability of trajectory prediction and path generation.

[0017] To achieve the above objectives, the present invention provides the following technical solution: Model-based predictive control vehicle collision detection and avoidance devices, applied to vehicle control equipment, including: The data acquisition and risk assessment module is used to obtain the current state vector of the main vehicle and the motion parameters of the target vehicle at the current moment. The current state vector includes at least the longitudinal position, lateral position, heading angle and longitudinal speed of the main vehicle. The target vehicle includes the vehicle in front and / or the vehicle behind. The motion parameters include at least the target speed of the target vehicle and the actual relative distance between the main vehicle and the target vehicle. The Kalman trajectory prediction module compares the actual relative distance with a preset distance threshold. When the collision risk triggering condition is met, the acquired motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, as well as the predicted position uncertainty parameters corresponding to the predicted motion trajectory. Based on the predicted motion trajectory of the target vehicle in the prediction time domain and the predicted position uncertainty parameters corresponding to the predicted motion trajectory, an updated safety distance constraint is obtained. The MPC constraint update and solution module is used to input the current state vector of the main vehicle as the initial state into the preset model predictive control (MPC) optimization model, and solve the MPC optimization model with the updated safety distance constraint, the preset vehicle kinematics constraint and the preset control quantity constraint as the constraint boundary to generate the optimal control quantity that satisfies all the constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain. The instruction execution module is used to apply the first optimal value of the front wheel steering angle and the first optimal value of the vehicle acceleration from the optimal control quantities in the current control cycle to the actuator of the master vehicle, so as to control the vehicle to drive and enter the next control cycle.

[0018] Thirdly, the purpose of this invention is to provide a vehicle control device that solves the problems of inaccurate path planning and low collision avoidance efficiency in complex environments, with a focus on improving the reliability of trajectory prediction and path generation.

[0019] To achieve the above objectives, the present invention provides the following technical solution: A vehicle control device includes: a memory for storing a computer program; and a processor communicatively connected to the memory for executing the computer program stored in the memory to implement the steps of the method of the first aspect.

[0020] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: This invention deeply couples the predicted position uncertainty parameter of the Kalman filter output to the underlying constraints of the Model Predictive Control (MPC) optimization model. It can adaptively expand the safety distance constraint boundary when the target vehicle's maneuvering intention is unclear or when sensor noise surges. When the target vehicle's motion state is stable and the prediction confidence is high, the safety boundary is dynamically contracted. This overcomes the shortcomings of traditional fixed thresholds, ensuring both the safety of extreme avoidance and traffic efficiency, and endowing the main vehicle with extremely high control robustness and dynamic adaptability.

[0021] In the face of potential front-and-rear vehicle collisions in complex traffic environments, this invention does not employ a consistent, universal intervention logic. Instead, it configures independent risk adjustment modes for the front and rear target vehicles. This asymmetric algebraic processing mechanism enables the Model Predictive Control (MPC) optimizer to accurately calculate the control command sequence with the global optimal solution in the conflict between avoiding the risk of emergency braking by the vehicle in front and mitigating the risk of rear-end collisions by the vehicle behind. This significantly enhances the obstacle avoidance robustness of the main vehicle in dense traffic interaction scenarios.

[0022] This invention designs a risk assessment and correction based on Monte Carlo simulation. In response to the prediction error exceeding a preset threshold, it quantifies the collision probability through on-demand random sampling logic and performs online secondary defensive reinforcement of safety distance constraints or adaptive gain adjustment of the objective function penalty parameters according to a preset probability compensation mapping table. While maintaining the structural stability of the main control MPC algorithm, this mechanism provides algorithm-level safety redundancy for the safe driving of autonomous driving systems in highly unpredictable environments through defensive compensation of underlying values.

[0023] This invention transforms underlying continuous motion state parameters (such as the algebraic relationship between actual relative distance, target speed, and preset thresholds) directly into discretized risk levels by pre-constructing a state machine decision logic and language mapping model. This lookup table matching mechanism avoids high-frequency redundant nonlinear rule derivation, significantly reducing the computational load on the underlying processor. Based on this mechanism, the system can perform millisecond-level seamless switching between various control modes such as cruise, warning, forced acceleration, and constrained curvilinear motion (CCM) with extremely low instruction transmission latency. This achieves efficient instruction flow from the physical perception layer to the chassis drive-by-wire execution layer, ensuring the continuity of vehicle dynamic control and driving smoothness. Attached Figure Description

[0024] Figure 1 This is the execution process of this application; Figure 2 This is the language table of MPC-MCDAS; Figure 3 This is a vehicle motion model according to an embodiment of the present invention; Figure 4 This is the vehicle phasor representation according to an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings: Figure 1 This is the execution flow of a Model Predictive Control (MPC) based vehicle collision detection system (MPC-MCDAS). Its basic idea is to establish a vehicle motion model, use sensors (ESR) to obtain motion parameters, determine the conditions of the preceding vehicle, execute different adaptive control methods, focus on generating avoidance paths through Kalman filtering and model predictive control, and finally verify the path safety through Monte Carlo simulation.

[0026] The first step is to establish a vehicle motion model: define the speed of the main vehicle. The distance between the following vehicle and the main vehicle The predefined distance between the following vehicle and the main vehicle The speed of the following vehicle The distance between the vehicle in front and the main vehicle The speed of the vehicle in front The predefined distance between the preceding vehicle and the master vehicle and A language table is defined. This language table is used to map continuous motion parameters to discrete risk levels for subsequent control strategy selection.

[0027] The system retrieves a language mapping table pre-stored in memory, which records the correspondence between continuous motion parameter ranges, discrete risk levels, and risk adjustment modes. The system inputs the currently acquired actual relative distance and the target speed as search indexes into the language mapping table for numerical range comparison and outputs the corresponding discrete risk level. Based on the matched discrete risk level, the system extracts the corresponding risk adjustment mode from the language mapping table.

[0028] For visualization purposes, the discrete risk levels in the application can be divided into extremely low risk level, rear-end collision risk level, critical early warning risk level, high-risk pincer attack risk level, and unknown risk level of perceived failure. Risk adjustment modes can be divided into non-acceleration mode, acceleration mode, warning mode, CCM mode, and fault-tolerant mode.

[0029] Figure 2 The language table content of MPC-MCDAS is shown. Specifically, in a further embodiment of the present invention, the system realizes risk assessment and control mode switching in multiple scenarios by retrieving a pre-configured language mapping table. The language mapping table is pre-stored in the memory of the vehicle control device, and it records a strict correspondence between the range of continuous motion parameters and discrete risk levels (i.e., the corresponding control modes).

[0030] In this embodiment, the continuous motion parameters (i.e., the retrieval index) used for input comparison include: the presence status of the vehicle ahead. The actual relative distance between the main vehicle and the vehicle in front target speed of the vehicle in front The actual relative distance between the main vehicle and the vehicles behind it and the target speed of the following vehicle. .

[0031] Meanwhile, the boundary constants in the language mapping table that divide the range of continuous motion parameters are preset to: a long-distance safety threshold ahead. Short-distance safety threshold ahead Rear warning distance threshold and the current longitudinal speed of the main vehicle .

[0032] The system uses the currently acquired actual relative distance and target speed, among other parameters, as search indices, inputs them into the language mapping table for numerical range comparison, and outputs the corresponding discrete risk level and control mode. The specific matching and mapping rules are as follows: Extremely low risk level (no acceleration mode): The search index is compared with the language mapping table. When a match is found that the continuous motion parameter range meets the following condition: no preceding vehicle was detected ( (None), forward actual relative distance And the actual relative distance backward When the mapping outputs an extremely low risk level, the system extracts the corresponding control strategy and enters a non-acceleration mode (maintaining the current cruise state).

[0033] Rear-end collision risk level (acceleration mode): The search index is compared with the language mapping table. When a match is found that the continuous motion parameter range meets the following criteria: No vehicle in front is detected ( (None), forward actual relative distance However, the actual relative distance backward And the target speed of the following vehicle At that time, the system maps and outputs the risk level of rear-end collisions, and extracts the corresponding control strategy accordingly, entering acceleration mode to avoid following vehicles.

[0034] Critical early warning risk level (warning mode): The search index is compared with the language mapping table. When a match is found that the continuous motion parameter range meets the following criteria: a vehicle ahead has been detected ( (If present), forward actual relative distance And the target speed of the vehicle in front Meanwhile, the actual relative distance backward And the target speed of the following vehicle When the critical risk level is mapped out, the system enters warning mode accordingly.

[0035] High-risk pincer attack level (CCM mode): The search index is compared with the language mapping table. When a match is found that the continuous motion parameter range meets the following conditions: a vehicle in front is detected ( (If present), forward actual relative distance Between and Between and Meanwhile, the actual relative distance backward and At that time, the system outputs a high-risk pincer attack level. Based on this discrete risk level, the system extracts the corresponding risk adjustment mode for initiating constrained curve motion, forces entry into CCM (Constrained Curve Motion) mode, and initiates MPC-based rolling optimization collision avoidance control.

[0036] Perceived failure unknown risk level (fault tolerance mode): When all search indexes match the "unknown" range, the system maps and outputs the discrete risk level with the highest prevention level based on the safety redundancy mechanism, and triggers the CCM (Constrained Curve Motion) mode by default.

[0037] After matching and obtaining the corresponding discrete risk levels and patterns, the system further extracts the risk adjustment pattern corresponding to the current working condition from the language mapping table, so as to be called when the MPC optimization model calculates the dynamic safety distance threshold.

[0038] To obtain the aforementioned high-precision motion parameters, this embodiment of the invention relies on environmental perception hardware (such as an onboard electronically scanned radar) deployed on the main vehicle to perform underlying data acquisition and signal analysis steps. Specifically, the vehicle control device continuously acquires the radio frequency echo signals transmitted and received by the onboard radar through a communication interface, and the built-in signal processing component performs physical layer analysis on the echo signals to extract the signal delay time characterizing the spatial position. Target azimuth and the Doppler frequency characterizing relative motion. .

[0039] After acquiring the aforementioned underlying signal characteristics, the system first determines the signal delay time. Calculate the radial distance of the target It satisfies the algebraic relation: in The preset signal transmission speed of the sensor.

[0040] Subsequently, the system will determine the target radial distance. With the target azimuth angle A mapping transformation from polar coordinates to a Cartesian two-dimensional coordinate system is performed to obtain the target vehicle's position coordinates in the current environmental coordinate system. The transformation relationship satisfies: as well as After completing the spatial mapping of the coordinate system, the system uses a Euclidean distance metric model to calculate the association parameters between the master vehicle and the target vehicle in a multi-vehicle interaction scenario. Specifically, the data processing unit will process the calculated target position coordinates... With the pre-calibrated known sensor origin coordinates Substituting into the distance equation, the actual relative distance between the main vehicle and the target vehicle is accurately calculated and output. The Euclidean distance calculation formula is specifically expressed as follows: Where c is the speed of the signal emitted by the sensor. X and Y represent the target locations of the main vehicles. For sensor coordinates, For the target speed, This refers to relative velocity.

[0041] Through the above calculations, the system can output the actual relative distance with high confidence in real time and continuously. relative to target speed The series of motion parameters output by the above calculations serve a dual technical function in the system control logic: Firstly, the actual relative distance is first used for high-frequency comparison with a preset distance threshold to serve as a rigid criterion for whether to trigger the subsequent collision risk avoidance strategy; secondly, after confirming that the collision risk triggering condition is met, these motion parameters are also used as full real-time observation data, synchronously fed into a pre-configured Kalman filter model for iterative prediction of the target trajectory and error, and finally used as the basic input variables for state feedback, flowing to the Model Predictive Control (MPC) optimization model for online updating of boundary constraints.

[0042] The actual relative distance is used to determine whether the collision risk triggering conditions are met, and serves as an input parameter for trajectory prediction and model prediction control.

[0043] In the absence of vehicles ahead, the system switches to a rear-vehicle cooperative adaptive control mode, which maintains the acceleration of the lead vehicle consistent with that of the rear vehicle. The specific calculation formula is as follows: in, For the maximum speed of the main vehicle, This represents the duty cycle.

[0044] Figure 3 The vehicle motion model of an embodiment of the present invention is shown, such as Figure 3 As shown in the attached figure, this illustrative diagram illustrates a typical multi-vehicle motion scenario when vehicles are traveling in the same direction on a structured road. Along the direction of vehicle movement, three key vehicle entities are defined: the main vehicle implementing the collision detection and avoidance method described in this invention, the target vehicle located in front of the main vehicle in the same lane (i.e., the vehicle in front), and the target vehicle located behind the main vehicle in the same lane (i.e., the vehicle behind).

[0045] During the movement of the main vehicle, the vehicle control equipment collects motion parameters of the main vehicle and surrounding target vehicles in real time through onboard sensors. Specifically, the current longitudinal speed of the main vehicle is defined as... The target speed of the vehicle ahead is defined as... The target speed of the vehicle behind is defined as... Meanwhile, in terms of measuring physical spatial distance, the actual relative distance between the front end of the main vehicle and the rear end of the vehicle in front is defined as... The actual relative distance between the rear end of the main vehicle and the front end of the rear vehicle is defined as... The aforementioned speed and distance parameters form the basis of the input data for subsequent risk assessment and system state machine switching.

[0046] To quantify and classify longitudinal collision risks of varying degrees, this embodiment pre-defines multiple distance-based static or dynamic threshold boundaries in the physical space surrounding the host vehicle. Specifically, this includes a forward long-distance safety threshold set in front of the host vehicle. This threshold is used to define whether the vehicle ahead is outside the far boundary of the safe cruising range, when the actual relative distance... Greater than the aforementioned long-distance safety threshold When the system determines that there is no imminent threat from following vehicles ahead, it also includes an additional boundary set in front of the main vehicle and whose value is less than the long-distance safety threshold ahead, i.e., the short-distance safety threshold ahead. This threshold is used to define the near-end high-risk boundary that requires triggering emergency collision avoidance; the forward long-distance safety threshold... Safety threshold for short distance ahead The space between them constitutes a sensitive transition zone for forward risks. Furthermore, a rear warning distance threshold is set behind the main vehicle. This threshold is used to define the warning boundary for the risk of a rear-end collision. When the rear vehicle intrudes into this warning boundary, i.e., the actual relative distance... Less than the rear warning distance threshold When there is a close relative speed difference, the system will trigger a backward high-risk warning or a forced acceleration avoidance action.

[0047] Furthermore, the aforementioned motion parameters and distance threshold boundaries defined by physical space collectively constitute the basis for the continuous motion parameter input of the language mapping table in this embodiment of the invention. In the actual control cycle, the vehicle control device will collect the actual relative distance in real time. , The system compares the speed parameters of each vehicle with the aforementioned distance threshold boundaries, and uses the parameter ranges and logical judgment results obtained from the comparison as a retrieval index, directly inputting them into a pre-stored language mapping table for lookup matching. Through this association mapping mechanism, the system can accurately convert continuously changing physical space parameters into discrete risk levels corresponding to the current operating condition, and then extract the corresponding risk adjustment mode from the language mapping table. After obtaining the discrete risk level and risk adjustment mode, the system activates the corresponding constraint curve motion mode or conventional acceleration / deceleration control mode accordingly, and uses the risk adjustment mode to generate or update the lower limit of the safe distance constraint in the model predictive control optimization model online. Finally, the system generates a control command sequence through a solution algorithm to achieve accurate and expected adaptive collision avoidance control in complex multi-vehicle interaction conditions.

[0048] Figure 4 This demonstrates the phasor representation of the vehicle, facilitating the description of its position. To further clarify the specific analytical process of the target vehicle's motion parameters in the underlying algorithm of this invention embodiment, combined with... Figure 4 The velocity vector decomposition logic of the target vehicle is explained in detail. The accompanying figure exemplifies the vector mapping relationship of the target vehicle's actual speed in a two-dimensional orthogonal reference coordinate system. After the onboard sensors acquire the motion feedback signal of the target vehicle, the vehicle control equipment decouples the velocity parameters in a preset two-dimensional environmental coordinate system. For example... Figure 4 As shown, the horizontal coordinate axis is defined to represent the direction of the first velocity (e.g., the lateral velocity direction). (represented), defining the vertical coordinate axis to characterize the direction of the second velocity (e.g., the longitudinal velocity direction, in...) (Representation). The system obtains a synthetic target velocity vector characterizing the overall motion trend of the target vehicle through the sensor's data preprocessing module. And simultaneously calculate the composite target velocity vector. The azimuth angle formed between the horizontal coordinate axis and the horizontal coordinate axis The included angle of this azimuth Essentially, it reflects the current yaw trend of the target vehicle.

[0049] Based on the aforementioned collected fundamental physical quantities, the data processing module built into the vehicle control device utilizes processor resources to perform trigonometric function mapping operations, thereby synthesizing the target velocity vector. Independent projections are performed along orthogonal coordinate axes. Specifically, the system calculates the first velocity component of the target vehicle on the horizontal coordinate axis based on the cosine transform law. Their algebraic relations satisfy Similarly, the system calculates the second velocity component of the target vehicle on the vertical coordinate axis based on the sine transform law. Their algebraic relations satisfy Through this vector orthogonal decomposition operation, the system losslessly transforms the velocity parameters, originally represented in polar coordinates, into two-dimensional Cartesian coordinate components suitable for computer matrix operations. The first velocity component output by the above calculation... Second velocity component This data will be used as a decoupled, independent state variable and directly input into the pre-configured Kalman filter model to participate in the recursive iteration of the state equation. This dimensionality reduction and decoupling of the underlying data can eliminate the prediction error caused by the coupling of lateral and longitudinal motion, thereby greatly improving the calculation accuracy of the target vehicle's two-dimensional planar motion trajectory prediction in the prediction time domain.

[0050] The second step involves Kalman filtering, which estimates and updates the position and velocity states of the target vehicle at discrete moments by establishing the state equation and measurement equation of the target vehicle.

[0051] Specifically, by establishing the state equation and measurement equation of the target vehicle, the position and velocity state of the vehicle at discrete moments are estimated and updated. This module synchronously estimates and predicts the two-dimensional position and velocity state of the target vehicle at discrete moments by establishing an 8-dimensional joint state equation and measurement equation covering the vehicles in front and behind.

[0052] The state equation describes the evolution of the target vehicle's motion state between adjacent sampling times, and the measurement equation fuses real-time high-frequency observation data acquired by onboard environmental perception hardware (such as electronically scanned radar) to correct prediction errors based on pure kinematics deduction. Specifically, First, define an 8-dimensional state vector: in: This refers to the position of the vehicle ahead in the global coordinate system. These are the velocity components of the vehicle ahead in the x and y directions. It is the position of the vehicle behind in the global coordinate system. These are the velocity components of the vehicle behind in the x and y directions.

[0053] Under discrete-time conditions, the state transition equation is: Wherein, the state transition matrix for: in, The sampling time interval, This is the process noise vector, used to describe the deviation between the vehicle motion model and the actual motion, representing the model uncertainty.

[0054] The measurement equation is expressed as: Among them, the measurement matrix for: The corresponding measurement vector is: in, For sensor observation vectors, For measuring noise, it is used to describe the sensor measurement error.

[0055] In the prediction phase of Kalman filtering, i.e., the time update phase, the system needs to determine the target vehicle's position coordinates and corresponding error variance within the prediction time domain. The underlying processor is based on the previous time step. Based on the historical state estimation information and its corresponding historical error covariance, the following matrix recursive prediction calculus is performed: in, Indicates at time Given information about time State prediction value, Indicates time The optimal state estimate, This represents the state prediction error covariance matrix. This represents the state estimation covariance matrix at the previous time step. The process noise covariance matrix is ​​used to characterize the uncertainty of the system model. Let it be its conjugate transpose; Through the recursive prediction process described above, the system completes matrix iteration of the multi-objective state space. Specifically, the prior state vector output by the above formula is... This can then be used as the predicted trajectory of the vehicle in front and the vehicle behind in the prediction time domain; and the calculated covariance matrix... This quantifies the degree of divergence of position coordinates at the physical level, at which point the covariance matrix... This can be used as a parameter for predicting the uncertainty of the collision risk in the system, i.e., the error range mentioned above in this embodiment. Ultimately, the predicted motion trajectory and its corresponding error range obtained by the system will directly cross the perception prediction layer and serve as a hard constraint and dynamic safety distance assessment basis in the subsequent model predictive control optimization problem, thereby fundamentally improving the system's forward-looking judgment ability and extreme avoidance robustness against potential front and rear vehicle sandwiching and nonlinear collision risks.

[0056] The third step is to comprehensively consider the kinematic constraints of the main vehicle itself and the predicted motion states of the vehicles in front and behind within the prediction time domain to generate a collision avoidance control strategy that meets safety constraints.

[0057] In this embodiment, the current state vector of the master vehicle is selected and constructed as follows: in, These represent the time of the main vehicle. The longitudinal and lateral positions, Indicates the vehicle's heading angle. This represents the vehicle's longitudinal velocity; the control vector is selected as follows: in, This refers to the front wheel steering angle. To accelerate the vehicle.

[0058] Based on the current state of the main vehicle and combined with the vehicle kinematics model, the state evolution of the main vehicle in the prediction time domain is predicted. At the same time, using the predicted trajectories of the vehicles in front and behind, output by the Kalman filter module in the second step, their position distribution information in the prediction time domain is obtained.

[0059] The fourth step is to introduce an adaptive parameter adjustment mechanism based on traffic risk assessment: First, a comprehensive risk index is constructed based on the relative motion state of the vehicles. : in, The collision time between the primary vehicle and the target vehicle. The relative speed between the main vehicle and the target vehicle. Main vehicle speed, To account for the uncertainty in trajectory prediction using Kalman filtering, Risk weighting coefficient; comprehensive risk index Used to characterize the level of potential collision risk in the current traffic environment.

[0060] The fifth step is to introduce an adaptive adjustment strategy, specifically: (1) Dynamic adjustment of safety distance According to risk indicators Safety distance in model predictive control Constraints are dynamically adjusted: in, Based on the safe distance, To predict the uncertainty adjustment coefficient, This is a risk adjustment mode; that is, when the system checks the risk indicators and they increase, the system automatically expands the safety distance constraint due to the above constraints.

[0061] in, The method for determining it is as follows: in, Main vehicle speed, For safety time (usually not less than 2 seconds).

[0062] Typically on city roads and highways : (2) Dynamic adjustment of objective function weights According to risk indicators Adjust the state weights in the MPC objective function: in, Based on the weight matrix, This is a risk adjustment model.

[0063] When the risk level increases, the system increases the obstacle avoidance weights, so that the control strategy prioritizes safety.

[0064] (3) Predictive time-domain adaptive adjustment The prediction time domain length is dynamically adjusted based on risk indicators: in, Based on the prediction time domain, This is the adjustment coefficient.

[0065] By extending the prediction time domain, the system can plan smoother collision avoidance paths in advance.

[0066] The sixth step, in the model predictive control process, is to comprehensively consider the above-mentioned safety distance constraints, vehicle kinematic constraints and control quantity constraints in the prediction time domain, so that the main vehicle can prioritize avoiding potential collision risks with vehicles in front or behind while satisfying driving stability.

[0067] Within each control cycle, the corresponding control problem is solved based on the current state of the master vehicle and the predicted information of the vehicles in front and behind. The first control variable in the obtained optimal control sequence is applied to the master vehicle, and then the next control cycle is entered. The above prediction and control process is repeated to achieve rolling optimization control.

[0068] In the model predictive control process, within each control cycle, the MPC optimization model is solved by comprehensively considering the aforementioned safety distance constraints, vehicle kinematic constraints, and control quantity constraints as a global boundary combination in the prediction time domain, and the first optimal front wheel steering angle is extracted. And the first vehicle acceleration It is applied to the actuator to achieve adaptive rolling optimized collision avoidance control.

[0069] The seventh step involves introducing an auxiliary risk assessment and adaptive control command correction mechanism based on the Monte Carlo simulation algorithm, in addition to the basic main control logic that generates optimal control commands using the model predictive control.

[0070] Specifically, the vehicle control device has a pre-set uncertainty threshold to monitor and evaluate the predicted position uncertainty parameter (i.e., error range) output by the Kalman filter model in real time. Under normal low-dynamic cruise conditions, when the error range is detected to be lower than or equal to the preset uncertainty threshold, the system determines that the confidence level of the predicted trajectory output by the current Kalman filter meets the direct control requirements, and therefore does not activate the Monte Carlo simulation assistance strategy. This on-demand triggering design not only avoids invalid random sampling calculations and significantly reduces the computational load on the underlying processor, but also ensures high real-time performance of conventional collision avoidance system control commands.

[0071] Example 2: Based on the same inventive concept as the aforementioned multi-vehicle collision detection and avoidance methods, this embodiment of the invention also provides a vehicle control device for implementing the above-mentioned control algorithm at the physical entity level. In practical automotive engineering applications, the vehicle control device can be physically implemented as a central domain controller of an advanced driver assistance system (ADAS), an on-board computing platform, or a highly integrated electronic control unit (ECU).

[0072] In terms of the underlying hardware topology, the vehicle control device mainly includes a processor and a memory interconnected via an internal system bus. The memory, serving as a non-volatile or volatile resident carrier of data and instructions, can be in the form of, but is not limited to, read-only memory (ROM), random access memory (RAM), flash memory, or other high-throughput solid-state storage media. This memory not only has a dedicated program storage area for storing computer program instructions containing all the underlying mathematical derivation logic, including Kalman filter prediction, model predictive control optimization, and Monte Carlo auxiliary evaluation, but also a data cache area for frequently reading and writing the pre-configured language mapping table, probability compensation mapping table, and storing massive amounts of dynamic calibration parameters such as historical state estimation information and error covariance matrix from the previous control cycle.

[0073] On the other hand, the processor, as the core computing and scheduling hub of the vehicle control device, can specifically adopt a general-purpose central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or an onboard artificial intelligence chip (NPU) optimized for tensor operations. The processor is configured to call and execute the computer program stored in the memory in real time via a high-bandwidth instruction bus to instantiate and connect all the control flows and data flows detailed in the foregoing embodiments. Specifically, by executing the program code, the processor can efficiently complete physical layer analysis of the underlying radio frequency signals of the onboard electronically scanned radar, high-order matrix multiplication and addition iterations for the 8-dimensional joint state space model, rolling solution of quadratic programming (QP) with hard boundary constraints for model predictive control, and massive random sampling operations triggered under extreme conditions within a millisecond-level control cycle.

[0074] Furthermore, to achieve closed-loop linkage with the physical actuators of the main vehicle, the vehicle control equipment also integrates a standardized on-board communication interface controller (such as a CAN bus transceiver or an on-board Ethernet interface). After the processor generates a control command sequence containing the optimal front wheel steering angle and the optimal vehicle acceleration, it sends the first optimal control quantity to the chassis steer-by-wire system and brake-by-wire / drive system of the main vehicle in the form of digital electrical signals through the on-board communication interface. This drives the vehicle to perform precise collision avoidance and trajectory tracking actions in real physical space, ultimately realizing the adaptive safety collision avoidance control steps described in the claims.

Claims

1. A model-based predictive control method for vehicle collision detection and avoidance, characterized in that, Includes the following steps: The current state vector of the main vehicle and the motion parameters of the target vehicle are obtained at the current moment. The current state vector includes at least the longitudinal position, lateral position, heading angle and longitudinal velocity of the main vehicle. The target vehicle includes the vehicle in front and / or the vehicle behind. The motion parameters include at least the target velocity of the target vehicle and the actual relative distance between the main vehicle and the target vehicle. The actual relative distance is compared with a preset distance threshold. When the collision risk triggering condition is met, the obtained motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, as well as the predicted position uncertainty parameter corresponding to the predicted motion trajectory. Based on the predicted motion trajectory of the target vehicle in the prediction time domain and the predicted position uncertainty parameter corresponding to the predicted motion trajectory, the updated safe distance constraint is obtained. The current state vector of the main vehicle is used as the initial state input into the preset model predictive control (MPC) optimization model. The updated safety distance constraint, the preset vehicle kinematics constraint, and the preset control quantity constraint are used as the constraint boundaries. The MPC optimization model is solved to generate the optimal control quantity that satisfies all the constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain. The first optimal front wheel steering angle and the first optimal vehicle acceleration value from the optimal control quantities in the current control cycle are applied to the actuator of the master vehicle to control the vehicle's movement and enter the next control cycle.

2. The method according to claim 1, characterized in that, The acquisition of the motion parameters of the target vehicle includes: The formula for calculating motion parameters is as follows: in, Let be the radial distance to the target, and c be the velocity of the signal emitted by the sensor. X and Y represent the target locations of the main vehicles. For sensor coordinates, For the target speed, This refers to relative velocity.

3. The method according to claim 1, characterized in that, The acquired motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, and the prediction position uncertainty parameters corresponding to the predicted motion trajectory, including: Obtain the target vehicle in Historical state estimation vector at time step and the corresponding historical error covariance matrix ; Obtain the preset state transition matrix and the system noise covariance matrix Calculate the output Predicted state vector at time step ,in ; Calculation output Prediction error covariance matrix at time step ,in ,in The state transition matrix The conjugate transpose of; The predicted state vector The predicted motion trajectory data points are output, and the prediction error covariance matrix is ​​used as the output. The predicted location uncertainty parameter is output as the output.

4. The method according to claim 1, characterized in that, The updated safety distance constraints are obtained by using the predicted motion trajectory of the target vehicle in the prediction time domain and the prediction position uncertainty parameters corresponding to the predicted motion trajectory, including: First, a comprehensive risk index is constructed based on the relative motion state of the vehicles. : in, The collision time between the primary vehicle and the target vehicle. The relative speed between the main vehicle and the target vehicle. Main vehicle speed, To account for the uncertainty in trajectory prediction using Kalman filtering, This refers to the risk weighting coefficient. According to risk indicators Dynamically adjust the safety distance constraints in model predictive control: in, Based on the safe distance, To predict the uncertainty adjustment coefficient, As a risk adjustment model; According to risk indicators Adjust the state weights in the objective function: in, Based on the weight matrix, As a risk adjustment model; The prediction time domain length is dynamically adjusted based on risk indicators: in, Based on the prediction time domain, This is the adjustment coefficient.

5. The method according to claim 4, characterized in that, The preset risk adjustment mode includes: Obtain a language mapping table pre-stored in the memory. The language mapping table records the correspondence between continuous motion parameter ranges, discrete risk levels, and risk adjustment modes. Use the currently obtained actual relative distance and the target speed as search indexes to input into the language mapping table for numerical range comparison, and output the corresponding discrete risk level. Based on the discrete risk level obtained from the matching, the corresponding risk adjustment mode is extracted from the language mapping table.

6. The method according to claim 1, characterized in that, Using the updated safety distance constraint, preset vehicle kinematics constraints, and preset control quantity constraints as constraint boundaries, the MPC optimization model is solved to generate the optimal control quantity that satisfies all the constraint boundaries, including: In response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, a finite number of random samples are performed based on the predicted motion trajectory and the predicted position uncertainty parameter to obtain the collision probability; according to the collision probability, the safety distance constraint or related penalty parameter in the MPC optimization model is adjusted and re-solved to update the optimal control quantity.

7. The method according to claim 6, characterized in that, In response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, a finite number of random samples are performed based on the predicted motion trajectory and the predicted position uncertainty parameter to obtain the collision probability; Based on the collision probability, the safety distance constraint or related penalty parameters in the MPC optimization model are adjusted and re-solved to update the optimal control quantity, including: In response to the predicted position uncertainty parameter exceeding a preset uncertainty threshold, the Monte Carlo simulation algorithm is invoked to perform a finite number of random samplings based on the predicted motion trajectory and the predicted position uncertainty parameter, and the collision probability is calculated. The collision probability is used as an index to input a preset probability compensation mapping table for numerical comparison, and the corresponding distance compensation increment and / or the corresponding penalty amplification coefficient are extracted. The dynamic safety distance threshold in the updated safety distance constraint is added to the distance compensation increment to generate the safety distance constraint after secondary expansion; and / or, the basic penalty parameter in the objective function of the MPC optimization model is multiplied with the penalty amplification coefficient to generate the updated objective function penalty parameter. Using the safety distance constraint after the secondary expansion, the preset vehicle kinematics constraint, and the preset control quantity constraint as constraint boundaries, and / or using the MPC optimization model containing the updated objective function penalty parameters as the solution objective, the solution is re-solved to generate the optimal control quantity that satisfies the current constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain.

8. The method according to claim 1, characterized in that, The method further includes: in response to the actual relative distance in the motion parameters being greater than a preset forward critical distance threshold and no forward target speed being detected, obtaining the historical duty cycle of the master vehicle in the previous control cycle. and the preset maximum main vehicle speed ; The historical duty cycle The target speed of the vehicle behind the target vehicle. and the maximum main vehicle speed enter Calculate and output the target duty cycle of the current control cycle. ; Based on the target duty cycle Determine the target coordinated acceleration used to keep the acceleration of the lead vehicle consistent with that of the following vehicle; The target cooperative acceleration is used as an additional longitudinal control constraint to update the MPC optimization model so that the optimal vehicle acceleration value solved and output in the current control cycle approximates the target cooperative acceleration.

9. A model-based predictive control vehicle collision detection and avoidance device, applied to vehicle control equipment, characterized in that, include: The data acquisition and risk assessment module is used to obtain the current state vector of the main vehicle and the motion parameters of the target vehicle at the current moment. The current state vector includes at least the longitudinal position, lateral position, heading angle and longitudinal speed of the main vehicle. The target vehicle includes the vehicle in front and / or the vehicle behind. The motion parameters include at least the target speed of the target vehicle and the actual relative distance between the main vehicle and the target vehicle. The Kalman trajectory prediction module compares the actual relative distance with a preset distance threshold. When the collision risk triggering condition is met, the acquired motion parameters are input into a pre-configured Kalman filter model for recursive prediction to obtain the predicted motion trajectory of the target vehicle in the prediction time domain, as well as the predicted position uncertainty parameters corresponding to the predicted motion trajectory. Based on the predicted motion trajectory of the target vehicle in the prediction time domain and the predicted position uncertainty parameters corresponding to the predicted motion trajectory, an updated safety distance constraint is obtained. The MPC constraint update and solution module is used to input the current state vector of the main vehicle as the initial state into the preset model predictive control (MPC) optimization model, and solve the MPC optimization model with the updated safety distance constraint, the preset vehicle kinematics constraint and the preset control quantity constraint as the constraint boundary to generate the optimal control quantity that satisfies all the constraint boundaries. The optimal control quantity includes the optimal value of the front wheel steering angle and the optimal value of the vehicle acceleration of the main vehicle in the prediction time domain. The instruction execution module is used to apply the first optimal value of the front wheel steering angle and the first optimal value of the vehicle acceleration from the optimal control quantities in the current control cycle to the actuator of the master vehicle, so as to control the vehicle to drive and enter the next control cycle.

10. A vehicle control device, characterized in that, include: Memory, used to store computer programs; A processor, communicatively connected to the memory, is configured to execute a computer program stored in the memory to implement the steps of the method as described in any one of claims 1 to 8.

Images