An autonomous vehicle dynamic environment modeling and active adaptation system

Abstract

The application relates to automatic driving, in particular to a dynamic environment modeling and active adaptation system for an automatic driving vehicle, a multi-frequency band cognitive radar module, a millimeter wave radar integrating multiple frequency bands, an intelligent switching and cooperative working mechanism, automatic optimization of a frequency band working cycle proportion according to an environment condition, emission and reception of radar waves of different frequency bands, acquisition of target multidimensional information, a data processing and dynamic environment modeling module, a multisensor data fusion algorithm for fusing target multidimensional information acquired by different frequency band radars, dynamic environment modeling around the vehicle, and real-time updating of a dynamic environment model, a decision and active adaptation control module, a deep learning and artificial intelligence algorithm for formulating reasonable driving decisions and planning paths and performing active adaptation control, and the technical scheme can effectively overcome the defects that it is difficult to construct an accurate and real-time dynamic environment model and a vehicle driving state cannot be timely and reasonably adjusted.

Description

TECHNICAL FIELD

[0001] The present application relates to automatic driving, in particular to a dynamic environment modeling and active adaptation system for automatic driving vehicles. BACKGROUND

[0002] In the field of automatic driving, with the rapid development of intelligent transportation systems and the increasing demand for travel safety and convenience, achieving autonomous and safe driving of vehicles in complex environments has become a key goal of industry development. High-precision environment perception and dynamic environment modeling technology, as the core support of automatic driving systems, can provide real-time and accurate environmental information for vehicles, helping vehicles make reasonable decisions in a timely manner, and has crucial significance for promoting the popularization and application of automatic driving technology.

[0003] Traditional automatic driving vehicle environment perception schemes mainly rely on single sensors such as laser radar, cameras, and single-band millimeter wave radar, or simple combinations. Single-band millimeter wave radar has limitations in detection distance, resolution, and anti-interference ability, making it difficult to comprehensively and accurately perceive dynamic targets in various complex environments. Although laser radar can obtain high-precision three-dimensional point cloud data, it is costly, limiting its large-scale application. Cameras are extremely sensitive to environmental light and weather conditions, and in harsh weather (such as rain, fog, snow, etc.) or night scenes, image quality decreases significantly, making target recognition difficult and unable to reliably provide effective information for automatic driving.

[0004] In addition, existing data fusion methods are relatively simple and fail to fully leverage the strengths of different sensors, resulting in insufficient data fusion depth and difficulty in constructing accurate and real-time dynamic environment models. In terms of vehicle adaptive control, traditional technologies lack effective analysis and rapid response capabilities in complex environments, making it difficult to adjust vehicle driving states in a timely and reasonable manner, and unable to meet the stringent requirements of automatic driving safety and reliability. The above problems seriously hinder the development of automatic driving technology, causing significant safety risks for vehicle driving in complex traffic scenarios, and hindering the process of moving automatic driving technology from the laboratory to practical applications. SUMMARY

[0005] (I) Technical problems solved

[0006] To overcome the above-mentioned shortcomings of the prior art, the present application provides a dynamic environment modeling and active adaptation system for automatic driving vehicles, which can effectively overcome the defects of the prior art in constructing accurate and real-time dynamic environment models and adjusting vehicle driving states in a timely and reasonable manner.

[0007] (II) Technical solutions

[0008] To achieve the above purposes, the present application is implemented by the following technical solutions:

[0009] An autonomous vehicle dynamic environment modeling and active adaptation system, comprising:

[0010] A multi-band cognitive radar module integrates multiple frequency bands of millimeter wave radars, adopts intelligent switching and collaborative working mechanism based on the characteristic differences of radars in different frequency bands, automatically optimizes the frequency band working cycle proportion according to the environmental conditions, transmits and receives radar waves of different frequency bands, and obtains multi-dimensional information of targets;

[0011] A data processing and dynamic environment modeling module receives the original radar data transmitted by the multi-band cognitive radar module, pre-processes the original radar data, and then adopts a multi-sensor data fusion algorithm to fuse the multi-dimensional information of targets obtained by radars in different frequency bands, dynamically models the environment around the vehicle, and updates the dynamic environment model in real time;

[0012] A decision and active adaptation control module formulates reasonable driving decisions and planning paths based on the dynamic environment model, and adopts deep learning and artificial intelligence algorithms to realize active adaptation control by adjusting the speed, direction and braking of the vehicle, so as to realize high-precision environment perception and active adaptation control of the autonomous vehicle in the dynamic environment.

[0013] Preferably, the multi-band cognitive radar module integrates millimeter wave radars in multiple frequency bands including 24GHz, 77GHz and 79GHz, and adopts intelligent switching and collaborative working mechanism based on the characteristic differences of radars in different frequency bands: 24GHz radars use the characteristic of wide detection range to preliminarily detect long-distance targets, and 77GHz and 79GHz radars use the characteristic of high resolution to accurately identify the shape and contour of targets, automatically optimize the frequency band working cycle proportion according to the environmental conditions, and transmit and receive radar waves of different frequency bands to collect multi-dimensional information of targets including distance, speed and angle.

[0014] Preferably, the multi-band cognitive radar module introduces a collaborative detection technology based on dynamic spectrum reconstruction and adaptive beamforming, which comprises:

[0015] S11, a spectrum resource dynamic allocation model is constructed, and a reinforcement learning algorithm is adopted to optimize the frequency band working cycle proportion of each frequency band radar:

[0016] An environment state vector is defined to describe the comprehensive characteristics of the current working environment of the radar, wherein E is the environmental complexity, T d is the target density, is the interference intensity;

[0017] An action space A is defined:

[0018] ;

[0019] in, , , The frequency band duty cycles of radars at 24GHz, 77GHz, and 79GHz are respectively.

[0020] To guide the reinforcement learning algorithm in finding the optimal combination of frequency band duty cycle proportions, a reward function R is designed:

[0021] ;

[0022] Where, p det Target detection probability represents the likelihood that the system can detect a target given the current frequency band's operating cycle ratios. res is the comprehensive resolution index, which considers the resolution performance of radars in different frequency bands. tf is the inter-band interference level, measuring the degree of interference between radars in different frequency bands during operation. , , All are weighting coefficients;

[0023] In the reinforcement learning process, a deep Q-network is used for iterative training, through a neural network. An approximate estimate of the value of performing action A in state S, where For the parameters of the neural network, at each time step t, the algorithm adjusts the parameters based on the state S at the current time step. t Select action A at the current time step. t and perform the action A. t Obtain the state S at the next time step t+1 and the reward R at the current time step t Then by minimizing the loss function Update the parameters of the neural network :

[0024] ;

[0025] Among them, A t+1 To determine the state S at the next time step t+1 Choose the action for the next time step. The parameters for the target network are periodically copied from the main network. Update As a discount factor, Represents the mathematical expectation;

[0026] S12. At the signal processing level, adaptive beamforming technology is introduced to improve the radar's detection performance:

[0027] For the received signals x of radars in each frequency band i(t), i = 24, 77, 79, the radar receives signals through an array antenna, and the array antenna has N array elements, so the received signal of the radar of each frequency band is expressed as a vector:

[0028] ;

[0029] wherein X i is the received signal vector of the radar of frequency band i, x iN (t) is the received signal of the Nth array element of the radar of frequency band i, and t is time;

[0030] In order to enhance the target signal and suppress the interference signal, an adaptive weighting vector w i is calculated by using a minimum variance distortion response (MVDR) criterion, and a covariance matrix R of the received signal is first estimated, wherein represents a mathematical expectation, is the conjugate transpose of X i , and a target steering vector a i describes the phase relationship of each array element when the target signal reaches the array antenna, and the adaptive weighting vector w is calculated according to the MVDR criterion:

[0031] ;

[0032] wherein, in actual application, since the covariance matrix R i is unknown, an estimated value is obtained by using limited time sample data: , is the received signal vector of the radar of frequency band i at the mth sample time, and M is the sample number;

[0033] The main lobe of the radar beam is always aligned with the direction of the target signal by dynamically adjusting the adaptive weighting vector w i , a null is formed in the direction of the interference signal, the sidelobe interference is effectively suppressed, and the cooperative enhancement of the multi-frequency band radar signals is realized;

[0034] S13, the received signals x i (t) of the radars of each frequency band are weighted and fused to obtain the fused received signal x fusion (t):

[0035] .

[0036] Preferably, the data processing and dynamic environment modeling module comprises a signal preprocessing unit and an environment modeling unit;

[0037] The signal preprocessing unit receives the original radar data transmitted by the multi-frequency band cognitive radar module, adopts preprocessing methods such as denoising, filtering and calibration, removes interference signals and noises, corrects data deviation, and improves data accuracy and reliability.

[0038] The environment modeling unit adopts a multi-sensor data fusion algorithm to deeply fuse the data transmitted by the signal preprocessing unit, combines the multi-dimensional information of the target obtained by the different frequency band radars, including distance, speed and angle, dynamically models the environment around the vehicle, and updates the dynamic environment model in real time, realizes accurate tracking and motion trajectory prediction of dynamic targets, and provides real-time and accurate environment information for automatic driving.

[0039] Preferably, the environment modeling unit introduces an improved spatiotemporal adaptive weighted multi-sensor data fusion algorithm, which includes:

[0040] Let x i (t) be the measurement value of the i-th sensor in the n sensors at time t, i=1, 2, …, n, and the covariance matrix of the measurement error be P i (t).

[0041] First, define the spatiotemporal correlation coefficient matrix C(t), whose element c ij (t) represents the spatiotemporal correlation between the measurement values of the i-th sensor and the j-th sensor at time t, which is determined by calculating the Euclidean distance of the measurement data of the two sensors in time and space:

[0042] ;

[0043] Wherein, , are the measurement values of the i-th sensor, the j-th sensor at time t on the k-th target parameter, and m is the dimension of the target parameter.

[0044] Then, the adaptive weight w of the i-th sensor is calculated:

[0045] ;

[0046] Finally, the estimated value of the fused target parameter x is:

[0047] .

[0048] Preferably, a deep learning model based on a hybrid structure of convolutional neural network (CNN) and long short-term memory network (LSTM) is constructed, and the preprocessed multi-dimensional information of the target obtained by the different frequency band radars is converted into image or sequence format data as model input to further improve the accuracy of data fusion.

[0049] For image format data, the powerful feature extraction capabilities of Convolutional Neural Networks (CNNs) are used to extract spatial features, including the shape and contour of the target.

[0050] For sequence-formatted data, the time-series information is processed through a Long Short-Term Memory (LSTM) network to learn the time dependencies between target motions.

[0051] The model output is an estimate of the fused target parameters. The model is trained using a large amount of real-world scene data to optimize network parameters, enabling the model to accurately fuse multi-band radar data.

[0052] Preferably, the environment modeling unit performs dynamic environment modeling around the vehicle and updates the dynamic environment model in real time, including:

[0053] S21. Dynamic environment modeling around the vehicle based on an improved octree:

[0054] 1) Improve the octree to an adaptive partitioning algorithm based on information entropy. Let the i-th octree node O i The covered spatial region contains N i Target data points Each target data point contains information including the target's type, location, and velocity, and an octree node O is defined. i Information entropy for:

[0055] ;

[0056] Where, p ik For octree node O i The probability of the k-th target type. n ik For octree node O i The number of data points of the kth target type, where K is the number of target types;

[0057] Set the information entropy threshold H th When octree node O i Information entropy H(O) i () greater than the information entropy threshold H th At that time, for octree node O i Perform octree partitioning; otherwise, stop partitioning. This adaptive partitioning method allows the octree structure to better fit the environmental characteristics and improves modeling efficiency.

[0058] 2) Considering the continuity of target motion and the uncertainty of sensor measurements, the calculation method for the occupancy probability of occupancy nodes is updated and optimized by introducing a time decay factor. , and measurement confidence factor octree node O at time t i Occupancy probability p(O) i ,t) is calculated using the following formula:

[0059] ;

[0060] Wherein, p(O i (t-1) represents the octree node O at time t-1. i The occupancy probability, n ij Let time t be the time at octree node O i The number of targets detected by the j-th sensor, and the measurement confidence factor. This indicates that time t is at node O in the octree. i The confidence level of the target being detected by the j-th sensor is given, where M is the number of sensors involved in the detection. This formula for calculating the occupancy probability of an occupancy node retains historical occupancy information while dynamically adjusting it based on the current measurement, making the occupancy probability more accurately reflect the environmental state.

[0061] S22. Real-time updating of dynamic environment model based on dynamic Bayesian network:

[0062] 1) To more accurately describe the dependencies between octree nodes, a hierarchical dynamic Bayesian network (H-DBN) is introduced, which divides octree nodes into different subnets according to the hierarchy. There are horizontal dependencies between octree nodes in the same subnet and vertical dependencies between octree nodes in different subnets.

[0063] Let octree node O i Located in the l-th layer subnet, its state transition probability depends not only on its own state at the previous time step and the states of neighboring nodes in the same layer subnet, but also on the state of the corresponding parent node in the next higher layer subnet. For a three-layer dynamic Bayesian network H-DBN, the octree node O i The state transition probability is:

[0064] ;

[0065] in, For the position located at the octree nodes in layered subnet , For the position located at the Regarding octree nodes within the layered subnet The set of adjacent nodes, For the position located at the -1 layer subnet and octree nodes The hierarchical structure can more efficiently process large-scale node data and improve the efficiency of dynamic environment model updating, t represents the current time, t-1 represents the last time, and t represents the current time.

[0066] For the calculation of the measurement probability model based on the Gaussian mixture model GMM, the radar measurement data z t is defined as a mixture of L Gaussian distributions in the state of the octree node

[0067] ;

[0068] wherein, represents the measurement probability of the radar measurement data z in the state of the octree node t at time t, represents that the radar measurement data z t obeys a Gaussian distribution with a mean of and a covariance matrix of in the state of the octree node , represents the weight of the lth Gaussian distribution to which the radar measurement data z t obeys in the state of the octree node O i , , and are estimated by expectation maximization EM training on historical measurement data, so as to more accurately calculate the measurement probability and improve the accuracy of dynamic environment model updating.

[0069] Preferably, the decision and active adaptive control module comprises a decision planning unit and a control execution unit.

[0070] The decision planning unit comprehensively analyzes the current environment, vehicle state, preset safety strategy and driving target based on the dynamic environment model, formulates a reasonable driving decision and planning path, and quickly generates a response strategy when detecting obstacles or dangerous situations in front, thereby providing decision guidance for vehicle driving.

[0071] The control execution unit receives the driving decision and planning path instructions transmitted by the decision planning unit, controls the actuators such as the power, steering and braking systems of the vehicle, and actively adjusts the speed, direction and braking of the vehicle to realize high-precision environment perception and active adaptive control of the autonomous vehicle in a dynamic environment.

[0072] Preferably, the decision planning unit introduces a decision algorithm based on the fusion of graph neural network GNN and reinforcement learning, which comprises:

[0073] A dynamic graph structure G=(V,E) is constructed based on octree nodes, where the node set V corresponds to the octree nodes, and the edge set E represents the adjacency relationship between the octree nodes.

[0074] For the octree node O i , a node feature vector x i is defined, which contains information such as the occupancy probability, target type and position of the node, and the graph neural network GNN updates the node feature by aggregating neighbor node information:

[0075] ;

[0076] wherein, is the hidden state of the octree node at the l-th layer, is the hidden state of the neighbor octree node O j at the l-th layer, , , are the learnable weight matrix and bias vector of the l-th layer respectively, N(i) is the neighbor node set of the octree node O i , is an activation function;

[0077] The node feature output by the graph neural network GNN is taken as the state input of reinforcement learning, the state space S is defined as the set of all node feature vectors, the action space A is the set of driving decisions, and the reward function R(s,a) is designed as:

[0078] ;

[0079] wherein, R(s,a) is the reward obtained by executing the action a in the state s, s' is the next state after executing the action a, Safety(s') is a safety parameter measuring the safety of the vehicle in the state s', calculated according to the distance from the surrounding obstacles, Cost(a) is a cost parameter of executing the action a, and Progress(s') is a progress parameter measuring the progress of the vehicle towards the driving target in the state s', , , are weight coefficients;

[0080] Through policy gradient training, the intelligent agent learns the optimal driving decision :

[0081] ;

[0082] wherein, R(s t ,a t ) is the reward obtained by executing the action a in the state s tperforming the action a of the current time step t obtained reward, is a discount factor, t is a time step, denotes a mathematical expectation, is a general driving decision.

[0083] Preferably, the control execution unit introduces a model predictive control and game theory collaborative control algorithm, including:

[0084] A vehicle motion model is established, and a state vector of the vehicle at time t is defined as where x t and y t are position coordinates, is a driving angle, v t is a speed, and a control input vector of the vehicle at time t is where o t is an acceleration, is a steering angle, and a dynamics model of the vehicle is represented as:

[0085] ;

[0086] wherein, is a state of the vehicle at time t+1, f is a kinematics equation, is process noise;

[0087] An optimization problem with a prediction horizon of N is constructed, and an objective function J is:

[0088] ;

[0089] wherein, is a state of the vehicle at time t+k predicted at time t, is an expected state of the vehicle at time t+k is a control input vector of the vehicle at time t+k-1 predicted at time t, and are both weight vectors;

[0090] Considering the behavior of surrounding dynamic targets, a Nash equilibrium in game theory is introduced, assuming that there are M dynamic targets around the vehicle, the strategy of the dynamic target j is , j=1,2,…,M, the strategy of the vehicle is , and a utility function is defined as , and the Nash equilibrium is solved:

[0091] ;

[0092] wherein, , , is the optimal strategy of the vehicle, is the optimal strategy of the vehicle, is the optimal strategy of the vehicle, is the optimal strategy of the dynamic target j, represents the total utility of the vehicle and all participants around it, is the strategy of the i-th participant itself, is the strategy combination of all participants except the i-th participant;

[0093] The Nash equilibrium solution is taken as the constraint condition of model predictive control, the control input of the vehicle is optimized, and more safe and efficient active adaptive control is realized.

[0094] (Three) beneficial effects

[0095] Compared with the prior art, the automatic driving vehicle dynamic environment modeling and active adaptation system provided by the present application has the following beneficial effects:

[0096] 1) At the environmental perception level, the multi-band cognitive radar module integrates millimeter wave radars of multiple frequency bands such as 24GHz, 77GHz and 79GHz, and fully utilizes the advantages of radars of various frequency bands by means of intelligent switching and cooperative working mechanism, so as to make up for the defects of single-band radars in terms of detection distance, resolution and anti-interference ability, and accurately and comprehensively collect multi-dimensional information of targets including distance, speed and angle;

[0097] 2) The data processing and dynamic environment modeling module adopts data preprocessing and multi-sensor data fusion algorithm, constructs and updates the dynamic environment model in real time, realizes accurate tracking and motion trajectory prediction of dynamic targets, and provides real-time and accurate environmental information for automatic driving;

[0098] 3) The decision and active adaptive control module adopts deep learning and artificial intelligence algorithm based on the dynamic environment model, formulates reasonable driving decisions and planning paths, and realizes high-precision environmental perception and active adaptive control of the automatic driving vehicle in the dynamic environment by adjusting the speed, direction and braking of the vehicle;

[0099] The present application significantly improves the environmental perception accuracy, adaptive control ability and driving safety of the automatic driving vehicle in complex environment, effectively reduces the risk of accidents, provides a reliable solution for the development and application of automatic driving technology, and effectively promotes the automatic driving industry to a higher level. BRIEF DESCRIPTION OF DRAWINGS

[0100] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other accompanying drawings can be obtained by those skilled in the art without creative labor on the premise of the accompanying drawings.

[0101] Figure 1 The first system schematic diagram of the present application;

[0102] Figure 2 The second system schematic diagram of the present application;

[0103] Figure 3 The flowchart of the cooperative detection technology based on dynamic spectrum reconstruction and adaptive beamforming in the present application;

[0104] Figure 4 The flowchart of constructing and updating the dynamic environment model in the present application;

[0105] Figure 5 The schematic diagram of the decision algorithm based on the fusion of graph neural network GNN and reinforcement learning in the present application;

[0106] Figure 6 The flowchart of active adaptive control of the vehicle in the present application. DETAILED DESCRIPTION

[0107] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other accompanying drawings can be obtained by those skilled in the art without creative labor on the premise of the accompanying drawings.

[0108] The specific functional modules (such as shown in FIG. 1) and technical effects of the automatic driving vehicle dynamic environment modeling and active adaptation system provided by the present application will be introduced below in combination with specific examples, and the system functional modules include: Figure 1

[0109] Multi-band cognitive radar module, integrating multiple frequency band millimeter wave radars, based on the characteristic differences of each frequency band radar, using intelligent switching and cooperative working mechanism, automatically optimizing the frequency band working cycle proportion according to the environment condition, and transmitting and receiving radar waves of different frequency bands to obtain target multi-dimensional information;

[0110] ​The data processing and dynamic environment modeling module receives the original radar data transmitted by the multi-band cognitive radar module, pre-processes the original radar data, and then adopts a multi-sensor data fusion algorithm to fuse the multi-dimensional information of the target obtained by the radars of different frequency bands, models the dynamic environment around the vehicle, and updates the dynamic environment model in real time.

[0111] The decision and active adaptive control module formulates reasonable driving decisions and planning paths based on the dynamic environment model, and adopts deep learning and artificial intelligence algorithms to actively adapt the control by adjusting the speed, direction and braking of the vehicle, so as to realize high-precision environment perception and active adaptive control of the autonomous vehicle in the dynamic environment.

[0112] I. Multi-band cognitive radar module (i.e. Figure 2 Data acquisition unit in it)

[0113] The multi-band cognitive radar module integrates millimeter wave radars of multiple frequency bands including 24GHz, 77GHz and 79GHz. Based on the characteristic differences of radars of different frequency bands: the 24GHz radar uses the characteristic of wide detection range to preliminarily detect long-distance targets, and the 77GHz and 79GHz radars accurately identify the shape and contour of the target by virtue of the characteristic of high resolution. The intelligent switching and cooperative working mechanism is adopted to automatically optimize the frequency band working cycle proportion according to the environmental conditions, and to transmit and receive radar waves of different frequency bands to collect multi-dimensional information of the target including distance, speed and angle.

[0114] As shown in Figure 3 , the multi-band cognitive radar module introduces a cooperative detection technology based on dynamic spectrum reconstruction and adaptive beamforming, including:

[0115] S11, a spectrum resource dynamic allocation model is constructed, and a reinforcement learning algorithm is adopted to optimize the frequency band working cycle proportion of each frequency band radar:

[0116] The environment state vector is defined to describe the comprehensive characteristics of the current working environment of the radar, where E is the environmental complexity, T d is the target density, is the interference intensity;

[0117] The action space A is defined as:

[0118] ;

[0119] Among them, , , are the frequency band working cycle proportions of 24GHz, 77GHz and 79GHz radars respectively;

[0120] To guide the reinforcement learning algorithm to find the optimal combination of frequency band working cycle ratio, a reward function R is designed:

[0121] ;

[0122] where p det is the target detection probability, reflecting the possibility that the system can detect the target under the current combination of frequency band working cycle ratio, res is the comprehensive resolution index, considering the resolution performance of different frequency band radars, tf is the inter-frequency interference degree, measuring the interference degree between different frequency band radars when working, 、 、 are weight coefficients;

[0123] In the reinforcement learning process, a deep Q network is used for iterative training, and a neural network is used to approximate the value of performing action A in state S, where is the parameter of the neural network, at each time step t, the algorithm selects the action A t of the current time step according to the state S t of the current time step, t performs the action A t+1 to get the state S t of the next time step and the reward R of the current time step, then the parameters of the neural network are updated by minimizing the loss function :

[0124] ;

[0125] where A t+1 is the action selected according to the state S t+1 of the next time step, is the parameter of the target network, which is updated periodically from the main network, is the discount factor, denotes the mathematical expectation;

[0126] S12, at the level of received signal processing, adaptive beamforming technology is introduced to improve the detection performance of the radar:

[0127] For the received signal x i of each frequency band radar, i=24,77,79, the radar receives the signal through an array antenna, and it is specified that the array antenna has N elements, then the received signal of each frequency band radar is represented as a vector:

[0128] ;

[0129] ​Among them, X i Let x be the received signal vector of the radar in frequency band i. iN (t) represents the received signal of the Nth element of the radar in frequency band i, where t is time;

[0130] To enhance the target signal and suppress interference signals, the adaptive weighting vector w is calculated using the Minimum Variance Distortionless Response (MVDR) criterion. i First, estimate the covariance matrix of the received signal. ,in Represents the mathematical expectation. For X i The conjugate transpose of the target guidance vector a i This describes the phase relationship of each array element when the target signal arrives at the array antenna, and calculates the adaptive weighting vector according to the MVDR criterion. :

[0131] ;

[0132] In practical applications, due to the covariance matrix R i Since it is unknown, an estimate is obtained using finite-time sample data. : , Let M be the received signal vector of the radar at frequency band i at the m-th sample time, where M is the number of samples;

[0133] By dynamically adjusting the adaptive weighting vector w i This ensures that the main lobe of the radar beam is always aligned with the target signal direction, while simultaneously creating nulls in the direction of interference signals, effectively suppressing sidelobe interference and achieving coordinated enhancement of multi-band radar signals.

[0134] S13, Received signals x from radars in each frequency band i (t) is weighted and fused to obtain the fused received signal x. fusion (t):

[0135] .

[0136] In terms of environmental perception, the above technical solution integrates millimeter-wave radars in multiple frequency bands such as 24GHz, 77GHz, and 79GHz. With intelligent switching and collaborative working mechanisms, it fully leverages the advantages of each frequency band radar, making up for the deficiencies of single-frequency radar in terms of detection distance, resolution, and anti-interference capabilities, and accurately and comprehensively collects multi-dimensional information of the target, including distance, speed, and angle.

[0137] II. Data Processing and Dynamic Environment Modeling Module

[0138] like Figure 2As shown, the data processing and dynamic environment modeling module includes a signal preprocessing unit and an environment modeling unit;

[0139] The signal preprocessing unit receives the original radar data transmitted by the multi-band cognitive radar module, adopts preprocessing methods such as denoising, filtering and calibration, removes interference signals and noises, corrects data deviation, and improves data accuracy and reliability;

[0140] The environment modeling unit adopts a multi-sensor data fusion algorithm to deeply fuse the data transmitted by the signal preprocessing unit, combines multi-dimensional information such as the distance, speed and angle of the target obtained by different frequency band radars, dynamically models the environment around the vehicle, and updates the dynamic environment model in real time to realize accurate tracking and motion trajectory prediction of dynamic targets (such as other vehicles, pedestrians, obstacles, etc.), and provides real-time and accurate environment information for automatic driving.

[0141] As shown in Figure 4 , the environment modeling unit introduces an improved spatiotemporal adaptive weighted multi-sensor data fusion algorithm, which includes:

[0142] Let x i (t) be the measurement value of the i-th sensor in the n sensors at time t for the target parameter, i=1,2,…,n, and the covariance matrix of the measurement error be P i (t).

[0143] First, define the spatiotemporal correlation coefficient matrix C(t), whose element c ij (t) represents the spatiotemporal correlation between the measurement values of the i-th sensor and the j-th sensor at time t for the target parameter, which is determined by calculating the Euclidean distance of the measurement data of the two sensors in time and space:

[0144] ;

[0145] Where, , are the measurement values of the i-th sensor and the j-th sensor at time t for the k-th target parameter, and m is the dimension of the target parameter.

[0146] Then, the adaptive weight of the i-th sensor is calculated:

[0147] ;

[0148] Finally, the estimated value of the fused target parameter is:

[0149] .

[0150] In the technical solution of this application, a deep learning model based on a hybrid structure of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) can be constructed. This model converts pre-processed multi-dimensional target information acquired by radar in different frequency bands into image or sequence format data as input to the model, thereby further improving the accuracy of data fusion.

[0151] For image format data, the powerful feature extraction capabilities of Convolutional Neural Networks (CNNs) are used to extract spatial features, including the shape and contour of the target.

[0152] For sequence-formatted data, the time-series information is processed through a Long Short-Term Memory (LSTM) network to learn the time dependencies between target motions.

[0153] The model output is an estimate of the fused target parameters. The model is trained using a large amount of real-world scene data to optimize network parameters, enabling the model to accurately fuse multi-band radar data.

[0154] like Figure 4 As shown, the environment modeling unit performs dynamic environment modeling around the vehicle and updates the dynamic environment model in real time, including:

[0155] S21. Dynamic environment modeling around a vehicle based on an improved octree: 1) The improved octree is an adaptive partitioning algorithm based on information entropy. Let the i-th octree node O i The covered spatial region contains N i Target data points Each target data point contains information including the target's type, location, and velocity, and an octree node O is defined. i Information entropy for:

[0156] ;

[0157] Where, p ik For octree node O i The probability of the k-th target type. n ik For octree node O i The number of data points of the kth target type, where K is the number of target types;

[0158] Set the information entropy threshold H th When octree node O i Information entropy H(O) i () greater than the information entropy threshold H th At that time, for octree node O i Perform octree partitioning; otherwise, stop partitioning. This adaptive partitioning method allows the octree structure to better fit the environmental characteristics and improves modeling efficiency.

[0159] 2) Considering the motion continuity of the target and the uncertainty of sensor measurement, the occupancy probability calculation method of octree node is updated and optimized, and a time decay factor , and a measurement confidence factor are introduced i The occupancy probability p(O i ,t) of octree node O i at time t is calculated as follows:

[0160] ;

[0161] Where p(O ij ,t-1) is the occupancy probability of octree node O i at time t-1, n is the number of targets detected by the jth sensor within octree node O i at time t, and the measurement confidence factor represents the confidence of detecting targets within octree node O i by the jth sensor at time t, and M is the number of participating sensors. This occupancy probability calculation formula for octree node not only retains historical occupancy information, but also dynamically adjusts according to current measurement, making the occupancy probability more accurately reflect the environment state.

[0162] S22, real-time update the dynamic environment model based on dynamic Bayesian network:

[0163] 1) To more accurately describe the dependency relationship between octree nodes, a hierarchical dynamic Bayesian network H-DBN is introduced, which divides octree nodes into different subnets according to levels, and there is a horizontal dependency relationship between octree nodes in the same level subnet, and a vertical dependency relationship between octree nodes in different level subnets.

[0164] Let octree node O i be located in the lth level subnet, its state transition probability depends not only on its own state and the state of adjacent nodes in the same level subnet at the previous time, but also on the state of the corresponding parent node in the previous level subnet. For a three-level dynamic Bayesian network H-DBN, the state transition probability of octree node O i is:

[0165] ;

[0166] Where is the occupancy probability of octree node O located in the lth level subnet, is the occupancy probability of octree node O located in the (l-1)th level subnet, and is the occupancy probability of octree node O located in the (l+1)th level subnet. a set of adjacent nodes of the octree node, a set of parent nodes corresponding to the octree node at the first layer subnet, t represents the current time, and t-1 represents the last time, and the hierarchical structure can more efficiently process large-scale node data and improve the efficiency of dynamic environment model updating;

[0167] For the calculation of , a measurement probability model based on a Gaussian mixture model (GMM) is constructed, and the radar measurement data z t in the state of the octree node obeys a mixture of L Gaussian distributions:

[0168] ;

[0169] wherein, represents the measurement probability of the radar measurement data z t in the state of the octree node at time t, represents that the radar measurement data z t obeys a Gaussian distribution with a mean of and a covariance matrix of in the state of the octree node O i , represents the weight of the radar measurement data z t obeys the L-th Gaussian distribution in the state of the octree node , parameters , and are estimated through expectation maximization (EM) training of historical measurement data, the measurement probability is more accurately calculated, and the accuracy of dynamic environment model updating is improved.

[0170] The above technical solution, the data processing and dynamic environment modeling module adopts a data preprocessing and multi-sensor data fusion algorithm, constructs and real-time updates a dynamic environment model, realizes accurate tracking and motion trajectory prediction of dynamic targets, and provides real-time and accurate environment information for automatic driving.

[0171] III. Decision and active adaptive control module

[0172] As shown in Figure 2 , the decision and active adaptive control module includes a decision planning unit and a control execution unit.

[0173] The decision-making and planning unit, based on a dynamic environment model, uses deep learning and artificial intelligence algorithms to comprehensively analyze the current environmental conditions, vehicle status, preset safety strategies and driving goals, formulate reasonable driving decisions and plan paths, and quickly generate response strategies when obstacles or dangerous situations are detected ahead, providing decision guidance for vehicle driving.

[0174] The control execution unit receives driving decisions and path planning instructions transmitted by the decision planning unit, controls the vehicle's power, steering, and braking systems, and performs active adaptive control by adjusting the vehicle's speed, direction, and braking to achieve high-precision environmental perception and active adaptive control of autonomous vehicles in dynamic environments.

[0175] like Figure 5 and Figure 6 As shown, the decision planning unit introduces a decision algorithm based on the fusion of graph neural networks (GNN) and reinforcement learning, including:

[0176] A dynamic graph structure G=(V,E) is constructed based on octree nodes, where the node set V corresponds to the octree nodes and the edge set E represents the adjacency relationship between the octree nodes.

[0177] For octree node O i Define the node feature vector x i This includes information such as the node's occupancy probability, target type, and location. Graph Neural Networks (GNNs) update node features by aggregating information from neighboring nodes.

[0178] ;

[0179] in, octree node In the +1 layer hidden state, Neighboring octree nodes The hidden state at layer l , These are the weight matrix and bias vector of the learnable l-th layer, respectively, and N(i) is the octree node. The set of neighboring nodes, For activation functions;

[0180] Using the node features output by the graph neural network (GNN) as the state input for reinforcement learning, we define the state space S as the set of all node feature vectors, the action space A as the set of driving decisions, and the reward function R(s,a) as follows:

[0181] ;

[0182] where R(s, a) is the reward obtained by performing action a in state s, s' is the next state after performing action a, Safety(s') is a parameter measuring the safety level of the vehicle in state s', calculated according to the distance from the surrounding obstacles, Cost(a) is a parameter of the cost of performing action a, and Progress(s') is a parameter measuring the progress of the vehicle towards the driving target in state s', 、 、 are weight coefficients;

[0183] Through policy gradient training, the intelligent agent learns the optimal driving decision :

[0184] ;

[0185] where R(s t ,a t ) is the reward obtained by performing action a t in the current time step s t , is a discount factor, t is a time step, denotes mathematical expectation, is a general driving decision.

[0186] As shown in Figure 6 , the control execution unit introduces a model predictive control and game theory collaborative control algorithm, including:

[0187] A vehicle motion model is established, and the state vector of the vehicle at time t is defined as where x t and y t are position coordinates, is the driving angle, and v t is the speed. The control input vector of the vehicle at time t is where o t is the acceleration, is the steering angle, and the dynamics model of the vehicle is represented as:

[0188] ;

[0189] where s is the state of the vehicle at time t+1, f is the kinematics equation, is the process noise;

[0190] An optimization problem with a prediction horizon of N is constructed, and the objective function J is:

[0191] ;

[0192] where, predicting the state of the vehicle at time t+k given the state of the vehicle at time t, predicting the desired state of the vehicle at time t+k given the state of the vehicle at time t, predicting the control input vector of the vehicle at time t+k-1 given the state of the vehicle at time t, and are weight vectors;

[0193] Considering the behavior of surrounding dynamic targets, introducing the Nash equilibrium in game theory, assuming that there are M dynamic targets around the vehicle, the strategy of dynamic target j is , j = 1, 2, …, M, the strategy of the vehicle is , defining the utility function , solving the Nash equilibrium:

[0194]

[0195] wherein, , , is the Nash equilibrium solution, is the optimal strategy of the vehicle, , is the optimal strategy of dynamic target j, represents the total utility of the vehicle and all participants around the vehicle, is the strategy of the i th participant itself, is the strategy combination of all other participants except the i th participant;

[0196] Taking the Nash equilibrium solution as the constraint condition of model predictive control, optimizing the control input of the vehicle, realizing more safe and efficient active adaptive control.

[0197] The above technical solution, the decision and active adaptive control module is based on the dynamic environment model, adopts deep learning and artificial intelligence algorithm, formulates reasonable driving decision and planning path, adjusts the speed, direction and brake of the vehicle for active adaptive control, realizes high-precision environment perception and active adaptive control of the autonomous vehicle in the dynamic environment.

[0198] On the basis of the above disclosure of the autonomous vehicle dynamic environment modeling and active adaptation system, the present application also provides an autonomous vehicle dynamic environment modeling and active adaptation device, comprising a memory, a processor, and a computer program stored in the memory and running in the processor. When the processor executes the computer program, the method corresponding to the above disclosed autonomous vehicle dynamic environment modeling and active adaptation system is adopted, realizing high-precision environment perception and active adaptive control of the autonomous vehicle.

[0199] ​The automatic driving vehicle dynamic environment modeling and active adaptation device provided by the embodiment is essentially a computer device for implementing the above system and the corresponding method. The computer device can be an embedded device installed on each vehicle, which can communicate and interact with other devices through a server to obtain required data and perform data processing in a manner of independent operation or edge computing. The computer device can also be a backend device independent of the vehicle, which then completes the data processing process when a user makes a request, and sends dynamic environment modeling information to the data terminal in each vehicle according to the data processing result, and assists the vehicle to complete active adaptation control.

[0200] The computer device in the embodiment can also be an intelligent terminal, a tablet computer, a notebook computer, a desktop computer, a rack server, a blade server, a tower server, or a cabinet server (including a standalone server or a server cluster composed of multiple servers).

[0201] The computer device in the embodiment at least includes, but is not limited to, a memory and a processor that can be connected to each other for communication through a system bus. The memory (i.e., a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (such as an SD, DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital card (SD), a flash card, etc. equipped on the computer device. Of course, the memory can include both the internal storage unit and the external storage device of the computer device. In the embodiment, the memory is usually used to store the operating system and various application software installed on the computer device, etc. In addition, the memory can also be used to temporarily store various data that have been output or will be output.

[0202] The processor in some embodiments can be a central processing unit (CPU), a graphics processing unit (GPU), a controller, a microcontroller, a microprocessor, a data processing chip, etc. The processor is usually used to control the overall operation of the computer device. In the embodiment, the processor is used to run the program code or process the data stored in the memory.

[0203] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those ordinarily skilled in the art should understand: the technical solutions recorded in the foregoing examples can still be modified, or some technical features can be replaced equivalently; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A dynamic environment modeling and active adaptation system for autonomous vehicles, characterized in that: include: The multi-band cognitive radar module integrates millimeter-wave radars of multiple frequency bands. Based on the differences in characteristics of each frequency band radar, it adopts an intelligent switching and collaborative working mechanism, automatically optimizes the frequency band working cycle ratio according to environmental conditions, and transmits and receives radar waves of different frequency bands to obtain multi-dimensional information about the target. The data processing and dynamic environment modeling module receives raw radar data transmitted from the multi-band cognitive radar module, preprocesses the raw radar data, and then uses a multi-sensor data fusion algorithm to fuse multi-dimensional target information acquired by radars of different frequency bands, to model the dynamic environment around the vehicle, and updates the dynamic environment model in real time. The decision-making and active adaptive control module, based on a dynamic environment model and employing deep learning and artificial intelligence algorithms, formulates reasonable driving decisions and plans paths. It performs active adaptive control by adjusting the vehicle's speed, direction, and braking, thereby achieving high-precision environmental perception and active adaptive control for autonomous vehicles in dynamic environments. The multi-band cognitive radar module integrates millimeter-wave radars in multiple frequency bands, including 24GHz, 77GHz, and 79GHz. Based on the differences in characteristics of each frequency band radar: the 24GHz radar utilizes its wide detection range for preliminary detection of long-distance targets, while the 77GHz and 79GHz radars accurately identify the shape and outline of targets due to their high resolution. It adopts an intelligent switching and collaborative working mechanism to automatically optimize the frequency band working cycle ratio according to environmental conditions, and transmits and receives radar waves from different frequency bands to collect multi-dimensional information of the target, including distance, speed, and angle.

2. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 1, characterized in that: The multi-band cognitive radar module introduces a collaborative detection technology based on dynamic spectrum reconstruction and adaptive beamforming, including: S11. Construct a dynamic spectrum resource allocation model and use reinforcement learning algorithms to optimize the frequency band duty cycle ratio of radars in each frequency band: Define the environment state vector This describes the comprehensive characteristics of the radar's current operating environment, where E represents environmental complexity and T... d Let I be the target density and I be the interference intensity. Define action space A: ； in, , , The frequency band duty cycles of radars at 24GHz, 77GHz, and 79GHz are respectively. To guide the reinforcement learning algorithm in finding the optimal combination of frequency band duty cycle proportions, a reward function R is designed: ； Where, p det Target detection probability represents the likelihood that the system can detect a target given the current frequency band's operating cycle ratios. res is the comprehensive resolution index, which considers the resolution performance of radars in different frequency bands. tf is the inter-band interference level, measuring the degree of interference between radars in different frequency bands during operation. , , All are weighting coefficients; In the reinforcement learning process, a deep Q-network is used for iterative training, through a neural network. An approximate estimate of the value of performing action A in state S, where For the parameters of the neural network, at each time step t, the algorithm adjusts the parameters based on the state S at the current time step. t Select action A at the current time step. t and perform the action A. t Obtain the state S at the next time step t+1 and the reward R at the current time step t Then by minimizing the loss function Update the parameters of the neural network : ； Among them, A t+1 To determine the state S at the next time step t+1 Choose the action for the next time step. The parameters for the target network are periodically copied from the main network. Update As a discount factor, Represents the mathematical expectation; S12. At the signal processing level, adaptive beamforming technology is introduced to improve the radar's detection performance: For the received signals x of radars in each frequency band i (t), i=24,77,79, the radar receives signals through an array antenna. The array antenna is defined as having N elements. The received signals of the radar in each frequency band are represented as vectors: ； Among them, X i Let x be the received signal vector of the radar in frequency band i. iN (t) represents the received signal of the Nth element of the radar in frequency band i, where t is time; To enhance the target signal and suppress interference signals, the adaptive weighting vector w is calculated using the Minimum Variance Distortionless Response (MVDR) criterion. i First, estimate the covariance matrix of the received signal. ,in Represents the mathematical expectation. For X i The conjugate transpose of the target guidance vector a i This describes the phase relationship of each array element when the target signal arrives at the array antenna, and calculates the adaptive weighting vector w according to the MVDR criterion. i : ； In practical applications, due to the covariance matrix R i Since it is unknown, an estimate is obtained using finite-time sample data. : X i (m) is the received signal vector of the radar in frequency band i at the m-th sample time, and M is the number of samples; By dynamically adjusting the adaptive weighting vector w i This ensures that the main lobe of the radar beam is always aligned with the target signal direction, while simultaneously creating nulls in the direction of interference signals, effectively suppressing sidelobe interference and achieving coordinated enhancement of multi-band radar signals. S13, Received signals x from radars in each frequency band i (t) is weighted and fused to obtain the fused received signal x. fusion (t): 。 3. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 1, characterized in that: The data processing and dynamic environment modeling module includes a signal preprocessing unit and an environment modeling unit. The signal preprocessing unit receives the raw radar data transmitted by the multi-band cognitive radar module and uses preprocessing methods including denoising, filtering and calibration to remove interference signals and noise, correct data deviations and improve data accuracy and reliability. The environment modeling unit employs a multi-sensor data fusion algorithm to deeply fuse the data transmitted by the signal preprocessing unit. It combines multi-dimensional information, including the distance, speed, and angle of the target obtained from radars of different frequency bands, to perform dynamic environment modeling around the vehicle and update the dynamic environment model in real time. This enables accurate tracking of dynamic targets and prediction of their motion trajectories, providing real-time and accurate environmental information for autonomous driving.

4. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 3, characterized in that: The environmental modeling unit introduces an improved spatiotemporal adaptive weighted multi-sensor data fusion algorithm, including: Let the measurement value of the target parameter by the i-th sensor among n sensors at time t be x. i (t), i=1,2,…,n, and the covariance matrix of the measurement error is P. i (t); First, define the spatiotemporal correlation coefficient matrix C(t), whose elements c ij (t) represents the spatiotemporal correlation between the measurements of the target parameter by the i-th sensor and the j-th sensor at time t, which is determined by calculating the Euclidean distance between the measurement data of the two sensors in time and space: ； in, , , respectively, are the measurement values ​​of the i-th sensor and the j-th sensor at time t for the k-th target parameter, where m is the dimension of the target parameter; Then, the adaptive weights of the i-th sensor are calculated. : ； Finally, the estimated values ​​of the fused target parameters for: 。 5. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 4, characterized in that: A deep learning model based on a hybrid structure of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) is constructed. Preprocessed multi-dimensional target information acquired by radar across different frequency bands is transformed into image or sequence data as model input to further improve the accuracy of data fusion. For image format data, the powerful feature extraction capabilities of Convolutional Neural Networks (CNNs) are used to extract spatial features, including the shape and contour of the target. For sequence-formatted data, the time-series information is processed through a Long Short-Term Memory (LSTM) network to learn the time dependencies between target motions. The model output is an estimate of the fused target parameters. The model is trained using a large amount of real-world scene data to optimize network parameters, enabling the model to accurately fuse multi-band radar data.

6. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 4, characterized in that: The environment modeling unit performs dynamic environment modeling around the vehicle and updates the dynamic environment model in real time, including: S21. Dynamic environment modeling around the vehicle based on an improved octree: 1) Improve the octree to an adaptive partitioning algorithm based on information entropy. Let the i-th octree node O i The covered spatial region contains N i Target data points Each target data point contains information including the target's type, location, and velocity, and an octree node O is defined. i Information entropy H(O) i )for: ； Where, p ik For octree node O i The probability of the k-th target type. n ik For octree node O i The number of data points of the kth target type, where K is the number of target types; Set the information entropy threshold H th When octree node O i Information entropy H(O) i () greater than the information entropy threshold H th At that time, for octree node O i Perform octree partitioning; otherwise, stop partitioning. This adaptive partitioning method allows the octree structure to better fit the environmental characteristics and improves modeling efficiency. 2) Considering the continuity of target motion and the uncertainty of sensor measurements, the calculation method for the occupancy probability of occupancy nodes is updated and optimized by introducing a time decay factor. , and measurement confidence factor octree node O at time t i Occupancy probability p(O) i ,t) is calculated using the following formula: ； Wherein, p(O i (t-1) represents the octree node O at time t-1. i The occupancy probability, n ij Let time t be the time at octree node O i The number of targets detected by the j-th sensor, and the measurement confidence factor. This indicates that time t is at node O in the octree. i The confidence level of the target being detected by the j-th sensor is given, where M is the number of sensors involved in the detection. This formula for calculating the occupancy probability of an occupancy node retains historical occupancy information while dynamically adjusting it based on the current measurement, making the occupancy probability more accurately reflect the environmental state. S22. Real-time updating of dynamic environment model based on dynamic Bayesian network: 1) To more accurately describe the dependencies between octree nodes, a hierarchical dynamic Bayesian network (H-DBN) is introduced, which divides octree nodes into different subnets according to the hierarchy. There are horizontal dependencies between octree nodes in the same subnet and vertical dependencies between octree nodes in different subnets. Let octree node O i Located in the l-th layer subnet, its state transition probability depends not only on its own state at the previous time step and the states of neighboring nodes in the same layer subnet, but also on the state of the corresponding parent node in the next higher layer subnet. For a three-layer dynamic Bayesian network H-DBN, the octree node O i The state transition probability is: ； in, For octree node O located in the l-th layer subnet i , For the octree node O located in the l-th subnet... i The set of adjacent nodes, For the octree node O located in the (l-1)th layer subnet... i The corresponding set of parent nodes, where t represents the current time and t-1 represents the previous time, this hierarchical structure can process large-scale node data more efficiently and improve the efficiency of dynamic environment model updates; for The calculation constructs a measurement probability model based on the Gaussian Mixture Model (GMM), specifying the radar measurement data z. t In octree node O i The state is a mixture of L Gaussian distributions: ； in, This indicates that at time t, given octree node O i Radar measurement data z under the condition t The probability of measurement, Represents radar measurement data z t In octree node O i Under the condition of following the mean The covariance matrix is Gaussian distribution, Represents radar measurement data z t In octree node O i The weights, which follow the l-th Gaussian distribution under each state, are estimated by training with Expectation-Maximization (EM) on historical measurement data. , and This allows for more accurate calculation of measurement probabilities and improves the accuracy of dynamic environment model updates.

7. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 1, characterized in that: The decision-making and proactive adaptive control module includes a decision-making planning unit and a control execution unit; The decision-making and planning unit, based on a dynamic environment model, uses deep learning and artificial intelligence algorithms to comprehensively analyze the current environmental conditions, vehicle status, preset safety strategies and driving goals, formulate reasonable driving decisions and plan paths, and quickly generate response strategies when obstacles or dangerous situations are detected ahead, providing decision guidance for vehicle driving. The control execution unit receives driving decisions and path planning instructions transmitted by the decision planning unit, controls the vehicle's power, steering, and braking systems, and performs active adaptive control by adjusting the vehicle's speed, direction, and braking to achieve high-precision environmental perception and active adaptive control of autonomous vehicles in dynamic environments.

8. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 7, characterized in that: The decision planning unit introduces a decision algorithm based on the fusion of graph neural networks (GNN) and reinforcement learning, including: A dynamic graph structure G=(V,E) is constructed based on octree nodes, where the node set V corresponds to the octree nodes and the edge set E represents the adjacency relationship between the octree nodes. For octree node O i Define the node feature vector x i This includes information such as the node's occupancy probability, target type, and location. Graph Neural Networks (GNNs) update node features by aggregating information from neighboring nodes. ； in, For octree node O i In the hidden state at layer l+1, For neighboring octree node O j The hidden state at layer l , These are the weight matrix and bias vector of the learnable l-th layer, respectively, and N(i) is the octree node O. i The set of neighboring nodes, For activation functions; Using the node features output by the graph neural network (GNN) as the state input for reinforcement learning, we define the state space S as the set of all node feature vectors, the action space A as the set of driving decisions, and the reward function R(s,a) as follows: ； Where R(s,a) is the reward obtained by performing action a in state s, s' is the next state after performing action a, Safety(s') is a parameter that measures the safety level of the vehicle in state s', calculated based on the distance to surrounding obstacles, Cost(a) is the cost parameter for performing action a, and Progress(s') is a parameter that measures the progress of the vehicle towards the driving target in state s'. , , All are weighting coefficients; Through policy gradient training, the agent learns the optimal driving decision. : ； Wherein, R(s) t ,a t ) represents the state s at the current time step. t Execute action a at the current time step t The rewards received Here, t is the discount factor, and t is the time step. Represents the mathematical expectation. This is a general driving decision.

9. The autonomous vehicle dynamic environment modeling and active adaptation system according to claim 8, characterized in that: The control execution unit incorporates a collaborative control algorithm based on model predictive control and game theory, including: Establish a vehicle motion model, and define the vehicle's state vector at time t as... , where x t and y t For position coordinates, v is the driving angle. t For velocity, the control input vector of the vehicle at time t is: , where o t For acceleration, Let be the steering angle. The vehicle's dynamic model is represented as: ； in, Let f be the state of the vehicle at time t+1, and f be the kinematic equation. This is process noise; Construct an optimization problem with prediction in the time domain N, and the objective function J is: ； in, To predict the state of the vehicle at time t+k at time t, Let be the desired state of the vehicle at time t+k. To predict the vehicle's control input vector at time t+k-1 at time t, and All are weight vectors; Considering the behavior of surrounding dynamic targets, we introduce the Nash equilibrium from game theory. Suppose there are M dynamic targets around the vehicle, and the strategy of target j is... For j=1,2,…,M, the vehicle's strategy is... Define utility function Solve for the Nash equilibrium: ； in, , ,…, For Nash equilibrium solutions, This is the optimal strategy for the vehicle. ,…, The optimal strategy for dynamic objective j is... This represents the sum of the utility of the vehicle and all participants in its M surrounding dynamic targets. For the strategy of the i-th participant, The strategy combination for all participants except the i-th participant; By using the Nash equilibrium solution as a constraint for model predictive control, the control input of the vehicle is optimized, achieving safer and more efficient active adaptive control.

Images