Rrt* and dwa hybrid path planning method based on gaussian mixture model

Abstract

The application discloses a kind of RRT* and DWA hybrid path planning method based on Gaussian mixture model, comprising: 1, establish grid map, set starting position and end position, 2, utilize RRT* algorithm to plan the global path of ego vehicle, select the coordinate closest to the current position of ego vehicle as the sub-target point of dynamic window, 3, use Gaussian mixture model to express the motion trajectory of surrounding vehicle, 4, establish Gaussian process regression model to predict the trajectory of surrounding vehicle, 5, construct dynamic window cost function, plan the path from the current position of ego vehicle to sub-target point, 6, with the last path point planned by step 5 dynamic window method as the current position of ego vehicle, return to step 3 sequentially until the current position is end position, so as to obtain the path of ego vehicle from starting position to end position.The application can accurately estimate the motion state of vehicle, so as to realize collision avoidance while preventing the safety distance between the vehicle and the obstacle from being too small.

Description

Technical Field

[0001] This invention belongs to the field of vehicle routing planning, specifically a hybrid routing planning method based on Gaussian mixture model (RRT* and DWA). Background Technology

[0002] Path planning technology primarily receives data processed by the perception module, plans the trajectory for the autonomous vehicle, and sends it to the control module. It is a key technology that needs to be solved in autonomous driving. Path planning is about obtaining a feasible or optimal path from the initial position to the target position given known environmental information.

[0003] Currently, path planning algorithms are divided into global path planning and local path planning. Global path planning algorithms include Dijkstra's algorithm, A* algorithm, and RRT algorithm; local path planning algorithms include artificial potential field method, neural network method, and fuzzy logic method. The dynamic window method has advantages such as simplicity, efficiency, real-time adaptability, path length control, path smoothness, and scalability among path planning algorithms. This makes it a commonly used path planning algorithm suitable for various application scenarios.

[0004] Traditional dynamic window methods are based on the sliding window concept, searching for the optimal path by adjusting the size and position of the window. However, when applied to environments with multiple dynamic obstacles, this algorithm often fails to accurately estimate the motion state of these obstacles. Consequently, vehicles may approach obstacles too closely during collision avoidance, resulting in insufficient safety space and increased danger. Summary of the Invention

[0005] The present invention aims to address the shortcomings of the existing technology by proposing a hybrid path planning method based on Gaussian mixture model (RRT*) and DWA, which can accurately estimate the motion state of dynamic obstacles, thereby enabling vehicles to avoid obstacles while preventing the safe distance between the vehicle and the obstacle from being too small.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] The present invention provides a hybrid path planning method based on Gaussian mixture models, combining RRT* and DWA, characterized by the following steps:

[0008] Step 1: Obtain information about the vehicle's surrounding environment using a lidar sensor and establish a grid map. Select a vertex of the bounding rectangle of the grid map as the origin o, and the two sides adjacent to the origin o are the x-axis and y-axis, respectively, thereby establishing a rectangular coordinate system xoy.

[0009] Set the starting and ending positions of the vehicle in the grid map. When the vehicle starts moving from the starting position to the ending position at time t, acquire the trajectory data Ω={(x j y j ), j=tm, t-m+1,...,t}, (x j y j () represents the location information of surrounding vehicles (car) at time j;

[0010] The starting position (x(t), y(t)) of the vehicle at time t is taken as the current position of the vehicle;

[0011] Step 2: Use the RRT* algorithm to plan the global path η = {(x g y g ), g = 1, 2, ..., p}, where (x g y g ) represents the coordinates of the g-th global path point; p represents the total number of global path points.

[0012] Step 3: Establish a Gaussian mixture model based on the location information Ω of surrounding vehicles to obtain the conditional mean of the location information of surrounding vehicles. Conditional covariance

[0013] Step 3.1: Use equation (1) to obtain the joint probability density function p(T, X) of the projection set X = {x(tm), x(t-m+1), ..., x(t-1)} of the surrounding vehicle car position information in the x-axis direction at historical time T = {tm, t-m+1, ..., t-1}, where x(t-1) represents the projection value of the surrounding vehicle car position information in the x-axis direction at time t-1:

[0014]

[0015] In equation (1): K represents the number of Gaussian models, α k Let be the weight coefficients of the k-th Gaussian model, and u k Let be the mean matrix of the k-th Gaussian model, and u k，T Let u represent the mean of the k-th Gaussian model over historical time T. k，X Let ∑ represent the mean of the x-axis projection set x of the k-th Gaussian model. k Let be the covariance matrix of the k-th Gaussian model, and ∑ k，TT ∑ represents the variance of historical time T itself. k，TX∑ represents the covariance between historical time T and the projection set X along the x-axis; k，XT ∑ represents the covariance between the x-axis projection set X and the historical time T. k，XX The variance of the projection set X of the location information of surrounding vehicles along the x-axis is represented by N(·), which is a multivariate normal distribution.

[0016] Step 3.2: Use the EM algorithm to process α k u k , ∑ k Perform iterative calculations to obtain the optimal parameters. in, This represents the optimal weight coefficients of the k-th Gaussian model. Let represent the optimal mean matrix of the k-th Gaussian model, where Let represent the optimal mean of the k-th Gaussian model at historical time T. The optimal mean of the projection set X of the k-th Gaussian model along the x-axis. Let represent the optimal covariance matrix of the k-th Gaussian model; where, This represents the optimal variance of the historical time period T itself. This represents the optimal covariance between historical time T and the projection set X along the x-axis. This represents the optimal covariance between the projection set X along the x-axis and the historical time T. This represents the optimal variance of the projection set X of the surrounding vehicle car location information onto the x-axis direction.

[0017] Step 3.3: Use equation (2) to obtain the conditional distribution of the projection set X of the surrounding vehicle car location information in the x-axis direction at historical time T.

[0018]

[0019] In equation (2), The mixture weights of the k-th Gaussian model are... The conditional mean of the projection set X of the location information of surrounding vehicles along the x-axis. The conditional covariance of the projection set X of the location information of surrounding vehicles along the x-axis;

[0020] Step 3.4: Following the process in Steps 3.1-3.3, obtain the conditional mean of the projection set Y = {y(tm), y(t-m+1), ..., y(t-1)} of the surrounding vehicle location information along the y-axis. Conditional covariance Where y(t-1) represents the projection value of the car position information of surrounding vehicles in the y-axis direction at time t-1;

[0021] Step 4: Use conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information from time tm to time t in the x-axis direction are sampled to generate a new projection set ξ of the surrounding vehicle car position information in the x-axis direction. x ={x(tm),x(t-m+1),…,x(t-1)};

[0022] Using conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information in the y-axis direction from time tm to time t are sampled to generate a new projection set ξ of the surrounding vehicle car position information in the y-axis direction. y ={y(tm),y(t-m+1),...y,(t-1)};

[0023] For the projection set ξ x and ξ y A Gaussian process regression model is established to predict the projection set of the new location information of surrounding vehicles (cars) along the x-axis from time t+1 to time t+n. Projection set along the y-axis Where x(t+n) represents the projection value of the surrounding vehicle car position information to be predicted at time t+n in the x-axis direction; y(t+n) represents the projection value of the surrounding vehicle car position information to be predicted at time t+n in the y-axis direction.

[0024] Step 5: Select the coordinates of the global path points (excluding the starting path point) that are closest to the vehicle's position at time t as the sub-target coordinates (x) of the dynamic window. sub y sub );

[0025] Based on sub-target point (x) sub y sub The projection set of the new location information of surrounding vehicles (cars) in the x-axis direction. Projection set along the y-axis A cost function J for a dynamic window is constructed, thereby using the dynamic window method to generate the distance from the vehicle at its current position (x(t), y(t)) to the sub-target point (x). sub y sub The driving route;

[0026] Step 6: When predicting the location information of surrounding vehicles at time t+1, project the predicted location information of surrounding vehicles at time t onto the x-axis. and y-axis projection set The first data is added to the surrounding vehicle car location information Ω, and the last path point in the driving path is taken as the current position of the vehicle. Then, the process returns to step 3 and executes sequentially until the current position is the destination position, thereby obtaining the path of the vehicle from the starting position to the destination position, so as to complete the hybrid path planning of RRT* and DWA.

[0027] The characteristic of the hybrid path planning method based on Gaussian mixture model and DWA described in this invention is that step 4 is obtained through the following process.

[0028] Step 4.1: Establish a Gaussian process regression model using equation (3):

[0029] ξ x (T)~GP(m(T), Ker(T, T′)) (3)

[0030] In equation (3), ξ x (T) represents the projection set of historical time T, m(T) is the mean function of historical time T, Ker(T, T′) is the covariance function of historical time T itself; T′ represents the transpose of historical time T, GP represents a Gaussian process that varies with time; ~ indicates obedience;

[0031] Step 4.2: Use equation (4) to establish a new projection set ξ of the surrounding vehicle car's circumferential position information in the x-axis direction. x Projection set of surrounding vehicle car position information in the x-axis direction Joint Gaussian distribution between them:

[0032]

[0033] In equation (4), T * = {t+1, t+2, ..., t+n} represents future times of length n; Ker(T, T) represents the covariance matrix of the historical time T itself, Ker(T) * (T) represents future time T * The covariance matrix between the historical time T and the historical time T, Ker(T, T) * Historical time T and future time T * The covariance matrix between them, Ker(T) * T * ) represents future time T * Its own covariance matrix;

[0034] Step 4.3: Construct the projection set of the surrounding vehicle car location information to be predicted in the x-axis direction using equation (5). Conditional distribution:

[0035]

[0036] In equation (5), T represents * Projection set of surrounding vehicle car location information in the x-axis direction at future time conditional mean, T represents * Projection set of surrounding vehicle car location information in the x-axis direction at future time Conditional covariance;

[0037] Step 4.4, by T * The conditional mean at different times in the future constitutes T * Projection set of surrounding vehicle car location information in the x-axis direction at future time

[0038] In step 5, the cost function J is constructed using equation (6):

[0039] min J=βJ1+γJ2 (6)

[0040] In equation (6), J1 represents the distance between the vehicle and the sub-target point (x). sub y sub The heading angle cost function between the vehicle and the surrounding vehicles is obtained from equation (7), J2 represents the distance cost function between the vehicle and the surrounding vehicles, and is obtained from equation (8), β and γ are two weighting coefficients;

[0041] J1=θ / π (7)

[0042]

[0043] In equation (7), θ represents the distance between the vehicle and the sub-target point (x). sub y sub The difference in heading angle between them;

[0044] In equation (8), d safe，ca r is the threshold for the safe distance between the vehicle and surrounding vehicles; d obs，car The dynamic window method is used to plan the vehicle's position information within a given time period and the future time T. * The Euclidean distance between the positions of surrounding vehicles (cars) at the same time; d obs_min The dynamic window method is used to plan the vehicle's position information within the specified time period and its position information at future time T. * The minimum Euclidean distance between the position information of surrounding vehicles at the same time; d cost Here, λ represents the distance penalty value; CAR denotes the set of all surrounding vehicles, λ is a coefficient, and we have:

[0045] λ=d ego_sub / rsenxor (9)

[0046] In equation (9), r sensor d represents the detection radius of the vehicle's lidar sensor. ego_sub Represents the relationship between the vehicle and its sub-target (x) sub y sub The Euclidean distance between them.

[0047] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing any of the RRT* and DWA hybrid path planning methods, and the processor is configured to execute the program stored in the memory.

[0048] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of any of the RRT* and DWA hybrid path planning methods.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] 1. This invention proposes using a Gaussian mixture model to represent the complex motion trajectories of surrounding vehicles. It generates new data using conditional mean and conditional covariance, better covering the dynamic range of the original data and capturing potential dynamics not covered by the original data. This helps reduce the risk of overfitting when the Gaussian process regression model predicts the trajectories of surrounding vehicles, thereby achieving accurate estimation of vehicle motion states.

[0051] 2. This invention proposes to improve the cost function of the dynamic window method by using the predicted values ​​of surrounding vehicle trajectories, thereby overcoming the problem of the vehicle being too close to dynamic obstacles during obstacle avoidance, improving the vehicle's safety level and reducing the risk of collision. Attached Figure Description

[0052] Figure 1 This is a flowchart of the hybrid path planning method based on Gaussian mixture model and DWA of the present invention. Detailed Implementation

[0053] In this embodiment, as Figure 1 As shown, a hybrid path planning method based on Gaussian mixture models (RRT* and DWA) includes the following steps:

[0054] Step 1: Acquire environmental information around the vehicle using a lidar sensor, preprocess the collected point cloud data including noise removal, filtering, and coordinate transformation, and establish a grid map. Select one vertex of the bounding rectangle of the grid map as the origin o, and the two sides adjacent to the origin o are the x-axis and y-axis, respectively, thereby establishing a rectangular coordinate system xoy.

[0055] Set the starting and ending positions of the vehicle in the grid map. When the vehicle starts moving from the starting position to the ending position at time t, acquire the trajectory data Ω={(x j y j ), j=tm, t-m+1,...,t}, (x j y j () represents the location information of surrounding vehicles (car) at time j;

[0056] The starting position (x(t), y(t)) of the vehicle at time t is taken as the current position of the vehicle;

[0057] Step 2: Use the RRT* algorithm to plan the global path η = {(x g y g ), g = 1, 2, ..., p}, where (x g y g ) represents the coordinates of the g-th global path point; p represents the total number of global path points.

[0058] Step 3: Establish a Gaussian mixture model based on the location information Ω of surrounding vehicles to obtain the conditional mean of the location information of surrounding vehicles. Conditional covariance Because vehicle driving is diverse, such as acceleration, deceleration, and curvilinear driving, a single Gaussian model may not be able to describe all trajectory patterns well. Using a Gaussian mixture model can describe different trajectory patterns by combining multiple Gaussian distributions, which helps to handle noise and uncertainty in the data, thereby better fitting vehicle trajectory data.

[0059] Step 3.1: Use equation (1) to obtain the joint probability density function p(T, X) of the projection set X = {x(tm), x(t-m+1), ..., x(t-1)} of the surrounding vehicle car position information in the x-axis direction at historical time T = {tm, t-m+1, ..., t-1}, where x(t-1) represents the projection value of the surrounding vehicle car position information in the x-axis direction at time t-1:

[0060]

[0061] In equation (1): K represents the number of Gaussian models, α k Let be the weight coefficients of the k-th Gaussian model, and u k Let be the mean matrix of the k-th Gaussian model, and u k，TLet u represent the mean of the k-th Gaussian model over historical time T. k，X Let ∑ represent the mean of the projection set X of the k-th Gaussian model along the x-axis. k Let be the covariance matrix of the k-th Gaussian model, and ∑ k，TT ∑ represents the variance of historical time T itself. k，TX ∑ represents the covariance between historical time T and the projection set X along the x-axis; k，XT ∑ represents the covariance between the x-axis projection set X and the historical time T. k，XX The variance of the projection set X of the location information of surrounding vehicles along the x-axis is represented by N(·), which is a multivariate normal distribution.

[0062] Step 3.2: Use the EM algorithm to process α k u k , ∑ k Perform iterative calculations to obtain the optimal parameters. in, This represents the optimal weight coefficients of the k-th Gaussian model. Let represent the optimal mean matrix of the k-th Gaussian model, where Let represent the optimal mean of the k-th Gaussian model at historical time T. The optimal mean of the projection set X of the k-th Gaussian model along the x-axis. Let represent the optimal covariance matrix of the k-th Gaussian model; where, This represents the optimal variance of the historical time period T itself. This represents the optimal covariance between historical time T and the projection set X along the x-axis. This represents the optimal covariance between the projection set X along the x-axis and the historical time T. This represents the optimal variance of the projection set X of the surrounding vehicle car location information onto the x-axis direction.

[0063] Step 3.3: Use equation (2) to obtain the conditional distribution of the projection set X of the surrounding vehicle car location information in the x-axis direction at historical time T.

[0064]

[0065] In equation (2), The mixture weights of the k-th Gaussian model are... The conditional mean of the projection set X of the location information of surrounding vehicles along the x-axis. The conditional covariance of the projection set X of the location information of surrounding vehicles along the x-axis;

[0066] Step 3.4: Following the process in Steps 3.1-3.3, obtain the conditional mean of the projection set Y = {y(tm), y(t-m+1), ..., y(t-1)} of the surrounding vehicle location information along the y-axis. Conditional covariance Where y(t-1) represents the projection value of the car position information of surrounding vehicles in the y-axis direction at time t-1;

[0067] Step 4: Use conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information from time tm to time t in the x-axis direction are sampled to generate a new projection set ξ of the surrounding vehicle car position information in the x-axis direction. x ={x(tm), x(t-m+1), ..., x(t-1)}; conditional covariance Performing Cholesky decomposition yields the lower triangular matrix L, which is: Where L′ denotes the transpose of the lower triangular matrix L; an m-dimensional standard normal distribution sample vector Z is generated, where the values ​​of each dimension come from independent and identically distributed standard normal distributions N(0, 1); the standard normal distribution sample vector Z is multiplied by the lower triangular matrix L to obtain a Gaussian distribution sample vector ξ. x ; Transform the sample vector ξ using a linear transformation x The sampled data was adjusted to meet the conditional mean and conditional covariance matrix requirements.

[0068] New input data is generated using the conditional mean and conditional covariance derived from the Gaussian mixture model. This new data will have different trends and ranges of change than the original data. It can be used to fill in the gaps in the data and ensure that the model fully covers the dynamics of the data. This enables more accurate predictions over a wider range of data, thereby improving the reliability and applicability of the model.

[0069] Specifically, it is obtained through the following process.

[0070] Step 4.1: Establish a Gaussian process regression model using equation (3):

[0071] ξ x (T)~GP(m(T), Ker(T, T′)) (3)

[0072] In equation (3), ξ x (T) represents the projection set of historical time T, m(T) is the mean function of historical time T, Ker(T, T′) is the covariance function of historical time T itself; T′ represents the transpose of historical time T, GP represents a Gaussian process that varies with time; ~ indicates obedience;

[0073] Step 4.2: Use equation (4) to establish a new projection set ξ of the surrounding vehicle car's circumferential position information in the x-axis direction. x Projection set of surrounding vehicle car position information in the x-axis direction Joint Gaussian distribution between them:

[0074]

[0075] In equation (4), T * = {t+1, t+2, ..., t+n} represents future times of length n; Ker(T, T) represents the covariance matrix of the historical time T itself, Ker(T) * (T) represents future time T * The covariance matrix between the historical time T and the historical time T, Ker(T, T) * Historical time T and future time T * The covariance matrix between them, Ker(T) * T * ) represents future time T * Its own covariance matrix;

[0076] Step 4.3: Construct the projection set of the surrounding vehicle car location information to be predicted in the x-axis direction using equation (5). Conditional distribution:

[0077]

[0078] In equation (5), T represents * Projection set of surrounding vehicle car location information in the x-axis direction at future time conditional mean, T represents * Projection set of surrounding vehicle car location information in the x-axis direction at future time Conditional covariance;

[0079] Step 4.4, by T * Conditional mean at different times in the future Composition of T * Projection set of surrounding vehicle car location information in the x-axis direction at future time

[0080] Step 4.5: Utilize the conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information in the y-axis direction from time tm to time t are sampled to generate a new projection set ξ of the surrounding vehicle car position information in the y-axis direction. y= {y(tm), y(t-m+1), ..., y(t-1)}; Repeat steps 4.1 to 4.4 to obtain T. * Projection set along the y-axis of future time

[0081] Step 5: Select the coordinates of the global path points (excluding the starting path point) that are closest to the vehicle's position at time t as the sub-target coordinates (x) of the dynamic window. sub y sub );

[0082] Based on sub-target point (x) sub y sub The projection set of the new location information of surrounding vehicles (cars) in the x-axis direction. Projection set along the y-axis A cost function J for a dynamic window is constructed, thereby using the dynamic window method to generate the distance from the vehicle at its current position (x(t), y(t)) to the sub-target point (x). sub y sub The driving route;

[0083] Construct the cost function J using equation (6):

[0084] min J=βJ1+γJ2 (6)

[0085] In equation (6), J1 represents the distance between the vehicle and the sub-target point (x). sub y sub The heading angle cost function between the vehicle and the surrounding vehicles is obtained from equation (7), J2 represents the distance cost function between the vehicle and the surrounding vehicles, and is obtained from equation (8), β and γ are two weighting coefficients;

[0086] J1=θ / π (7)

[0087]

[0088] In equation (7), θ represents the distance between the vehicle and the sub-target point (xs). ub ys ub The difference in heading angle between them;

[0089] In equation (8), d sαfe，car The threshold for the safe distance between the vehicle and surrounding vehicles is t; the dynamic window planning time is t. * , and t * If the distance is less than or equal to t+n, then the vehicle needs to be moved from t to t+t. * The location information at each time point is used to calculate d in a one-to-one correspondence with the location information of surrounding vehicles (cars) from time t to t+n. obs，car d obs，carThe dynamic window method is used to plan the vehicle's position information within a given time period and the future time T. * The Euclidean distance between the positions of surrounding vehicles (cars) at the same time; d obs_min The dynamic window method is used to plan the vehicle's position information within the specified time period and its position information at future time T. * The minimum Euclidean distance between the position information of surrounding vehicles at the same time; d cost λ is the distance penalty value; CAR represents the set of all surrounding vehicles; λ is a coefficient that indicates that when the vehicle is extremely close to the sub-target, the distance cost function will not account for too large a proportion, and we have:

[0090] λ=d ego_sub / r se，sor (9)

[0091] In equation (9), r sensor d represents the detection radius of the vehicle's lidar sensor. ego_sub Represents the relationship between the vehicle and its sub-target (x) sub y sub The Euclidean distance between them.

[0092] Step 6: When predicting the location information of surrounding vehicles at time t+1, project the predicted location information of surrounding vehicles at time t onto the x-axis. and y-axis projection set The first data is added to the surrounding vehicle car location information Ω, and the last path point in the driving path is taken as the current position of the vehicle. Then, the process returns to step 3 and executes sequentially until the current position is the destination position, thereby obtaining the path of the vehicle from the starting position to the destination position, so as to complete the hybrid path planning of RRT* and DWA.

[0093] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0094] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A Gaussian mixture model-based approach The hybrid path planning method with DWA is characterized by, The steps include the following: Step 1: Obtain information about the vehicle's surrounding environment using a lidar sensor and establish a grid map. Select a vertex of the bounding rectangle of the grid map as the origin o, and the two sides adjacent to the origin o are the x-axis and y-axis, respectively, thereby establishing a rectangular coordinate system xoy. The starting and ending positions of the vehicle are set in the grid map. When the vehicle starts moving from the starting position to the ending position at time t, the trajectory data of any car in the vicinity from time tm to time t is acquired by the vehicle's sensors. , This represents the location information of surrounding vehicles (car) at time j; The starting position of the vehicle at time t As the current location of the vehicle; Step 2: Use the RRT* algorithm to plan the global path of the vehicle from the starting position to the ending position. ,in, Indicates the first The coordinates of a global path point; p represents the total number of coordinates of global path points; Step 3: Based on the location information of surrounding vehicles (cars) Establish a Gaussian mixture model to obtain the conditional mean of the car location information of surrounding vehicles. , Conditional covariance , ; Step 3.1: Obtain historical time using equation (1) Projection set of surrounding vehicle car position information along the x-axis joint probability density function ,in, This represents the projection of the car position information of surrounding vehicles along the x-axis at time t-1: (1) In equation (1): K represents the number of Gaussian models, Let be the weight coefficients of the k-th Gaussian model, and , Let be the mean matrix of the k-th Gaussian model, and , Let represent the mean of the k-th Gaussian model over historical time T. This represents the projection set of the k-th Gaussian model along the x-axis. The mean, Let be the covariance matrix of the k-th Gaussian model, and , This represents the variance of the historical time period T itself. Represents the projection set of historical time T along the x-axis. Covariance between them; Represents the projection set along the x-axis Covariance with historical time T This represents the projection set of surrounding vehicle car location information along the x-axis. Its own variance It follows a multivariate normal distribution; Step 3.2: Use the EM algorithm to... , , Perform iterative calculations to obtain the optimal parameters. ;in, This represents the optimal weight coefficients of the k-th Gaussian model. Let represent the optimal mean matrix of the k-th Gaussian model, where Let represent the optimal mean of the k-th Gaussian model at historical time T. The projection set of the k-th Gaussian model along the x-axis The optimal mean, Let represent the optimal covariance matrix of the k-th Gaussian model; where, This represents the optimal variance of the historical time period T itself. Represents the projection set of historical time T along the x-axis. The optimal covariance between them Represents the projection set along the x-axis The optimal covariance between the historical time T and the historical time T This represents the projection set of surrounding vehicle car location information along the x-axis. Its own optimal variance; Step 3.3: Use equation (2) to obtain the historical time. Projection set of surrounding vehicle car position information along the x-axis Conditional distribution : (2) In equation (2), The mixture weights of the k-th Gaussian model are... Projection set of car position information of surrounding vehicles in the x-axis direction conditional mean, Projection set of car position information of surrounding vehicles in the x-axis direction Its own conditional covariance; Step 3.4: Following the process in steps 3.1-3.3, obtain the projection set of the surrounding vehicle car position information along the y-axis. conditional mean Conditional covariance ;in, This represents the projection of the car position information of surrounding vehicles along the y-axis at time t-1; Step 4: Use conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information from time tm to time t in the x-axis direction are sampled to generate a new projection set of the surrounding vehicle car position information in the x-axis direction. ; Using conditional mean Conditional covariance The projection values ​​of the surrounding vehicle car position information from time tm to time t in the y-axis direction are sampled to generate a new projection set of the surrounding vehicle car position information in the y-axis direction. ; For projection set and A Gaussian process regression model is established to predict the projection set of the new location information of surrounding vehicles (cars) along the x-axis from time t+1 to time t+n. Projection set along the y-axis ;in, express Projection of the location information of surrounding vehicles (cars) to be predicted at any given time onto the x-axis; express The projection of the location information of surrounding vehicles (cars) to be predicted at any given time onto the y-axis. Step 5: Select the global path The coordinates of the vehicle's position at time t among all global path points other than the starting path point are used as the sub-target coordinates of the dynamic window. ; Based on sub-target points The projection set of the new location information of surrounding vehicles (cars) in the x-axis direction. Projection set along the y-axis The cost function J of the dynamic window is constructed, thereby using the dynamic window method to generate the vehicle's current position. Towards the target point The driving route; Step 6: When predicting the location information of surrounding vehicles at time t+1, project the predicted location information of surrounding vehicles at time t onto the x-axis. and y-axis projection set The first data is added to the surrounding vehicle car location information. In the process, the last path point in the driving path is taken as the current position of the vehicle, and then the process returns to step 3 and executes sequentially until the current position is the destination position, so as to obtain the path of the vehicle from the starting position to the destination position, thus completing the hybrid path planning of RRT* and DWA.

2. The Gaussian mixture model based on claim 1 The hybrid path planning method with DWA is characterized by, Step 4 is obtained through the following process: : Step 4.1: Establish a Gaussian process regression model using equation (3): (3) In equation (3), Represents the projection set of historical time T. Let T be the mean function of the historical time period. The covariance function of the historical time T itself; This represents the transpose of historical time T. This represents a Gaussian process that varies over time. ~ indicates obedience; Step 4.2: Use equation (4) to establish a new projection set of the perimeter position information of the vehicle car in the x-axis direction. Projection set of surrounding vehicle car position information in the x-axis direction Joint Gaussian distribution between them: (4) In equation (4), Let n be the future time. This represents the covariance matrix of historical time T itself. Indicates future time The covariance matrix between the historical time T and the historical time T Historical time T and future time The covariance matrix between them Indicates future time Its own covariance matrix; Step 4.3: Construct the projection set of the surrounding vehicle car location information to be predicted in the x-axis direction using equation (5). Conditional distribution: (5) In equation (5), express Projection set of surrounding vehicle car location information in the x-axis direction at future time conditional mean, express Projection set of surrounding vehicle car location information in the x-axis direction at future time Conditional covariance; Step 4.4, from Composition of conditional means at different times in the future Projection set of surrounding vehicle car location information in the x-axis direction at future time .

3. A Gaussian mixture model based on claim 1 The hybrid path planning method with DWA is characterized by, In step 5, the cost function J is constructed using equation (6): (6) In equation (6), For the vehicle and the sub-target point The heading angle cost function is obtained from equation (7). Let represent the distance cost function between the vehicle and surrounding vehicles, and obtain it from equation (8). There are two weighting coefficients; (7) (8) In equation (7), For the vehicle and the target point The difference in heading angle between them; In equation (8), The threshold for the safe distance between the vehicle and surrounding vehicles; The dynamic window method is used to plan the vehicle's position information within a given time period and its future position information. The Euclidean distance between the positions of surrounding vehicles (cars) at the same time; To plan the vehicle's position information within the current time frame and its position in the future time frame using the dynamic window method The minimum Euclidean distance between the position information of surrounding vehicles at the same time; This is the distance penalty value; It represents the collection of all surrounding vehicles. Let be the coefficient, and we have: (9) In equation (9), The detection radius of the vehicle's lidar sensor; Indicates the vehicle and its sub-targets The Euclidean distance between them.

4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store information that supports the processor in executing any one of claims 1-3. A program that combines DWA with a hybrid path planning method, wherein the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by a processor to perform any one of the claims 1-3. The steps of the hybrid path planning method with DWA.

Images