A method for continuous planning of dynamic obstacle avoidance path in intelligent vehicle racing

Abstract

This invention relates to the field of obstacle avoidance path planning technology and discloses a continuous planning method for dynamic obstacle avoidance paths in intelligent vehicle racing, including the following steps: S1, acquiring a pre-stored track map and constructing a track reference coordinate system; In the field of obstacle avoidance path planning technology, this application predicts the probability distribution of the future position of obstacles using a recursive Bayesian filtering algorithm based on historical obstacle observation data, transforms it into a geometric confidence region through multivariate normal distribution modeling, and forms a confidence occupancy set covering the uncertainty of obstacles; combining the intelligent vehicle's shape envelope sequence and the track boundary, a spatiotemporal safe passage is generated through set operations; without relying on a single predicted trajectory, by characterizing the uncertainty of obstacle movement, even if there is a deviation between the actual position of the obstacle and the prediction, the intelligent vehicle can still drive within the safe passage, effectively alleviating the problems of planning failure and vehicle-obstacle conflict caused by deterministic prediction of the actual position of obstacles in the prior art.

Description

Technical Field

[0001] This invention relates to the field of obstacle avoidance path planning technology, and in particular to a method for continuous planning of dynamic obstacle avoidance paths in intelligent vehicle racing. Background Technology

[0002] In the field of intelligent vehicle racing, the ability to dynamically avoid obstacles on the track is a core factor affecting the driving safety and racing performance of intelligent vehicles. The key lies in the accurate prediction of dynamic obstacles and robust path planning.

[0003] On the one hand, existing technologies for predicting the trajectory of dynamic obstacles generally rely on simple deterministic models, such as uniform velocity extrapolation models and uniform acceleration extrapolation models. These models can only linearly predict the future trajectory based on the obstacle's historical motion state, completely ignoring the possible speed changes, temporary stops, and nonlinear turning movements that obstacles may experience on intelligent vehicle racing tracks. Due to the lack of an effective characterization of the uncertainty of obstacle motion, when the actual motion pattern of the obstacle deviates from the linear law preset by the model, the prediction results cannot accurately reflect the actual position of the obstacle at future moments, thus causing subsequent obstacle avoidance planning based on the prediction results to lose its reference value and leading to planning failure problems.

[0004] On the other hand, existing obstacle avoidance planning schemes are usually based on a single deterministic predicted trajectory for path optimization design. Once the actual position of the obstacle deviates from the predicted trajectory, the intelligent vehicle may encounter spatial conflicts with the obstacle when traveling along the planned path. At the same time, due to the lack of modeling capabilities for obstacle uncertainties, existing technologies often adopt overly conservative obstacle avoidance strategies to avoid collision risks, such as increasing detour distances and reducing driving speeds. This results in the intelligent vehicle being unable to travel along a near-optimal path, thus limiting its overall racing performance. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention proposes a continuous planning method for dynamic obstacle avoidance paths in intelligent vehicle racing.

[0006] This invention proposes a continuous path planning method for dynamic obstacle avoidance in intelligent vehicle racing, comprising the following steps:

[0007] S1. Obtain the pre-stored track map and construct the track reference coordinate system;

[0008] S2. Construct the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track's reference coordinate system;

[0009] S3. Collect track geometry data and dynamic obstacle data through sensors to generate an observation set;

[0010] S4. Generate a confidence occupancy set based on the observation set;

[0011] S5. Generate a spatiotemporal secure channel based on the confidence occupancy set;

[0012] S6. Generate the driving path of the intelligent vehicle within the spatiotemporal safety channel.

[0013] Preferably, in S1, a pre-stored track map is obtained, and a track reference coordinate system is constructed as follows:

[0014] Based on the pre-stored track map, the track reference coordinate system is established with the midpoint of the track starting line as the origin, the tangent direction along the track design center line as the vertical coordinate axis, and the horizontal coordinate axis perpendicular to the vertical coordinate axis and pointing to the inside of the track.

[0015] For clarification, the right-hand rule refers to the right-hand rule, where the thumb points in the positive X-axis direction, the index finger points in the positive Y-axis direction, and the middle finger points in the positive Z-axis direction perpendicular to the ground.

[0016] Preferably, in S2, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track reference coordinate system is constructed as follows:

[0017] Acquire GNSS / IMU fusion data from the intelligent vehicle. The GNSS / IMU fusion data includes the intelligent vehicle's positioning data and motion status data. The motion status data includes the intelligent vehicle's real-time heading angle and speed.

[0018] By using positioning data, the position coordinates of the intelligent vehicle in the track reference coordinate system can be obtained;

[0019] By using the real-time heading angle of the intelligent vehicle based on its motion state data, the angle between the longitudinal coordinate axis of the intelligent vehicle's local coordinate system and the track's reference coordinate system can be obtained.

[0020] Based on the position coordinates of the intelligent vehicle in the track reference coordinate system and the angle between the vertical coordinate axis of the intelligent vehicle's local coordinate system and the track reference coordinate system, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track reference coordinate system is obtained through the Euclidean transformation algorithm.

[0021] For clarification: the intelligent vehicle's local coordinate system refers to a coordinate system with the intelligent vehicle's center of mass as the origin and the vehicle's heading as the X-axis; the Euclidean transformation algorithm is an existing algorithm used to implement combined translation and rotation operations.

[0022] Preferably, in S3, track geometry data and dynamic obstacle data are collected by sensors to generate an observation set, as follows:

[0023] The track geometry data and dynamic obstacle data are collected through sensors;

[0024] Sensors include LiDAR, cameras, and millimeter-wave radar; track geometry data includes track boundary data and track design centerline curvature data;

[0025] The first perception dataset is obtained by fusing GNSS / IMU fusion data, track geometry data, and dynamic obstacle data using multi-sensor data fusion technology.

[0026] The first perception dataset is transformed to the track reference coordinate system using a coordinate transformation matrix to obtain the second perception dataset. The second perception dataset is then processed by Kalman filtering to obtain the third perception dataset.

[0027] Based on the third perception dataset, historical observation data of the intelligent vehicle reference centerline, track boundary, and obstacles are generated, and the historical observation data of the intelligent vehicle reference centerline, track boundary, and obstacles form an observation set.

[0028] Preferably, in S3, based on the third perception dataset, historical observation data of the intelligent vehicle's reference centerline, track boundaries, and obstacles are generated as follows;

[0029] Based on the third perception dataset;

[0030] The intelligent vehicle reference centerline is generated by fitting the track geometry data in the third perception dataset with GNSS / IMU fusion data.

[0031] Extract the track boundary data from the track geometry data in the third perception dataset as the track boundary;

[0032] The dynamic obstacle data in the third perception dataset is sorted by timestamp to form historical obstacle observation data.

[0033] Preferably, in S4, a confidence occupancy set is generated based on the observation set, as follows:

[0034] Based on the historical obstacle observation data in the observation set, the recursive Bayesian filtering algorithm is used to predict the probability distribution of the position of the dynamic obstacle at each time point in the future unit of time.

[0035] The probability distribution of position at all times within a unit of time in the future forms a set of position probability distributions, which is used as the probability occupied area.

[0036] Based on the fact that the probability distribution of the position of a dynamic obstacle at each time step within a future unit of time is modeled using a multivariate normal distribution, the multivariate normal distribution model S of the probability distribution of the position at each time step is obtained. k k = 1, 2, ..., T1, where T1 is the total number of moments in a future unit of time;

[0037] Multivariate normal distribution model S k This includes the position mean vector and the 2×2 position covariance matrix at the corresponding time point;

[0038] Based on the preset confidence intervals and the corresponding chi-square distribution critical values, the multivariate normal distribution model S at each time step is calculated using the confidence interval calculation method of the multivariate normal distribution. k Transform into a geometric confidence region;

[0039] The geometrically defined confidence regions at all times form a confidence occupancy set;

[0040] For clarification: the future unit time can be set to 10ms; the preset confidence interval can be set to 95%.

[0041] The geometric confidence region can be understood as follows: taking the position mean vector at the corresponding time as the center, the 2×2 position covariance matrix is ​​decomposed using analytical methods to obtain two eigenvalues ​​λ1 and λ2, where λ1≥λ2. The elliptical region obtained by multiplying λ1 by the chi-square distribution critical value is taken as the major semi-axis, and the square root of λ2 by the chi-square distribution critical value is taken as the minor semi-axis. If λ1=λ2, a circular region is formed as the geometric confidence region, where the center of the circle is the position mean vector at the corresponding time, and the radius is λ1 or λ2 multiplied by the chi-square distribution critical value.

[0042] The method for calculating the confidence interval of a multivariate normal distribution is existing technology. The method for calculating the confidence interval of a multivariate normal distribution can be found in Section 5.3 of the existing public literature "Probabilistic Robots" (by Sebastian Thrun), which can be directly reproduced by those skilled in the art.

[0043] Preferably, in S5, a spatiotemporal secure channel is generated based on the confidence occupancy set, as follows:

[0044] Obtain the geometric shape parameters of the intelligent vehicle in the track reference coordinate system;

[0045] Based on the GNSS / IMU fusion data in the third perception dataset, the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time unit in the future are generated by machine learning algorithm.

[0046] Based on the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time in the future, the shape envelope region of the intelligent vehicle at each time in the future is generated in the track reference coordinate system through the geometric shape parameters of the intelligent vehicle, thus forming a vehicle shape envelope sequence.

[0047] The vehicle's outer shape envelope is used to characterize the spatial range occupied by the intelligent vehicle at a certain moment in the track reference coordinate system;

[0048] By performing set operations on the confidence occupancy set and the vehicle shape envelope sequence, the repulsion region of dynamic obstacles to the intelligent vehicle in the future per unit time is obtained.

[0049] As an explanation: performing set operations on the confidence occupancy set and the vehicle shape envelope sequence means that for each moment, the geometric confidence region at that moment is merged with the shape envelope region of the intelligent vehicle at the corresponding moment to obtain the joint restricted area at that moment; sorting the joint restricted areas at all moments according to the time sequence is the area where dynamic obstacles will exclude the intelligent vehicle in the future per unit time. This design is because the current planned trajectory of the intelligent vehicle cannot enter the space that it will occupy at some point in the future. For example, if the current planned trajectory is exactly the position that the vehicle will occupy in 100ms, the vehicle will overlap with its own trajectory, which is equivalent to scraping itself. In actual driving, this will cause problems such as steering jerking and vehicle deviation.

[0050] Based on the track boundary in the observation set, the repulsion area of ​​dynamic obstacles to intelligent vehicles in future unit time time is combined with the track boundary area to form a spatiotemporal safety channel in the track reference coordinate system.

[0051] The spatiotemporal safety channel is used to characterize the drivable planning space of an intelligent vehicle within the track boundary to meet obstacle avoidance constraints under the condition of obstacle uncertainty;

[0052] As an explanation, based on the track boundary in the observation set, the region where dynamic obstacles will repel the intelligent vehicle in a future unit of time is combined with the track boundary region by a set difference operation. This operation yields the collision-free drivable space at each moment in the future unit of time. By connecting all collision-free drivable spaces in a time sequence, a spatiotemporal safety channel can be formed in the track reference coordinate system.

[0053] Preferably, in S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows:

[0054] Obtain the preset curvature continuity constraint. Within the spatiotemporal safe channel, using the intelligent vehicle reference centerline in the observation set as the path reference, perform local path point search on the intelligent vehicle reference centerline using the A* algorithm to generate a discrete obstacle avoidance path point sequence for future unit time. Use cubic spline interpolation to smoothly fit the discrete obstacle avoidance path point sequence for future unit time to generate a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

[0055] As an explanation, when using the A* algorithm to search for local path points on the reference centerline of an intelligent vehicle, the search range is limited to the spatiotemporal safe channel.

[0056] Preferably, in S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows:

[0057] Obtain the preset curvature continuity constraint, and within the spatiotemporal safe channel, take the intelligent vehicle reference centerline in the observation set as the path reference, and use the A* algorithm to search for local path points on the intelligent vehicle reference centerline to generate multiple sets of discrete obstacle avoidance path point sequences in future unit time as candidate path point sequences.

[0058] Set a preset vehicle speed, obtain the distance between path points in each candidate path point sequence, and obtain the total path time T based on the preset vehicle speed and the distance between path points in the candidate path point sequence.

[0059] Obtain the sum L of the distances between adjacent points in each candidate path point sequence;

[0060] Normalize T and L to obtain T1 and L1;

[0061] For each candidate path point sequence, calculate the comprehensive cost J. The formula for comprehensive cost J is:

[0062] J = α × T1 + β × L1, α + β = 1, where α and β are preset weights;

[0063] The candidate path point sequence with the minimum comprehensive cost J is selected as the chosen path point sequence.

[0064] Cubic spline interpolation is used to smoothly fit the selected path point sequence to generate a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

[0065] The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method proposed in this invention has the following beneficial technical effects:

[0066] 1. This application first acquires a pre-stored track map to construct a track reference coordinate system, providing a unified spatial reference for the data; then, combining GNSS / IMU fusion data from intelligent vehicles, it generates a transformation matrix between the local coordinate system and the track reference coordinate system through obstacle uncertainty modeling and spatiotemporal safety boundary construction, using an Euclidean transformation algorithm to eliminate coordinate deviation; after collecting track geometric data and dynamic obstacle data, it performs multi-sensor fusion, coordinate transformation, and Kalman filtering to obtain an observation set containing the reference centerline, track boundaries, and historical obstacle observation data; finally, based on the historical obstacle observation data, a recursive Bayesian filtering algorithm is used to predict obstacles. The probability distribution of future positions is transformed into a geometric confidence region through multivariate normal distribution modeling, forming a confidence occupancy set that covers the uncertainty of obstacles. Combining the intelligent vehicle's shape envelope sequence and the track boundary, a spatiotemporal safe passage is generated through set operations. This design does not rely on a single predicted trajectory. By characterizing the uncertainty of obstacle movement, even if there is a deviation between the actual position of the obstacle and the prediction, the intelligent vehicle can still drive within the safe passage, improving planning robustness and effectively alleviating the problems of planning failure and vehicle-obstacle conflict caused by deterministic prediction of the actual position of obstacles in existing technologies, thereby improving the reliability of intelligent vehicles in track obstacle avoidance.

[0067] 2. Within the pre-defined constraint of the spatiotemporal safety channel, this application uses the intelligent vehicle reference centerline in the observation set as the path reference, generates a discrete obstacle avoidance path point sequence for future unit time through the A* algorithm, and uses cubic spline interpolation to fit the driving path of the intelligent vehicle, thus alleviating the problems of excessive detours and sudden speed drops caused by the lack of clear safety boundaries in traditional obstacle avoidance.

[0068] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0069] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0070] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0071] like Figure 1 The method for continuous planning of dynamic obstacle avoidance path in intelligent vehicle racing, as shown, includes the following steps:

[0072] S1. Obtain the pre-stored track map and construct the track reference coordinate system;

[0073] In an optional embodiment, in S1, a pre-stored track map is obtained, and a track reference coordinate system is constructed as follows:

[0074] Based on the pre-stored track map, the track reference coordinate system is established with the midpoint of the track starting line as the origin, the tangent direction along the track design center line as the vertical coordinate axis, and the horizontal coordinate axis perpendicular to the vertical coordinate axis and pointing to the inside of the track.

[0075] As an explanation, the right hand refers to the right-hand rule, with the thumb pointing in the positive X-axis direction, the index finger pointing in the positive Y-axis direction, and the middle finger pointing in the positive Z-axis direction perpendicular to the ground.

[0076] S2. Construct the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track's reference coordinate system;

[0077] In an optional embodiment, in S2, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track reference coordinate system is constructed as follows:

[0078] Acquire GNSS / IMU fusion data from the intelligent vehicle. The GNSS / IMU fusion data includes the intelligent vehicle's positioning data and motion status data. The motion status data includes the intelligent vehicle's real-time heading angle (direction of the vehicle) and speed.

[0079] By using positioning data, the position coordinates of the intelligent vehicle in the track reference coordinate system can be obtained;

[0080] By using the real-time heading angle of the intelligent vehicle based on its motion state data, the angle between the longitudinal coordinate axis of the intelligent vehicle's local coordinate system and the track's reference coordinate system can be obtained.

[0081] Based on the position coordinates of the intelligent vehicle in the track reference coordinate system and the angle between the vertical coordinate axis of the intelligent vehicle's local coordinate system and the track reference coordinate system, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track reference coordinate system is obtained through the Euclidean transformation algorithm.

[0082] For clarification: the intelligent vehicle's local coordinate system refers to a coordinate system with the intelligent vehicle's center of mass as the origin and the vehicle's heading as the X-axis; the Euclidean transformation algorithm is an existing algorithm used to implement combined translation and rotation operations;

[0083] S3. Collect track geometry data and dynamic obstacle data through sensors to generate an observation set;

[0084] In an optional embodiment, in S3, track geometry data and dynamic obstacle data are collected by sensors to generate an observation set, as follows:

[0085] The track geometry data and dynamic obstacle data are collected through sensors;

[0086] Sensors include LiDAR, cameras, and millimeter-wave radar; track geometry data includes track boundary data and track design centerline curvature data;

[0087] The first perception dataset is obtained by fusing GNSS / IMU fusion data, track geometry data, and dynamic obstacle data using multi-sensor data fusion technology.

[0088] The first perception dataset is transformed to the track reference coordinate system using a coordinate transformation matrix to obtain the second perception dataset. The second perception dataset is then processed by Kalman filtering to obtain the third perception dataset.

[0089] Based on the third perception dataset, historical observation data of the intelligent vehicle reference centerline, track boundary and obstacles are generated, and the historical observation data of the intelligent vehicle reference centerline, track boundary and obstacles form an observation set;

[0090] In an optional embodiment, in S3, historical observation data of the intelligent vehicle reference centerline, track boundaries, and obstacles are generated based on the third perception dataset, as follows;

[0091] Based on the third perception dataset;

[0092] The intelligent vehicle reference centerline is generated by fitting the track geometry data in the third perception dataset with GNSS / IMU fusion data.

[0093] Extract the track boundary data from the track geometry data in the third perception dataset as the track boundary;

[0094] The dynamic obstacle data in the third perception dataset is sorted by timestamp to form historical obstacle observation data;

[0095] S4. Generate a confidence occupancy set based on the observation set;

[0096] In an optional embodiment, in S4, a confidence occupancy set is generated based on the observation set, as follows:

[0097] Based on the historical obstacle observation data in the observation set, the recursive Bayesian filtering algorithm is used to predict the probability distribution of the position of the dynamic obstacle at each time point in the future unit of time.

[0098] The probability distribution of position at all times within a unit of time in the future forms a set of position probability distributions, which is used as the probability occupied area.

[0099] Based on the fact that the probability distribution of the position of a dynamic obstacle at each time step within a future unit of time is modeled using a multivariate normal distribution, the multivariate normal distribution model S of the probability distribution of the position at each time step is obtained. kk = 1, 2, ..., T1, where T1 is the total number of moments in a future unit of time;

[0100] Multivariate normal distribution model S k This includes the position mean vector and the 2×2 position covariance matrix at the corresponding time point;

[0101] Based on the preset confidence intervals and the corresponding chi-square distribution critical values, the multivariate normal distribution model S at each time step is calculated using the confidence interval calculation method of the multivariate normal distribution. k Transform into a geometric confidence region;

[0102] The geometrically defined confidence regions at all times form a confidence occupancy set;

[0103] For clarification: the future unit time can be set to 10ms; the preset confidence interval can be set to 95%.

[0104] The geometric confidence region can be understood as follows: taking the position mean vector at the corresponding time as the center, the 2×2 position covariance matrix is ​​decomposed using analytical methods to obtain two eigenvalues ​​λ1 and λ2, where λ1≥λ2. The elliptical region obtained by multiplying λ1 by the chi-square distribution critical value is taken as the major semi-axis, and the square root of λ2 by the chi-square distribution critical value is taken as the minor semi-axis. If λ1=λ2, a circular region is formed as the geometric confidence region, where the center of the circle is the position mean vector at the corresponding time, and the radius is λ1 or λ2 multiplied by the chi-square distribution critical value.

[0105] The method for calculating the confidence interval of a multivariate normal distribution is existing technology. The method for calculating the confidence interval of a multivariate normal distribution can be found in Section 5.3 of the existing published literature "Probabilistic Robots" (by Sebastian Thrun). Those skilled in the art can directly reproduce it.

[0106] S5. Generate a spatiotemporal secure channel based on the confidence occupancy set;

[0107] In an optional embodiment, in S5, a spatiotemporal secure channel is generated based on the confidence occupancy set, as follows:

[0108] Obtain the geometric shape parameters of the intelligent vehicle in the track reference coordinate system;

[0109] Based on the GNSS / IMU fusion data in the third perception dataset, the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time unit in the future are generated by machine learning algorithm.

[0110] Based on the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time in the future, the shape envelope region of the intelligent vehicle at each time in the future is generated in the track reference coordinate system through the geometric shape parameters of the intelligent vehicle, thus forming a vehicle shape envelope sequence.

[0111] The vehicle's outer shape envelope is used to characterize the spatial range occupied by the intelligent vehicle at a certain moment in the track reference coordinate system;

[0112] By performing set operations on the confidence occupancy set and the vehicle shape envelope sequence, the repulsion region of dynamic obstacles to the intelligent vehicle in the future per unit time is obtained.

[0113] As an explanation: performing set operations on the confidence occupancy set and the vehicle shape envelope sequence means that for each moment, the geometric confidence region at that moment is merged with the shape envelope region of the intelligent vehicle at the corresponding moment to obtain the joint restricted area at that moment; sorting the joint restricted areas at all moments according to the time sequence is the area where dynamic obstacles will exclude the intelligent vehicle in the future per unit time. This design is because the current planned trajectory of the intelligent vehicle cannot enter the space that it will occupy at some point in the future. For example, if the current planned trajectory is exactly the position that the vehicle will occupy in 100ms, the vehicle will overlap with its own trajectory, which is equivalent to scraping itself. In actual driving, this will cause problems such as steering jerking and vehicle deviation.

[0114] Based on the track boundary in the observation set, the repulsion area of ​​dynamic obstacles to intelligent vehicles in future unit time time is combined with the track boundary area to form a spatiotemporal safety channel in the track reference coordinate system.

[0115] The spatiotemporal safety channel is used to characterize the drivable planning space of an intelligent vehicle within the track boundary to meet obstacle avoidance constraints under the condition of obstacle uncertainty;

[0116] As an explanation, based on the track boundary in the observation set, the region where dynamic obstacles will repel the intelligent vehicle in a future unit of time is combined with the track boundary region by a set difference operation. This operation yields the collision-free drivable space at each moment in the future unit of time. By connecting all collision-free drivable spaces in a time sequence, a spatiotemporal safety channel can be formed in the track reference coordinate system.

[0117] This application first acquires a pre-stored track map to construct a track reference coordinate system, providing a unified spatial reference for the data. Then, combining GNSS / IMU fusion data from intelligent vehicles, it generates a transformation matrix between the local coordinate system and the track reference coordinate system using an Euclidean transformation algorithm after obstacle uncertainty modeling and spatiotemporal safety boundary construction, eliminating coordinate deviations. After collecting track geometric data and dynamic obstacle data, multi-sensor fusion, coordinate transformation, and Kalman filtering are performed to obtain an observation set containing historical obstacle observation data, including the reference centerline, track boundaries, and obstacle observation data. Next, based on the historical obstacle observation data, a recursive Bayesian filtering algorithm is used to predict obstacle failures. The location probability distribution is modeled using a multivariate normal distribution and transformed into a geometric confidence region, forming a confidence occupancy set that covers the uncertainty of obstacles. Combining the intelligent vehicle's shape envelope sequence with the track boundary, a spatiotemporal safe passage is generated through set operations. This design does not rely on a single predicted trajectory. By characterizing the uncertainty of obstacle movement, even if the actual position of the obstacle deviates from the prediction, the intelligent vehicle can still drive within the safe passage, improving planning robustness and effectively alleviating the planning failure and vehicle-obstacle conflict problems caused by deterministic prediction of the actual position of obstacles in existing technologies, thus improving the reliability of intelligent vehicles in track obstacle avoidance.

[0118] S6. Generate the driving path of the intelligent vehicle within the spatiotemporal safety channel.

[0119] In an optional embodiment, in S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows:

[0120] Obtain the preset curvature continuity constraint. Within the spatiotemporal safe channel, using the intelligent vehicle reference centerline in the observation set as the path reference, perform local path point search on the intelligent vehicle reference centerline using the A* algorithm to generate a discrete obstacle avoidance path point sequence for future unit time. Use cubic spline interpolation to smoothly fit the discrete obstacle avoidance path point sequence for future unit time to generate a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

[0121] As an explanation, when using the A* algorithm to search for local path points on the reference centerline of an intelligent vehicle, the search range is limited to the spatiotemporal safe channel.

[0122] In an optional embodiment, in S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows:

[0123] Obtain the preset curvature continuity constraint, and within the spatiotemporal safe channel, take the intelligent vehicle reference centerline in the observation set as the path reference, and use the A* algorithm to search for local path points on the intelligent vehicle reference centerline to generate multiple sets of discrete obstacle avoidance path point sequences in future unit time as candidate path point sequences.

[0124] Set a preset vehicle speed, obtain the distance between path points in each candidate path point sequence, and obtain the total path time T based on the preset vehicle speed and the distance between path points in the candidate path point sequence.

[0125] Obtain the sum L of the distances between adjacent points in each candidate path point sequence;

[0126] Normalize T and L to obtain T1 and L1;

[0127] For each candidate path point sequence, calculate the comprehensive cost J. The formula for comprehensive cost J is:

[0128] J = α × T1 + β × L1, α + β = 1, where α and β are preset weights;

[0129] The candidate path point sequence with the minimum comprehensive cost J is selected as the chosen path point sequence.

[0130] Cubic spline interpolation is used to smoothly fit the selected path point sequence to generate a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

[0131] Within the pre-defined constraints of the spatiotemporal safety channel, the application uses the intelligent vehicle reference centerline in the observation set as the path benchmark, generates a discrete obstacle avoidance path point sequence for future unit time using the A* algorithm, and uses cubic spline interpolation to fit the intelligent vehicle's driving path, thus alleviating the problems of excessive detours and sudden speed drops caused by the lack of clear safety boundaries in traditional obstacle avoidance.

[0132] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0133] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0134] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0136] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0137] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for continuous path planning for dynamic obstacle avoidance in intelligent vehicle racing, characterized in that, Includes the following steps: S1. Obtain the pre-stored track map and construct the track reference coordinate system; S2. Construct the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track's reference coordinate system; S3. Collect track geometry data and dynamic obstacle data through sensors to generate an observation set; S4. Based on the observation set, generate the confidence occupancy set as follows: Based on the historical obstacle observation data in the observation set, the recursive Bayesian filtering algorithm is used to predict the probability distribution of the position of the dynamic obstacle at each time point in the future unit of time. The probability distribution of position at all times within a unit of time in the future forms a set of position probability distributions, which is used as the probability occupied area. Based on the fact that the probability distribution of the position of a dynamic obstacle at each time step within a future unit of time is modeled using a multivariate normal distribution, the multivariate normal distribution model S of the probability distribution of the position at each time step is obtained. k k = 1, 2, ..., T1, where T1 is the total number of moments in a future unit of time; Multivariate normal distribution model S k This includes the position mean vector and the 2×2 position covariance matrix at the corresponding time point; Based on the preset confidence intervals and the corresponding chi-square distribution critical values, the multivariate normal distribution model S at each time step is calculated using the confidence interval calculation method of the multivariate normal distribution. k Transform into a geometric confidence region; The geometrically defined confidence regions at all times form a confidence occupancy set; S5. Generate a spatiotemporal secure channel based on the confidence occupancy set; S6. Generate the driving path of the intelligent vehicle within the spatiotemporal safety channel.

2. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 1, characterized in that, In S1, the pre-stored track map is retrieved, and the track's baseline coordinate system is constructed as follows: Based on the pre-stored track map, the track reference coordinate system is established with the midpoint of the track starting line as the origin, the tangent direction along the track design center line as the vertical coordinate axis, and the horizontal coordinate axis perpendicular to the vertical coordinate axis and pointing to the inside of the track.

3. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 1, characterized in that, In S2, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track's reference coordinate system is constructed as follows: Acquire GNSS / IMU fusion data from the intelligent vehicle. The GNSS / IMU fusion data includes the intelligent vehicle's positioning data and motion status data. The motion status data includes the intelligent vehicle's real-time heading angle and speed. By using positioning data, the position coordinates of the intelligent vehicle in the track reference coordinate system can be obtained; By using the real-time heading angle of the intelligent vehicle based on its motion state data, the angle between the longitudinal coordinate axis of the intelligent vehicle's local coordinate system and the track's reference coordinate system can be obtained. Based on the position coordinates of the intelligent vehicle in the track reference coordinate system and the angle between the vertical coordinate axis of the intelligent vehicle's local coordinate system and the track reference coordinate system, the coordinate transformation matrix between the intelligent vehicle's local coordinate system and the track reference coordinate system is obtained through Euclidean transformation algorithm.

4. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 3, characterized in that, In S3, track geometry data and dynamic obstacle data are collected through sensors to generate an observation set, as follows: The track geometry data and dynamic obstacle data are collected through sensors; Sensors include LiDAR, cameras, and millimeter-wave radar; track geometry data includes track boundary data and track design centerline curvature data; The first perception dataset is obtained by fusing GNSS / IMU fusion data, track geometry data, and dynamic obstacle data using multi-sensor data fusion technology. The first perception dataset is transformed to the track reference coordinate system using a coordinate transformation matrix to obtain the second perception dataset. The second perception dataset is then processed by Kalman filtering to obtain the third perception dataset. Based on the third perception dataset, historical observation data of the intelligent vehicle reference centerline, track boundary, and obstacles are generated, and the historical observation data of the intelligent vehicle reference centerline, track boundary, and obstacles form an observation set.

5. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 4, characterized in that, In S3, based on the third perception dataset, historical observation data of the intelligent vehicle's reference centerline, track boundaries, and obstacles are generated, as follows; Based on the third perception dataset; The intelligent vehicle reference centerline is generated by fitting the track geometry data in the third perception dataset with GNSS / IMU fusion data. Extract the track boundary data from the track geometry data in the third perception dataset as the track boundary; The dynamic obstacle data in the third perception dataset is sorted by timestamp to form historical obstacle observation data.

6. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 5, characterized in that, In S5, a spatiotemporal secure channel is generated based on the confidence occupancy set, as follows: Obtain the geometric shape parameters of the intelligent vehicle in the track reference coordinate system; Based on the GNSS / IMU fusion data in the third perception dataset, the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time unit in the future are generated by machine learning algorithm. Based on the position and heading angle of the intelligent vehicle in the track reference coordinate system at each time in the future, the shape envelope region of the intelligent vehicle at each time in the future is generated in the track reference coordinate system through the geometric shape parameters of the intelligent vehicle, thus forming a vehicle shape envelope sequence. By performing set operations on the confidence occupancy set and the vehicle shape envelope sequence, the repulsion region of dynamic obstacles to the intelligent vehicle in the future per unit time is obtained. Based on the track boundary in the observation set, the repulsion area of ​​dynamic obstacles to intelligent vehicles per unit time in the future is combined with the track boundary area to form a spatiotemporal safety channel in the track reference coordinate system.

7. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 6, characterized in that, In S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows: By obtaining the preset curvature continuity constraint, within the spatiotemporal safe channel, using the intelligent vehicle reference centerline in the observation set as the path reference, the A* algorithm is used to search for local path points on the intelligent vehicle reference centerline, generating a discrete obstacle avoidance path point sequence for future unit time. Cubic spline interpolation is then used to smoothly fit the discrete obstacle avoidance path point sequence for future unit time, generating a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

8. The intelligent vehicle racing dynamic obstacle avoidance path continuous planning method according to claim 6, characterized in that, In S6, within the spatiotemporal safety channel, the driving path of the intelligent vehicle is generated as follows: Obtain the preset curvature continuity constraint, and within the spatiotemporal safe channel, take the intelligent vehicle reference centerline in the observation set as the path reference, and use the A* algorithm to search for local path points on the intelligent vehicle reference centerline to generate multiple sets of discrete obstacle avoidance path point sequences in future unit time as candidate path point sequences. Set a preset vehicle speed, obtain the distance between path points in each candidate path point sequence, and obtain the total path time T based on the preset vehicle speed and the distance between path points in the candidate path point sequence. Obtain the sum L of the distances between adjacent points in each candidate path point sequence; Normalize T and L to obtain T1 and L1; For each candidate path point sequence, calculate the comprehensive cost J. The formula for comprehensive cost J is: J = α × T1 + β × L1, α + β = 1, where α and β are preset weights; The candidate path point sequence with the minimum comprehensive cost J is selected as the chosen path point sequence. Cubic spline interpolation is used to smoothly fit the selected path point sequence to generate a smooth path segment with continuous curvature. The smooth path segment with continuous curvature satisfies the preset curvature continuity constraint, thus obtaining the driving path of the intelligent vehicle.

Images