A method for discrete boundary proportional point tracking control of intelligent vehicles

By using the discrete boundary proportional point tracking control method for intelligent vehicles, and utilizing lidar and point cloud data processing, continuous and interference-resistant target points are generated, solving the steering stability problem of intelligent vehicles in discrete boundary channels and achieving efficient path planning and motion control.

CN123104673APending Publication Date: 2026-11-17SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611333596.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In outdoor low-speed intelligent vehicle application scenarios, when the vehicle is running in a channel composed of discrete boundary markers such as cones, road signs, tape, and baffles, the existing technology is prone to sudden changes in steering commands due to changes in the target point or deviation of the center line, and the anti-interference capability is insufficient.

Method used

The system uses LiDAR to acquire point cloud data around the vehicle. Through coordinate transformation, effective area filtering, clustering, and deduplication, it separates the inner and outer boundary point sets and generates candidate target points on local cross-sections according to a preset ratio. Combining the aiming distance and target point sorting, it outputs speed and steering commands.

Benefits of technology

It achieves good continuity of target points, strong anti-interference ability, adaptability to discrete boundaries, real-time performance and engineering feasibility, security degradation capability, and strong parameter adjustability, and is suitable for temporary channels composed of various discrete or semi-discrete boundaries.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104673A_ABST
    Figure CN123104673A_ABST
Patent Text Reader

Abstract

This invention relates to a discrete boundary proportional point tracking control method for intelligent vehicles, belonging to the field of intelligent vehicle path planning and motion control. The method includes: acquiring lidar data and performing coordinate transformation; setting a rectangular effective area, filtering out points outside the rectangular effective area to obtain an effective point set; clustering the effective point set and removing duplicate center points; selecting an inner boundary seed point, and iteratively expanding the inner boundary point set starting from the seed point, with other points assigned to the outer boundary point set; segmenting the inner boundary point set according to direction, retaining points within the lower envelope of each segment, and deleting outliers on the outer side; generating candidate target points; performing pre-aiming distance filtering and distance sorting on the candidate target points to obtain the current target point; calculating the direction error based on the current target point; and outputting linear velocity, angular velocity, or equivalent front wheel steering angle commands. This invention has advantages such as adaptability to discrete boundaries, good target point continuity, strong anti-interference capability, good real-time performance, and good engineering feasibility.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a discrete boundary proportional point tracking control method for intelligent vehicles, belonging to the field of intelligent vehicle path planning and motion control technology. Background Technology

[0002] In outdoor low-speed intelligent vehicle application scenarios, vehicles can be used for park inspection, low-speed logistics in campuses or factories, warehousing and transfer, teaching and scientific research training, intelligent vehicle competition verification, agricultural or industrial site inspection, and automatic driving on temporarily closed roads.

[0003] In the above scenarios, vehicles often operate within lanes composed of cones, road signs, tape, barriers, or other discrete boundary markers. Unlike continuous lane lines, discrete boundary points typically exhibit characteristics such as uneven spacing, local absence, inconsistent numbers of left and right boundaries, and interference from adjacent lane markers. If the nearest obstacle point or a single-sided boundary point is directly selected as the target point, the vehicle is prone to target point jumps at curves or boundary transitions, leading to abrupt steering commands. If a global centerline is directly fitted to the left and right boundaries, centerline offset is likely to occur when boundary points are missing or there are many interfering points. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a discrete boundary proportional point tracking control method for intelligent vehicles. It utilizes lidar to acquire point cloud data around the vehicle, performs coordinate transformation, effective region filtering, clustering, and deduplication on the point cloud to separate inner and outer boundary point sets. Then, it performs nearest neighbor pairing on the inner and outer boundary points, generating candidate target points on local cross-sections according to a preset ratio. Finally, it combines pre-aiming distance, target point sorting, and control state machine output speed and steering commands. This method possesses advantages such as adaptability to discrete boundaries, good target point continuity, strong anti-interference capability, good real-time performance, and good engineering feasibility.

[0005] The present invention adopts the following technical solution:

[0006] A discrete boundary proportional point tracking control method for intelligent vehicles includes the following steps:

[0007] S1 collects LiDAR data around the intelligent vehicle body and performs coordinate transformation to convert the LiDAR data into two-dimensional points in the vehicle coordinate system.

[0008] S2, set a rectangular effective area based on the intelligent vehicle body length, vehicle width B, channel width, radar installation position and vehicle speed, filter out points outside the rectangular effective area to obtain the effective point set;

[0009] S3, cluster the effective point set to obtain several candidate boundary center points, and remove duplicate center points to avoid generating multiple duplicate center points for the same marker;

[0010] S4. Select the inner boundary seed point based on the lateral position and distance relationship, and iteratively expand the inner boundary point set starting from the seed point. Candidate boundary center points that are not included in the inner boundary point set are included in the outer boundary point set.

[0011] S5, divide the inner boundary point set into segments according to direction, retain the points of the lower envelope in each segment, and delete the outliers on the outer side;

[0012] S6. For each inner boundary point, find the nearest outer boundary point to form a local transverse line segment, and generate candidate target points according to the proportion.

[0013] S7 performs pre-aiming distance filtering and distance sorting on candidate target points to obtain the current target point. Based on the current target point, it calculates the direction error and combines the speed gear and state machine outputs linear velocity, angular velocity, or equivalent front wheel rotation angle commands.

[0014] Preferably, the coordinate transformation process in step S1 is as follows:

[0015] x_i =(r_i + Δr) cos(α_i + Δα) + Δx

[0016] y_i =(r_i + Δr) sin(α_i + Δα) + Δy

[0017] Where x_i is the X coordinate in the vehicle coordinate system, y_i is the Y coordinate in the vehicle coordinate system, the X-axis points in the direction of vehicle movement, the Y-axis points to the left side of the vehicle and is perpendicular to the X-axis, r_i is the ranging value of the i-th laser beam, α_i is the scanning angle of the i-th laser beam, Δr is the distance compensation amount, Δα is the angle compensation amount, and Δx and Δy are the installation offsets of the lidar relative to the origin of the vehicle coordinate system in the front-back direction and the left-right direction.

[0018] Preferably, in step S2, the front distance of the intelligent vehicle body is 2.0m to 2.5m, the rear redundancy distance is 0.2m to 0.5m, and the left and right widths are B / 2 plus a safety margin of 0.3m to 0.6m to determine the rectangular effective area;

[0019] Valid points are retained as follows, while points outside the valid rectangular region are filtered out:

[0020] x_min ≤ x_i ≤ x_max, y_min ≤ y_i ≤ y_max

[0021] In the formula, x_min, x_max, y_min, and y_max are the minimum, maximum, minimum, and maximum values ​​of the rectangular effective region along the X-axis, Y-axis, and Y-axis, respectively.

[0022] Preferably, in step S3, one of KMeans, DBSCAN, and Euclidean clustering is used, and the number of clusters K is 10 to 20.

[0023] The deduplication radius refers to the Euclidean distance threshold between the center points of two candidate boundaries. When the distance between the center points of two candidate boundaries is less than the deduplication radius, they are considered to correspond to the same marker or the same local boundary. The center point that is closer to the intelligent vehicle body and has more points is retained, or the average of the two is used as the merged center point.

[0024] Preferably, in step S4, the lateral position refers to the Y-coordinate of the candidate boundary center point in the vehicle coordinate system. Based on the vehicle's current driving direction or the site's detour direction, the inner boundary is pre-determined to be located on the left or right side of the vehicle. Inner boundary seed points are preferentially selected from the candidate center points closest to the vehicle. Starting from seed point q, the remaining candidate boundary center points are traversed. When the point p to be judged meets the following conditions, the point p to be judged is added to the inner boundary point set:

[0025] (x_p - x_q)^2 + (y_p - y_q)^2 ≤D, (y_p - y_q)^2 ≤T

[0026] Where D is the square of the Euclidean distance threshold, and T is the square of the lateral difference threshold; x_p and y_p represent the forward and lateral coordinates of the point p to be judged in the vehicle coordinate system, respectively, and x_q and y_q represent the forward and lateral coordinates of the seed point q in the vehicle coordinate system, respectively.

[0027] Preferably, in step S5, the inner boundary point set is segmented along the X-axis direction, and multiple intervals Ω_q are divided from x_min to x_max with a fixed segment length Δx. Within each segment interval, the lateral coordinate reference value of the inner boundary point is calculated, and points whose lateral coordinates are not greater than the reference value plus the segment tolerance τ are retained. These points are used to retain the boundary points close to the lower envelope of the lane within each segment. The formula is as follows:

[0028] y_i≤ min(y_j | x_j ∈ Ω_q) + τ

[0029] Where y_i represents the lateral coordinate of the inner boundary point i to be screened in the vehicle coordinate system, y_j represents the lateral coordinate of other inner boundary points in the same segment interval Ω_q, x_j represents the forward coordinate of other inner boundary points in the same segment interval Ω_q, and τ represents the segment tolerance;

[0030] Preferred, obviously erroneous outlier points include: points located on the same side of the inner boundary, points whose local cross-section segment formed with the inner boundary point has a length less than the lower limit of the channel width or greater than the upper limit of the channel width, points whose lateral residual relative to the fitted curve exceeds 0.3m to 0.5m, and points whose X direction is obviously outside the effective area. The channel width refers to the nominal lateral width of the driving channel formed by discrete boundary markers (cones / road signs), that is, the nominal vertical distance from the inner boundary to the outer boundary. The nominal value for outdoor intelligent vehicle competition scenarios is 1.5m to 1.8m.

[0031] The lower limit of the channel width is 70% of the channel width, for example, 1.0m. If it is less than this value, it is judged as an interference point on the same side and no target point is generated. The upper limit of the channel width is 130% of the channel width, for example, 2.2m. If it is greater than this value, it is judged as an interference point in the adjacent lane and no target point is generated.

[0032] Preferably, Δx is between 0.5m and 1.5m;

[0033] τ is taken as 0.2m to 0.5m.

[0034] Preferably, in step S6, the process of generating candidate target points is as follows:

[0035] j = arg min ||I_i - O_j|| 2

[0036] Z_i = I_i + λ(O_j - I_i), 1 / 3 ≤ λ ≤ 2 / 3

[0037] In the formula, I_i represents the i-th inner boundary point, O_j represents the outer boundary point closest to it, and j represents the set of outer boundary points such that ||I_i - O_j|| 2 The index of the outer boundary point that yields the minimum value; λ represents the proportion, ranging from 1 / 3 to 2 / 3; if the distance between I_i and O_j is less than the lower limit of the channel width or greater than the upper limit of the channel width, the local transverse segment is judged as abnormal and does not participate in the generation of candidate target points.

[0038] Preferably, in step S7, after the candidate target points are generated, points with X coordinates less than the minimum aiming distance L_min are first deleted, then the distance between each candidate target point and the vehicle origin is calculated, and the points are sorted in ascending order of distance. The Nth nearest target point is selected as the current target point based on the vehicle speed and the curve conditions.

[0039] L_min is taken as 0.6m to 1.2m.

[0040] Preferably, in step S7, the formula for calculating the direction error e is as follows:

[0041] e = atan2(y_t, x_t)

[0042] Where (x_t, y_t) are the coordinates of the current target point;

[0043] The formula for calculating the steering control quantity u is as follows:

[0044] u = K_p(v)e + K_i(v)∑e + K_d(v)(e - e_last)

[0045] Where K_p(v) represents the proportional term parameter corresponding to the speed gear v; K_i(v) represents the integral term parameter corresponding to the speed gear v; K_d(v) represents the differential term parameter corresponding to the speed gear v; e_last represents the direction error of the previous control cycle, which is cached by the controller to obtain the direction error e calculated at the previous moment as the e_last of the current cycle. When the system is initialized, e_last is set to 0.

[0046] The linear velocity is determined as follows: Pre-set the normal tracking speed v_base, cornering deceleration speed v_turn, degraded speed v_safe, and stopping speed 0. When |e| is less than the first angle threshold, output v_base; when |e| is between the first and second angle thresholds, output v_turn; when |e| is greater than the second angle threshold, it indicates excessive steering deviation and a risk of the vehicle running off the road, so output the degraded speed v_safe; if the deviation persists for more than 5 frames without convergence, output 0 to trigger stopping protection; when target points are continuously invalid (no valid candidate target points detected for 3-5 consecutive frames), it indicates a failure of the perception system, so output v_safe for slow creep; if invalid for more than 10 frames, output 0 for stopping and waiting.

[0047] The first angle threshold (straight-line tracking threshold) is set to 10°~15°, and the second angle threshold (curve safety threshold) is set to 30°~45°. For example, if the first angle threshold is 12° (about 0.21 rad), driving at full speed on a straight road will be allowed if the value is less than this. If the second angle threshold is 35° (about 0.61 rad), driving at full speed on a straight road will be allowed if the value is greater than this. If the value is greater than this, safety degradation will be triggered.

[0048] The angular velocity ω is determined as follows: ω = clamp(u, -ω_max, ω_max), where clamp() represents the clamping function, which restricts the input value to the closed interval [-ω_max, ω_max]. If the input is greater than ω_max, the output is ω_max; if the input is less than -ω_max, the output is -ω_max; otherwise, the original value is output.

[0049] ω_max represents the maximum permissible angular velocity of the vehicle, which can be determined based on the chassis steering performance and the maximum vehicle speed. The example value for an outdoor low-speed intelligent vehicle is 1.0 rad / s (approximately 57° / s), and the range can be 0.8 rad / s to 1.5 rad / s.

[0050] The equivalent front wheel steering angle δ is determined as follows: when v is greater than zero, δ = atan(L_w·ω / v), where L_w is the wheelbase, and the wheelbase of the small racing intelligent car is 0.3m~0.5m, for example 0.4m; v is the actual linear velocity value output in the current control cycle, which is output by the speed state machine according to the cornering conditions in v_base / v_turn / v_safe, and is not a fixed gear.

[0051] For any details not covered in this invention, please refer to the prior art.

[0052] The beneficial effects of this invention are as follows:

[0053] 1. Good continuity of target points: By pairing the nearest neighbors of the inner and outer boundaries and generating proportional points, the target points are restricted to the local cross-section segment; the present invention generates tracking target points according to the channel cross-section relationship, which can reduce the jumps caused by the single-sided nearest point method when switching boundary points.

[0054] 2. Strong anti-interference capability: Through rectangular region filtering, cluster center deduplication, segmented envelope screening, inner curve fitting and outer outlier correction, it can reduce the impact of vehicle reflection, ground clutter and adjacent channel markers on target point generation.

[0055] 3. Adaptable to discrete boundaries: This invention does not rely on continuous lane lines and existing planned paths, nor does it require the number of left and right boundary points to be exactly the same. It is suitable for temporary passages composed of discrete or semi-discrete boundaries such as cones, road signs, tape, and temporary barriers.

[0056] 4. Good real-time performance and engineering feasibility: Perception computing and control computing are separated into independent nodes. The perception node is responsible for point cloud filtering, clustering, boundary separation and target point generation, while the control node is responsible for state judgment, speed selection and steering control, which can reduce the impact of point cloud processing time on the control cycle.

[0057] 5. Possesses safety degradation capability: When the sensing node cannot generate a normal scale target point, it does not forcibly output an incorrect target point, but outputs mode flags according to different abnormal working conditions, and the control node executes strategies such as yellow line bend turning, right angle bend handling, right correction, maintaining the previous normal turn, or stopping and waiting accordingly.

[0058] 6. High parameter adjustability: The proportional coefficient λ, rectangular filter range, number of clusters, deduplication radius of the center point, segment length, segment tolerance, minimum aiming distance and PID parameters can all be adjusted according to vehicle size, vehicle speed, channel width and site conditions. Attached Figure Description

[0059] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0060] Figure 1 This is a flowchart of the intelligent vehicle discrete boundary proportional point tracking control method of the present invention;

[0061] Figure 2 This is a schematic diagram of cluster center extraction according to the present invention;

[0062] Figure 3 This is a schematic diagram illustrating the generation of candidate target points according to the present invention;

[0063] Figure 4 A flowchart illustrating the decoupling communication process between sensing nodes and control nodes;

[0064] Figure 5 This is a schematic diagram of target point tracking control and state machine.

[0065] Figure 6 This is a schematic diagram of the recognition results for the sensing nodes. Detailed Implementation

[0066] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. However, this is not the only description; all aspects not described in detail herein are based on conventional techniques in the art.

[0067] Example 1

[0068] A discrete boundary proportional point tracking control method for intelligent vehicles includes the following steps:

[0069] S1 collects LiDAR data around the intelligent vehicle body and performs coordinate transformation to convert the LiDAR data into two-dimensional points in the vehicle coordinate system.

[0070] S2, set a rectangular effective area based on the intelligent vehicle body length, vehicle width B, channel width, radar installation position and vehicle speed, filter out points outside the rectangular effective area to obtain the effective point set;

[0071] S3, cluster the effective point set to obtain several candidate boundary center points, and remove duplicate center points to avoid generating multiple duplicate center points for the same marker;

[0072] S4. Select the inner boundary seed point based on the lateral position and distance relationship, and iteratively expand the inner boundary point set starting from the seed point. Candidate boundary center points that are not included in the inner boundary point set are included in the outer boundary point set.

[0073] S5, divide the inner boundary point set into segments according to direction, retain the points of the lower envelope in each segment, and delete the outliers on the outer side;

[0074] S6. For each inner boundary point, find the nearest outer boundary point to form a local transverse line segment, and generate candidate target points according to the proportion.

[0075] S7 performs pre-aiming distance filtering and distance sorting on candidate target points to obtain the current target point. Based on the current target point, it calculates the direction error and combines the speed gear and state machine outputs linear velocity, angular velocity, or equivalent front wheel rotation angle commands.

[0076] Example 2

[0077] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Example 1, except that in step S1, the intelligent vehicle collects scanning data around the vehicle body by using a two-dimensional lidar installed at the front or top of the vehicle body, and receives the vehicle motion status output by the odometer or positioning module.

[0078] The coordinate transformation process is as follows:

[0079] x_i =(r_i + Δr) cos(α_i + Δα) + Δx

[0080] y_i =(r_i + Δr) sin(α_i + Δα) + Δy

[0081] Where x_i is the X coordinate in the vehicle coordinate system, y_i is the Y coordinate in the vehicle coordinate system, the X-axis points in the direction of vehicle movement, the Y-axis points to the left side of the vehicle and is perpendicular to the X-axis, r_i is the ranging value of the i-th laser beam, α_i is the scanning angle of the i-th laser beam, Δr is the distance compensation amount, Δα is the angle compensation amount, and Δx and Δy are the installation offsets of the lidar relative to the origin of the vehicle coordinate system in the front-back direction and the left-right direction.

[0082] The installation offset can be obtained through actual vehicle measurement or static calibration, that is, measuring the distance between the radar center and the origin of the vehicle coordinate system in the front-rear and left-right directions; in one embodiment, Δx can be 0.05m to 0.30m, Δy can be -0.10m to 0.10m, and if the radar is installed on the center line of the vehicle, Δy can be 0; Δr and Δα can be corrected by a target at a known distance in the calibration site, or can be taken as 0 when the installation error is small.

[0083] Example 3

[0084] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Example 2, differs in that, in step S2, the distance to the front of the intelligent vehicle is 2.0m to 2.5m, the redundant distance to the rear is 0.2m to 0.5m, and the width to the left and right is B / 2 plus a safety margin of 0.3m to 0.6m to determine the rectangular effective area;

[0085] Valid points are retained as follows, while points outside the valid rectangular region are filtered out:

[0086] x_min ≤ x_i ≤ x_max, y_min ≤ y_i ≤ y_max

[0087] In the formula, x_min, x_max, y_min, and y_max are the minimum, maximum, minimum, and maximum values ​​of the rectangular effective region along the X-axis, Y-axis, and Y-axis, respectively.

[0088] Preferably, x_min is -0.2m to -0.5m, x_max is 2.0m to 2.5m, y_min is -1.3m to -1.8m, and y_max is 1.5m to 2.0m. When vehicle speed increases, x_max can be appropriately increased; when the channel width decreases, y_min and y_max can be appropriately narrowed. This step can reduce site-irrelevant points while preserving channel boundary points.

[0089] Example 4

[0090] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Example 3, except that step S3 uses one of KMeans, DBSCAN, and Euclidean clustering, with the number of clusters K ranging from 10 to 20, and K being 15 in this example.

[0091] The deduplication radius refers to the Euclidean distance threshold between the center points of two candidate boundaries. When the distance between the center points of two candidate boundaries is less than the deduplication radius, they are considered to correspond to the same marker or the same local boundary. The center point that is closer to the intelligent vehicle body and has more points is retained, or the average of the two is used as the merged center point.

[0092] The de-duplication radius can be determined based on the cone diameter, marker spacing, and radar angular resolution, and is preferably 0.10m to 0.20m. In this embodiment, it is 0.15m.

[0093] Example 5

[0094] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Embodiment 4, differs in that, in step S4, the lateral position refers to the Y coordinate of the candidate boundary center point in the vehicle coordinate system. The inner boundary is pre-determined to be located on the left or right side of the vehicle based on the vehicle's current driving direction or the site's detour direction. Inner boundary seed points are preferentially selected from candidate center points closest to the vehicle. Starting from seed point q, the remaining candidate boundary center points are traversed. When the point p to be judged meets the following conditions, the point p to be judged is added to the inner boundary point set:

[0095] (x_p - x_q)^2 + (y_p - y_q)^2 ≤D, (y_p - y_q)^2 ≤T

[0096] Where D is the square of the Euclidean distance threshold, and T is the square of the lateral difference threshold; x_p and y_p represent the forward and lateral coordinates of the point p to be judged in the vehicle coordinate system, respectively, and x_q and y_q represent the forward and lateral coordinates of the seed point q in the vehicle coordinate system, respectively. By simultaneously constraining the distance and lateral difference, the probability of mistakenly adding the other side boundary or adjacent channel markers to the inner boundary point set can be reduced.

[0097] In this embodiment, the inner boundary seed point is preferentially selected from the left near-field region of y∈(0,2.0m) and x∈(-1.0m,1.0m), and the candidate center point with the smallest Euclidean distance to the vehicle origin is selected as the seed point. When expanding, the Euclidean distance squared threshold D=1.69 and the lateral difference squared threshold T=1.21 are used, and combined with the front and rear direction constraints, to avoid adding points that are too far behind, boundary points on the other side, or adjacent channel points to the inner boundary by mistake.

[0098] Example 6

[0099] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Example 4, differs in that, in step S5, the inner boundary point set is segmented along the X-axis direction, and multiple intervals Ω_q are divided from x_min to x_max with a fixed segment length Δx. Within each segment interval, the lateral coordinate reference value of the inner boundary points is calculated, and points whose lateral coordinates are not greater than the reference value plus the segment tolerance τ are retained. These points are used to retain boundary points close to the lower envelope of the current lane within each segment, eliminating outliers caused by reflection, occlusion, or adjacent channels. The formula is as follows:

[0100] y_i≤ min(y_j | x_j ∈ Ω_q) + τ

[0101] Where y_i represents the lateral coordinate of the inner boundary point i to be screened in the vehicle coordinate system, y_j represents the lateral coordinate of other inner boundary points in the same segment interval Ω_q, x_j represents the forward coordinate of other inner boundary points in the same segment interval Ω_q, and τ represents the segment tolerance;

[0102] Preferred, obviously erroneous outlier points include: points located on the same side of the inner boundary, points whose local cross-section segment formed with the inner boundary point has a length less than the lower limit of the channel width or greater than the upper limit of the channel width, points whose lateral residual relative to the fitted curve exceeds 0.3m to 0.5m, and points whose X direction is obviously outside the effective area. The channel width refers to the nominal lateral width of the driving channel formed by discrete boundary markers (cones / road signs), that is, the nominal vertical distance from the inner boundary to the outer boundary. The nominal value for outdoor intelligent vehicle competition scenarios is 1.5m to 1.8m.

[0103] The lower limit of the channel width is 70% of the channel width, for example, 1.0m. If it is less than this value, it is judged as an interference point on the same side and no target point is generated. The upper limit of the channel width is 130% of the channel width, for example, 2.2m. If it is greater than this value, it is judged as an interference point in the adjacent lane and no target point is generated.

[0104] Preferably, Δx is between 0.5m and 1.5m;

[0105] τ is taken from 0.2m to 0.5m, preferably 0.3m.

[0106] Example 7

[0107] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Example 6, differs in that, in step S6, as... Figure 3 The process of generating candidate target points is as follows:

[0108] j = arg min ||I_i - O_j|| 2

[0109] Z_i = I_i + λ(O_j - I_i), 1 / 3 ≤ λ ≤ 2 / 3

[0110] In the formula, I_i represents the i-th inner boundary point, O_j represents the outer boundary point closest to it, and j represents the set of outer boundary points such that ||I_i - O_j|| 2 The index of the outer boundary point that yields the minimum value; λ represents the proportion, ranging from 1 / 3 to 2 / 3. When it is 0.5, it is the midpoint of the transverse segment, and when it is 1 / 3 or 2 / 3, it is the trisection point; if the distance between I_i and O_j is less than the lower limit of the channel width or greater than the upper limit of the channel width, the local transverse segment is judged as abnormal and does not participate in the generation of candidate target points.

[0111] Z_i represents the candidate target point. By adjusting λ, the vehicle can be kept in the middle of the channel or offset to one side according to the safety margin. In terms of computational implementation, the inner and outer point sets can be constructed as two-dimensional arrays respectively. For each inner point, the squared distance between it and all outer points is vectorized and calculated. The outer point with the smallest squared distance is then selected for pairing, thereby improving real-time performance.

[0112] Example 8

[0113] A discrete boundary proportional point tracking control method for intelligent vehicles, as described in Embodiment 7, differs in that, in step S7, after the candidate target points are generated, points with X coordinates less than the minimum pre-aiming distance L_min are first deleted. Then, the distance between each candidate target point and the vehicle origin is calculated and sorted in ascending order of distance. The Nth nearest target point is selected as the current target point based on vehicle speed and curve conditions. N can be increased at higher vehicle speeds or on straightaways, and decreased at curves or low speeds. If no candidate target point meets the pre-aiming conditions, the nearest candidate target point in front of the vehicle is selected as the downgraded target point. The purpose of this step is to obtain the current target point for the current control cycle, and the control node calculates the direction error based on the current target point, combining the speed gear and the state machine outputting linear velocity, angular velocity, or equivalent steering angle commands.

[0114] L_min is taken from 0.6m to 1.2m, preferably 0.8m.

[0115] In step S7, the formula for calculating the direction error e is as follows:

[0116] e = atan2(y_t, x_t)

[0117] Where (x_t, y_t) are the coordinates of the current target point;

[0118] The formula for calculating the steering control quantity u is as follows:

[0119] u = K_p(v)e + K_i(v)∑e + K_d(v)(e - e_last)

[0120] Wherein, K_p(v) represents the proportional term parameter corresponding to the speed gear v; K_i(v) represents the integral term parameter corresponding to the speed gear v; and K_d(v) represents the derivative term parameter corresponding to the speed gear v. These parameters can be calibrated through real-vehicle step steering tests, low-speed straight-line tracking tests, and cornering tracking tests. Alternatively, empirical values ​​can be used first, and then adjusted based on target point overshoot, steering oscillation, and response lag. Generally, a larger K_p(v) can be used in low-speed gears to improve steering response, while K_p(v) should be appropriately reduced and K_d(v) or the steering change rate limited in medium and high-speed gears to improve smoothness. The control node also switches between normal tracking, temporary deceleration, stop-and-wait, and resume tracking states based on stop signals, lap count information, target point validity, and positioning feedback.

[0121] e_last represents the direction error of the previous control cycle. The direction error e calculated in the previous moment is cached by the controller and used as e_last for the current cycle. e_last is set to 0 during system initialization.

[0122] The linear velocity is determined as follows: Pre-set the normal tracking speed v_base, cornering deceleration speed v_turn, degraded speed v_safe, and stopping speed 0. When |e| is less than the first angle threshold, output v_base; when |e| is between the first and second angle thresholds, output v_turn; when |e| is greater than the second angle threshold, it indicates excessive steering deviation and a risk of the vehicle running off the road, so output the degraded speed v_safe; if the deviation persists for more than 5 frames without convergence, output 0 to trigger stopping protection; when target points are continuously invalid (no valid candidate target points detected for 3-5 consecutive frames), it indicates a failure of the perception system, so output v_safe for slow creep; if invalid for more than 10 frames, output 0 for stopping and waiting.

[0123] The first angle threshold (straight-line tracking threshold) is set to 10°~15°, and the second angle threshold (curve safety threshold) is set to 30°~45°. For example, if the first angle threshold is 12° (about 0.21 rad), driving at full speed on a straight road will be allowed if the value is less than this. If the second angle threshold is 35° (about 0.61 rad), driving at full speed on a straight road will be allowed if the value is greater than this. If the value is greater than this, safety degradation will be triggered.

[0124] The angular velocity ω is determined as follows: ω = clamp(u, -ω_max, ω_max), where clamp() represents the clamping function, which restricts the input value to the closed interval [-ω_max, ω_max]. If the input is greater than ω_max, the output is ω_max; if the input is less than -ω_max, the output is -ω_max; otherwise, the original value is output.

[0125] ω_max represents the maximum permissible angular velocity of the vehicle, which can be determined based on the chassis steering performance and the maximum vehicle speed. The example value for an outdoor low-speed intelligent vehicle is 1.0 rad / s (approximately 57° / s), and the range can be 0.8 rad / s to 1.5 rad / s.

[0126] The equivalent front wheel steering angle δ is determined as follows: when v is greater than zero, δ = atan(L_w·ω / v), where L_w is the wheelbase, and the wheelbase of the small racing intelligent car is 0.3m~0.5m, for example 0.4m; v is the actual linear velocity value output in the current control cycle, which is output by the speed state machine according to the cornering conditions in v_base / v_turn / v_safe, and is not a fixed gear.

[0127] For servo / motor chassis, linear velocity and steering angle can be output through equivalent control quantity encoding. For example, with 1500 as the speed median, the speed level v is superimposed on the linear velocity channel, and with 80 as the steering median, the PID output or fixed cornering angle is superimposed on the angular velocity / steering channel, and the steering output is limited from 0 to 180.

[0128] Example 9

[0129] A discrete boundary proportional point tracking control method for intelligent vehicles is presented. An outdoor discrete cone-tracking intelligent vehicle is built based on the ROS2 architecture. Hardware configuration includes a 16-line LiDAR, a racing intelligent vehicle chassis, and an industrial control processing unit. The vehicle wheelbase L_w = 0.4m, and the nominal track width is 1.6m. The method fully implements all steps S1-S7 of this invention, with decoupled operation of the perception node and control node. The specific process is as follows: Figures 1 to 6 .

[0130] Step S1: LiDAR coordinate transformation

[0131] The lidar acquires the original polar coordinate point cloud at 10Hz, and the coordinates are corrected using the formulas x_i =(r_i + Δr) cos(α_i + Δα) + Δx and y_i =(r_i + Δr) sin(α_i + Δα) + Δy.

[0132] Hardware compensation parameters: Δr=0.02m distance offset, Δα=0rad angle offset, radar installation offset Δx=0m, Δy=0m, all converted to the vehicle coordinate system (X forward, Y left).

[0133] Step S2: Rectangular effective region filtering

[0134] Rectangular boundary parameters: maximum 2.3m in front of the vehicle, 0.35m redundancy behind, and lateral range of -1.6m to 1.8m.

[0135] Only the point cloud within the range of -0.35 ≤ x ≤ 2.3 and -1.6 ≤ y ≤ 1.8 is retained, while vehicle body and long-distance external clutter are filtered out to obtain the effective point set, as shown in the rectangular box filtering logic in Figure 2.

[0136] Step S3 as follows Figure 2 Clustering and deduplication of centroids

[0137] KMeans clustering is used, with the dynamic number of clusters K = min(15, number of valid points) and a maximum of 300 iterations;

[0138] For points in the same cone with a deduplication radius of 0.15m and an Euclidean distance less than the threshold, the coordinates are averaged and merged to output a unique candidate center point for the boundary.

[0139] Step S4: Separation of inner and outer boundaries

[0140] The inner boundary of the preset track is located on the left side of the vehicle (positive Y direction), and the leftmost point closest to the vehicle is selected as the seed point;

[0141] Iterative expansion conditions: the square of the Euclidean distance between two points ≤ 1.69, and the square of the lateral difference ≤ 1.21; points that do not meet the conditions are all classified as outer boundary points, and the distinction results are shown in Figure 2 with red and blue markings.

[0142] Step S5: X-axis segmented lower envelope filtering

[0143] The segment step size Δx = 1.0m, and the segment tolerance τ = 0.3m;

[0144] Each X interval retains only the inner boundary point with the smallest lateral width, and removes redundant interference points on the same side; at the same time, it filters out inner and outer paired line segments with a cross-sectional width of <1.0m or >2.2m, and removes outlier points in adjacent lanes.

[0145] Step S6: Generation of proportional candidate target points

[0146] Each inner point is matched with the nearest outer point in Euclidean geometry, and the target point is generated according to the midline ratio of λ=0.5. The calculation formula is Z_i = I_i + 0.5(O_j - I_i). Only the matching line segments with compliant channel width are retained. The generated midline point is shown in the triangular mark in Figure 3.

[0147] Step S7: Pre-targeting and filtering + State machine PID control

[0148] 1. Pre-aiming filtering: Remove point clouds with X < 0.8m, sort the remaining points in ascending order of distance, and select the second closest point as the real-time target point;

[0149] 2. Error calculation: e = atan2(y_t, x_t), the PID differential term is substituted with the previous frame error e_last (initialized to 0);

[0150] 3. Speed ​​Adaptive PID Parameters:

[0151] The control node selects the proportional coefficient, integral coefficient, and derivative coefficient from multiple preset PID parameters based on the actual speed range of the vehicle; when the vehicle speed crosses a preset threshold, it switches to the corresponding parameter group.

[0152] The specific values ​​of each speed threshold and Kp, Ki, and Kd are determined through actual vehicle calibration based on chassis steering characteristics, vehicle load, tire adhesion conditions, and site curvature, in order to balance response speed and steering stability.

[0153] 4. The speed state machine is shown in Table 1:

[0154] Table 1. Speed ​​State Machine

[0155] 5. Steering Limiting: The PID output u is limited to a preset steering range by a limiting function, and then converted into angular velocity, equivalent front wheel steering angle, or servo control quantity according to the chassis type before being sent for execution. Figure 4 .

[0156] The perception node subscribes to the / scan laser topic, and after processing, publishes the / target, / circle, and / stop_signal topics via BEST_EFFORT QoS.

[0157] The control node subscribes to the above topics, independently runs the PID and lap count state machine, and outputs the / cmd_vel speed and steering command; the sensing node processes the point cloud according to the arrival frequency of the LiDAR data; the control node runs according to the preset control cycle, and can set independent high-frequency check callbacks for safety events such as parking.

[0158] Figure 6 The light blue sector represents the detection area 2m ahead, as shown in the node recognition results.

[0159] The blue dots represent the original point cloud from the lidar.

[0160] The red markers represent the cluster centers at the inner boundaries;

[0161] The yellow markers indicate the cluster centers at the outer boundaries;

[0162] The purple pentagram represents the final selected real-time tracking target point;

[0163] The green pentagon represents the origin of the intelligent vehicle.

[0164] Actual test results:

[0165] The outdoor cone track was followed completely for two consecutive laps with no significant changes in the target point. The track was stable and downgraded in scenarios with missing corners and boundaries, and there was no run-off from the track.

[0166] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for discrete boundary proportional point tracking control of an intelligent vehicle, characterized in that, Includes the following steps: S1 collects LiDAR data around the intelligent vehicle body and performs coordinate transformation to convert the LiDAR data into two-dimensional points in the vehicle coordinate system. S2, set a rectangular effective area based on the intelligent vehicle body length, vehicle width B, channel width, radar installation position and vehicle speed, filter out points outside the rectangular effective area to obtain the effective point set; S3, cluster the effective point set to obtain several candidate boundary center points, and remove duplicate center points to avoid generating multiple duplicate center points for the same marker; S4. Select the inner boundary seed point based on the lateral position and distance relationship, and iteratively expand the inner boundary point set starting from the seed point. Candidate boundary center points that are not included in the inner boundary point set are included in the outer boundary point set. S5, divide the inner boundary point set into segments according to direction, retain the points of the lower envelope in each segment, and delete the outliers on the outer side; S6. For each inner boundary point, find the nearest outer boundary point to form a local transverse line segment, and generate candidate target points according to the proportion. S7 performs pre-aiming distance filtering and distance sorting on candidate target points to obtain the current target point. Based on the current target point, it calculates the direction error and combines the speed gear and state machine outputs linear velocity, angular velocity, or equivalent front wheel rotation angle commands.

2. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 1, characterized in that, The coordinate transformation process in step S1 is as follows: x_i =(r_i + Δr) cos(α_i + Δα) + Δx y_i =(r_i + Δr) sin(α_i + Δα) + Δy Where x_i is the X coordinate in the vehicle coordinate system, y_i is the Y coordinate in the vehicle coordinate system, the X-axis points in the direction of vehicle movement, the Y-axis points to the left side of the vehicle and is perpendicular to the X-axis, r_i is the ranging value of the i-th laser beam, α_i is the scanning angle of the i-th laser beam, Δr is the distance compensation amount, Δα is the angle compensation amount, and Δx and Δy are the installation offsets of the lidar relative to the origin of the vehicle coordinate system in the front-back direction and the left-right direction.

3. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 2, characterized in that, In step S2, the front distance of the intelligent vehicle body is set at 2.0m to 2.5m, the rear redundancy distance is set at 0.2m to 0.5m, and the left and right widths are set at B / 2 plus a safety margin of 0.3m to 0.6m to determine the rectangular effective area; Valid points are retained as follows, while points outside the valid rectangular region are filtered out: x_min ≤ x_i ≤ x_max, y_min ≤ y_i ≤ y_max In the formula, x_min, x_max, y_min, and y_max are the minimum, maximum, minimum, and maximum values ​​of the rectangular effective region along the X-axis, Y-axis, and Y-axis, respectively.

4. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 3, characterized in that, In step S3, one of KMeans, DBSCAN, or Euclidean clustering is used, and the number of clusters K is between 10 and 20. The deduplication radius refers to the Euclidean distance threshold between the center points of two candidate boundaries. When the distance between the center points of two candidate boundaries is less than the deduplication radius, they are considered to correspond to the same marker or the same local boundary. The center point that is closer to the intelligent vehicle body and has more points is retained, or the average of the two is used as the merged center point.

5. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 4, characterized in that, In step S4, the lateral position refers to the Y-coordinate of the candidate boundary center point in the vehicle coordinate system. Based on the vehicle's current driving direction or the site's detour direction, the inner boundary is pre-determined to be located on the left or right side of the vehicle. Seed points for the inner boundary are preferentially selected from the candidate center points closest to the vehicle. Starting from seed point q, the remaining candidate boundary center points are traversed. When the point p to be judged meets the following conditions, it is added to the inner boundary point set: (x_p - x_q)^2 + (y_p - y_q)^2 ≤D, (y_p - y_q)^2 ≤T Where D is the square of the Euclidean distance threshold, and T is the square of the lateral difference threshold; x_p and y_p represent the forward and lateral coordinates of the point p to be judged in the vehicle coordinate system, respectively, and x_q and y_q represent the forward and lateral coordinates of the seed point q in the vehicle coordinate system, respectively.

6. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 5, characterized in that, In step S5, the inner boundary point set is segmented along the X-axis, dividing the range from x_min to x_max into multiple intervals Ω_q with a fixed segment length Δx. Within each segment interval, the lateral coordinate reference value of the inner boundary points is calculated, and points whose lateral coordinates are not greater than the reference value plus the segment tolerance τ are retained. These points are used to retain the boundary points close to the lower envelope of the lane within each segment. The formula is as follows: y_i≤ min(y_j | x_j ∈ Ω_q) + τ Where y_i represents the lateral coordinate of the inner boundary point i to be screened in the vehicle coordinate system, y_j represents the lateral coordinate of other inner boundary points in the same segment interval Ω_q, x_j represents the forward coordinate of other inner boundary points in the same segment interval Ω_q, and τ represents the segment tolerance; Preferably, obviously erroneous outlier points include: points located on the same side of the inner boundary, points whose local cross-section segment formed with the inner boundary point has a length less than the lower limit of the channel width or greater than the upper limit of the channel width, points whose lateral residual relative to the fitted curve exceeds 0.3m to 0.5m, and points whose X direction falls outside the effective area. The channel width refers to the nominal lateral width of the driving channel formed by the discrete boundary markers. The lower limit of the channel width is 70% of the channel width, and the upper limit of the channel width is 130% of the channel width.

7. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 6, characterized in that, Δx is taken as 0.5m to 1.5m; τ is taken as 0.2m to 0.5m.

8. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 7, characterized in that, In step S6, the process of generating candidate target points is as follows: j = arg min ||I_i - O_j|| 2 Z_i = I_i + λ(O_j - I_i), 1 / 3 ≤ λ ≤ 2 / 3 In the formula, I_i represents the i-th inner boundary point, O_j represents the outer boundary point closest to it, and j represents the set of outer boundary points such that ||I_i - O_j|| 2 The index of the outer boundary point that yields the minimum value; λ represents the proportion, ranging from 1 / 3 to 2 / 3; if the distance between I_i and O_j is less than the lower limit of the channel width or greater than the upper limit of the channel width, the local transverse segment is judged as abnormal and does not participate in the generation of candidate target points.

9. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 8, characterized in that, In step S7, after the candidate target points are generated, points with X coordinates less than the minimum aiming distance L_min are first deleted. Then, the distance between each candidate target point and the vehicle origin is calculated and sorted in ascending order of distance. The Nth nearest target point is selected as the current target point based on vehicle speed and curve conditions. L_min is taken as 0.6m to 1.2m.

10. The intelligent vehicle discrete boundary proportional point tracking control method according to claim 9, characterized in that, In step S7, the formula for calculating the direction error e is as follows: e = atan2(y_t, x_t) Where (x_t, y_t) are the coordinates of the current target point; The formula for calculating the steering control quantity u is as follows: u = K_p(v)e + K_i(v)∑e + K_d(v)(e - e_last) Where K_p(v) represents the proportional term parameter corresponding to the speed level v; K_i(v) represents the integral term parameter corresponding to the speed level v; K_d(v) represents the derivative term parameter corresponding to the speed level v; and e_last represents the direction error of the previous control cycle. The linear velocity is determined as follows: Pre-set the normal tracking speed v_base, cornering deceleration speed v_turn, degraded speed v_safe, and stopping speed 0. When |e| is less than the first angle threshold, output v_base; when |e| is between the first and second angle thresholds, output v_turn; when |e| is greater than the second angle threshold, it indicates excessive steering deviation and a risk of the vehicle running off the road, so output the degraded speed v_safe; if the deviation persists for more than 5 frames without convergence, output 0 to trigger stopping protection; when the target point is continuously invalid, it indicates a failure of the perception system, so output v_safe for slow creep; if it remains invalid for more than 10 frames, output 0 to stop and wait. The first angle threshold is set at 10°~15°, and the second angle threshold is set at 30°~45°. The angular velocity ω is determined as follows: ω = clamp(u, -ω_max, ω_max), where clamp() represents the clamping function; ω_max represents the maximum permissible angular velocity of the vehicle. The equivalent front wheel steering angle δ is determined as follows: when v is greater than zero, δ = atan(L_w·ω / v), where L_w is the wheelbase and v is the actual linear velocity value output in the current control cycle.