A method for predicting and quantifying collision risk of a road ahead of an unmanned mine truck

Abstract

The application discloses a kind of unmanned mine truck front road collision risk prediction and quantification method, including the following steps: determine the reference path of the given ODD length front as the running design domain of collision risk prediction;Calculate the maximum sweep range of the vehicle body when the vehicle travels along the reference path of the ODD length front, as the absolute risk area;Based on the absolute risk area, the maximum lateral deviation of the vehicle is offset outward, to define the left and right strip area outside the absolute risk zone as the relative risk area;Outside the relative risk range is defined as a risk-free area;The position information of each corner point of the polygonal obstacle of the front road is converted to the driving path coordinate system;Determine whether the obstacle will cause risk and give the collision risk quantification result one by one.The method is based on the driving path coordinate system to predict the collision, and converts the collision of different intrusion levels into quantifiable risk output, which can realize more intuitive collision prediction, quantifiable prediction results, and can classify and control different risks to avoid.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks. Under the condition that the planned path, the "obstacle" ahead output by the perception system, and the vehicle's own motion state are known, the method predicts whether the obstacle will collide with the vehicle and outputs a quantitative risk. This method is applicable to collision prediction and quantitative risk output for autonomous driving systems (ADS) and driver assistance systems (ADAS) for highway vehicles and off-road vehicles. Background Technology

[0002] In recent years, with the continuous development of artificial intelligence, autonomous driving has gradually become a research hotspot in the transportation field. Research on real-time collision detection methods during autonomous driving is of great significance for improving road traffic safety. Currently, existing research mainly focuses on pixel-based real-time collision detection algorithms and real-time collision detection algorithms based on the separating axis theorem.

[0003] Pixel-based real-time collision detection algorithms are the most accurate, testing each pixel of an object and identifying collisions when pixels overlap. However, this algorithm is computationally intensive and rarely used. When high accuracy is not required, the bounding sphere algorithm can be used. This involves enclosing the object with its circumcircle. To test if two objects collide, it's only necessary to calculate whether the distance between the centers of the two circles is greater than the sum of their radii. This method has poor precision and is only applicable in specific scenarios, such as V2V collision prediction, where it's useful for identifying vehicles with turning radii like excavators and electric shovels.

[0004] A real-time collision detection algorithm based on the Separating Axis Theorem (SAT) is used to detect collisions of convex polygons. In the collision prediction process, the bounding box is first assumed to travel along a reference path to each path point. Then, it performs separation axis collision detection on each obstacle's convex polygon's xy-axis projection (ignoring height) given by the perception system—determining whether the two intersect—and outputs a collision risk conclusion. For example, Chinese invention patent CN202110713798.4 discloses a method for detecting convex polygon collisions, which uses a hill-climbing algorithm to calculate the support points and potential separation axis set of the convex polygon, deriving the minimum travel distance to determine whether a collision has occurred.

[0005] Although the separation axis algorithm is relatively accurate, it requires traversing every edge of the two convex polygons and is only suitable for collision prediction of convex polygons. It can only output a collision or non-collision result at each instant, without knowing which edge will cause a collision, nor can it output the collision range or the degree of collision risk. In extreme cases, the collision determination actually depends on the degree of discretization of collision detection when the vehicle travels along the reference path. A low degree of discretization may result in potential collision risks not being detected in time, while a high degree of discretization will cause the computational load to increase exponentially. Figure 1 As shown, there is an obstacle in front of the vehicle. Discrete sampling is performed every 5 meters along the reference path ahead. The collision risk is judged using a polygon collision detection algorithm with a split axis. In this case, the collision risk cannot be detected in time, but the left front corner of the vehicle that is actually traveling according to the reference path has already collided with the obstacle, resulting in a missed detection.

[0006] Furthermore, the collision risk assessment results based on polygon collision detection algorithms such as the separation axis cannot intuitively display the collision intrusion range. To intuitively see the effect of the "collision", the two polygons must be redrawn in the same coordinate system so that the viewer can understand the degree of collision that will occur, but the degree of collision interference cannot be obtained through calculation.

[0007] To provide a more intuitive and quantifiable method for collision prediction, this invention proposes a collision prediction method with quantifiable risk. This method, when the reference path is known, combines...<s,l> The coordinate system predicts whether the "obstacle" output by the perception system will collide with itself, outputting collision risk attributes and quantitative risk. A patent search reveals that the existing patent in this field is Chinese invention patent CN201911398952.2, which discloses a collision prediction and judgment method and device for autonomous vehicles. This method predicts collisions based on the movement trajectories of different vehicles and uses a whisker algorithm for collision detection, which is significantly different from the method of this invention. Summary of the Invention

[0008] To address the above issues, a travel path coordinate system based on the existing reference path is used.<s,l> The purpose of this invention is to establish a driving path coordinate system.<s,l> This invention performs collision prediction, transforming collisions of varying intrusion levels into quantifiable risk outputs. This results in more intuitive collision prediction, quantifiable prediction outcomes, and the ability to classify and control different risks. The technical solution adopted in this invention is as follows:

[0009] A method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks includes the following steps:

[0010] S1, Obtain the forward reference path: Determine the forward reference path for the given ODD length as the operational design domain for collision risk prediction;

[0011] S2, Calculate the absolute risk area: Calculate the maximum sweep range of the vehicle body when the vehicle travels along the reference path ahead of the ODD length. The maximum sweep range of the vehicle body is the maximum envelope area that the vehicle's corner points can sweep when the vehicle travels accurately along the reference path, which is the absolute risk area.

[0012] S3, Calculate the relative risk area: Based on the lateral deviation of the daily autonomous vehicle driving along the reference path obtained from the statistical data of actual operation, and in accordance with the principle of statistical safety, determine the maximum value of the vehicle's lateral deviation LatErrMax. Based on the absolute risk area, shift outward from LatErrMax to define the left and right strip areas outside the absolute risk area as the relative risk area.

[0013] S4, Determine the area in front of the vehicle where there is no risk of collision: The area outside the relative risk range is defined as a no-risk area, and obstacles appearing in this area are considered to pose no risk of collision;

[0014] S5, Obtain obstacle location information: During real-time driving, the vehicle obtains obstacle location information on the road ahead based on the vehicle perception system, and converts the position information of each corner point of the polygonal obstacle into the driving path coordinate system;

[0015] S6: Collision Risk Assessment: Following the decision-making logic of absolute collision risk, relative collision risk, and no collision risk, the system assesses whether obstacles on the road ahead will pose a risk, and calculates the quantitative result of the collision risk of the obstacle for each case.

[0016] Furthermore, in step S3, the principle of statistical safety refers to the fact that the lateral deviation of the vehicle obtained from a large amount of data statistics follows a Gaussian distribution, and the value taken is the maximum value of the lateral deviation, LatErrMax.

[0017] Furthermore, in step S5, the driving path coordinate system<s,l> The path length when traveling forward along the nearest reference path point to the vehicle's current position is the s-axis, with s being positive for the front and negative for the back. The left and right offsets of each path point are the l-axis, with l being positive for the left and negative for the right.

[0018] Furthermore, step S5 includes the following sub-steps:

[0019] S51, Match each corner point of the polygonal obstacle with the nearest reference path point, calculate the lateral distance between each corner point of the polygonal obstacle and the nearest reference path point in the normal direction of the path point's heading, and use it as the l value of each corner point of the polygonal obstacle;

[0020] S52, calculate the path distance between the nearest reference path point of each corner point of the polygonal obstacle and the current vehicle position, and use it as the s value of each corner point of the polygonal obstacle;

[0021] S53, thus, the coordinates (s_i, l_i) of each corner point of the polygonal obstacle in the driving path coordinate system are obtained, where i is the corner point number of the polygonal obstacle.

[0022] Furthermore, step S6 includes the following sub-steps:

[0023] S61, If ​​an obstacle has a corner point in the absolute risk zone, record the absolute risk zone quantification risk value as 100%, and simultaneously record the s value and l value at the corner point with the smallest lateral distance from the obstacle.

[0024] S62, if no corner point is in the absolute risk zone, but a corner point is in the relative risk zone, then the quantified risk value is recorded according to 1%-99%; the quantified risk value is the ratio of the distance ΔL from the nearest corner point of the obstacle to the relative risk zone to the width of the relative risk zone; at the same time, the s value and l value of the obstacle corner point that produces the deepest lateral interference in the relative risk zone are recorded.

[0025] S63, if an obstacle has no corner points in the absolute or relative risk zone, its quantified risk value is 0%.

[0026] The beneficial effects of this invention compared to the prior art are as follows:

[0027] 1. Collision detection is more intuitive:

[0028] Driving path coordinate system<s,l> In autonomous driving scenarios, the coordinate system is defined with the path trajectory line as the center line, where 's' represents the direction of the path trajectory line and 'l' represents the direction perpendicular to the path trajectory line. Driving path coordinate system.<s,l> It has the advantages of being intuitive and easy to understand, and is more in line with the actual needs of obstacle prediction than the XY coordinate system. It adopts the driving path coordinate system.<s,l> For collision prediction of static obstacles, it is easier to understand than the separation axis algorithm, and the prediction results are more intuitive and easier to visualize.

[0029] 2. Collision risk is quantifiable:

[0030] For absolute risk attributes, outputting the longitudinal distance allows for timely parking in front of obstacles.

[0031] For obstacles in relatively high-risk areas, a quantitative output of 1-99% is given based on the degree of lateral collision interference.

[0032] For obstacles in risk-free areas, no risk is output.

[0033] The aforementioned quantitative output principles facilitate subsequent control functions in limiting speed or stopping in a timely manner, and can also trigger decision-making functions to replan reference paths and avoid obstacles.

[0034] For expected functional safety, the goal is to reduce uncertain risks. The risk quantification output of this invention can help the decision-making function align with the requirements of expected functional safety, enabling subsequent processes such as stopping, speed limiting, and replanning obstacle avoidance.

[0035] 3. Interpolation fitting can be used for judgment to avoid missed detections:

[0036] For absolute risk areas, the representation of their discrete point set allows for convenient interpolation and fitting, avoiding the omission of potential risks due to excessively large step selections in the risk area data preparation process.

[0037] 4. Reduced algorithm complexity:

[0038] Determining the separation axis requires calculating the bounding box within the longitudinal perception range of the vehicle after stepping along the reference path at each current position. In the second loop, the separation axis is determined for the convex hull of each perceived "obstacle".

[0039] The risk area data preparation in this method can be performed asynchronously when the reference path is obtained. Obstacle detection only occurs after the "obstacle" perception output event occurs.<s,l> By performing coordinate system transformation, the risk area can be determined through size comparison. Then, the interference distance ΔL can be determined through ± operation, followed by a division operation to obtain the linearly quantized risk output. The algorithm load in a typical running cycle is significantly reduced. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of collision risk prediction based on polygon collision detection algorithms in existing technologies.

[0041] Figure 2 This is a flowchart of the method for predicting and quantifying the collision risk of unmanned mining trucks ahead, according to the present invention.

[0042] Figure 3 This is a schematic diagram illustrating the principle of absolute risk zones for the vehicle body during driving.

[0043] Figure 4 It is on the reference path<s,l> Coordinates and a schematic diagram of road risk zone delineation.

[0044] Figure 5 Based on the present invention<s,l> Coordinate system obstacle collision risk quantification assessment process. Detailed Implementation

[0045] To make the purpose and technical solution of the invention clearer, they are illustrated in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0046] A method for predicting and quantifying the collision risk of unmanned mining trucks on the road ahead, such as... Figure 2 As shown, the process mainly includes six steps: obtaining the reference path ahead, calculating the absolute risk area, calculating the relative risk area, determining the no-collision-risk area in front of the vehicle, obtaining obstacle location information, and judging the collision risk.

[0047] S1, Obtain the forward reference path. Determine the forward ODD = 50m (Operational Design Domain) as the operational design domain for collision risk prediction, and obtain the pose information [longitude; latitude; heading] of the discrete points of the forward reference path along the ODD length.

[0048] S2, Calculate the absolute risk area. Calculate the maximum sweep range of the vehicle body as it travels along the reference path of the preceding ODD length, and obtain the left and right boundary envelopes of the vehicle body sweep range (e.g., ...). Figure 3 (As shown). The maximum sweep range of the vehicle body is the maximum envelope area swept by the vehicle's corner points when the vehicle accurately travels along the reference path, serving as the absolute risk area. The maximum left boundary value l_left and right boundary value l_right along the normal direction of the heading are calculated for each discrete point on the reference path. This yields the left and right boundaries of the absolute risk area assigned to each discrete point on the reference path (e.g., ...). Figure 4 As shown), the left and right boundary values ​​are defined as positive on the left and negative on the right. This yields the discrete point information of the reference path with the absolute risk area [longitude; latitude; heading; left boundary value of the absolute risk area l_left; right boundary value of the absolute risk area l_right].

[0049] S3, Calculate the relative risk area. Based on the lateral deviation of the autonomous vehicle's performance along the reference path, obtained from statistical analysis of actual operational data, and following the principle of statistical safety, determine the maximum lateral deviation, LatErrMax. Based on the absolute risk area, offset outwards by LatErrMax, define the left and right strip-shaped areas outside the absolute risk area as the relative risk area. The statistical safety principle means that if the lateral deviation of the vehicle, obtained from a large amount of data, follows a Gaussian distribution, then the value at that point is the maximum lateral deviation mentioned above. Thus, determine the left safety margin LatErrMax and the right safety margin LatErrMax of the relative risk area for each discrete reference path point.

[0050] Thus, the reference path discrete point information with absolute risk area and relative risk area is obtained [longitude; latitude; heading; left boundary value of absolute risk area l_left; right boundary value of absolute risk area l_right; left safety margin of relative risk area LatErrMax; right safety margin of relative risk area LatErrMax].

[0051] S4, Determine if there is no risk of collision in front of the vehicle. Figure 4 The area outside the dotted line. The area outside the relative risk range is defined as a risk-free zone, where obstacles appearing in this zone are considered to pose almost no risk of collision.

[0052] S5, Obtain obstacle location information. During real-time driving, the vehicle uses its perception system to obtain the location information of obstacles on the road ahead and transforms the position information of each corner point of the polygonal obstacle into the driving path coordinate system. Define the driving path coordinate system.<s,l> The path length when traveling forward along the nearest reference path point to the vehicle's current position is the s-axis, with s being positive for the front and negative for the back. The left and right offsets of each path point are the l-axis, with l being positive for the left and negative for the right.

[0053] S51, match each corner point of the polygonal obstacle with the nearest reference path point, calculate the lateral distance between each corner point of the polygonal obstacle and the nearest reference path point in the normal direction of the path point's heading, and use it as the l value of each corner point of the polygonal obstacle.

[0054] S52 calculates the path distance between the nearest reference path point to each corner of the polygonal obstacle and the current vehicle position, and uses it as the s value for each corner of the polygonal obstacle.

[0055] S53, thus, the coordinates (s_i, l_i) of each corner point of the polygonal obstacle in the driving path coordinate system are obtained, where i is the corner point number of the polygonal obstacle.

[0056] S6, Collision Risk Assessment. For example... Figure 5 As shown, based on the following decision-making logic, it is determined one by one whether an obstacle on the road ahead will pose a risk: absolute / relative / no collision risk. For each case, the collision risk quantification result of the obstacle is calculated.

[0057] S61, if the obstacle has a risky corner point in the absolute risk zone, then record the absolute risk zone quantified risk value as 100%; at the same time, record the s value and l value at the corner point with the smallest lateral distance from the obstacle, and transmit them to the downstream control module to make speed decisions and control the speed according to different collision risk values.

[0058] S62: If no obstacle corner is in the absolute risk zone, but an obstacle corner is in the relative risk zone, then the quantified risk value is recorded as 1%-99%. The output of the quantization result is the ratio (0%~100%) of the distance ΔL from the nearest obstacle corner's lateral intrusion into the relative risk zone to the width of the relative risk zone. The quantization algorithm can use linear calculation or nonlinear algorithms such as hyperbolic tangent, sigmoid, etc. Taking a linear algorithm as an example:

[0059] The relative risk zone quantification risk value = distance ΔL of the nearest corner of the obstacle laterally intruding into the relative risk zone ÷ width of the relative risk zone * 100%; at the same time, the s value and l value of the obstacle corner that generates the deepest lateral interference in the relative risk zone are recorded and transmitted to the downstream control module for speed decision-making, and speed control is performed based on different collision risk values;

[0060] S63. If an obstacle has no risk corners located in an absolute or relative risk zone, its quantified risk value is 0%.

[0061] This invention enables a collision prediction method with quantifiable risks, offering advantages such as more intuitive collision judgment, quantifiable collision risks, interpolation-based judgment, avoidance of missed detections, and lower complexity compared to split-axis collision detection algorithms.

Claims

1. A method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks, characterized in that, Includes the following steps: S1, Obtain the forward reference path: Determine the forward reference path for the given ODD length as the operational design domain for collision risk prediction; S2, Calculate the absolute risk area: Calculate the maximum sweep range of the vehicle body when the vehicle travels along the reference path ahead of the ODD length. The maximum sweep range of the vehicle body is the maximum envelope area that the vehicle's corner points can sweep when the vehicle travels accurately along the reference path, which is the absolute risk area. Calculate the maximum left boundary value l_left and right boundary value l_right along the normal direction of the heading for each reference path discrete point. This yields the left and right boundaries of the absolute risk area assigned to each reference path discrete point. The left and right boundary values ​​are defined as positive on the left and negative on the right. This results in the reference path discrete point information [longitude; latitude; heading] containing the absolute risk area. The left boundary value of the absolute risk region is l_left; The right boundary value of the absolute risk region is l_right; S3, Calculate the relative risk area: Based on the lateral deviation of the daily autonomous vehicle driving along the reference path obtained from the statistical data of actual operation, and in accordance with the principle of statistical safety, determine the maximum value of the vehicle's lateral deviation LatErrMax. Based on the absolute risk area, shift outward from LatErrMax to define the left and right strip areas outside the absolute risk area as the relative risk area. S4, Determine the area in front of the vehicle where there is no risk of collision: The area outside the relative risk range is defined as a no-risk area, and obstacles appearing in this area are considered to pose no risk of collision; S5, Obtain obstacle location information: During real-time driving, the vehicle obtains obstacle location information on the road ahead based on the vehicle perception system, and converts the position information of each corner point of the polygonal obstacle into the driving path coordinate system; S6: Collision Risk Assessment: Following the decision-making logic of absolute collision risk, relative collision risk, and no collision risk, the system assesses whether obstacles on the road ahead will pose a risk, and calculates the quantitative result of the collision risk of the obstacle for each case. In step S3, the principle of statistical safety refers to the fact that the lateral deviation of the vehicle obtained based on a large amount of statistical data follows a Gaussian distribution. ,Pick The value at that point is the maximum value of the lateral deviation, LatErrMax.

2. The method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks according to claim 1, characterized in that, In step S5, the driving path coordinate system<s,l> The path length when traveling forward along the nearest reference path point to the vehicle's current position is the s-axis, with s being positive for the front and negative for the back. The left and right offsets of each path point are the l-axis, with l being positive for the left and negative for the right.

3. The method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks according to claim 1, characterized in that, Step S5 includes the following sub-steps: S51, Match each corner point of the polygonal obstacle with the nearest reference path point, calculate the lateral distance between each corner point of the polygonal obstacle and the nearest reference path point in the normal direction of the path point's heading, and use it as the l value of each corner point of the polygonal obstacle; S52, calculate the path distance between the nearest reference path point of each corner point of the polygonal obstacle and the current vehicle position, and use it as the s value of each corner point of the polygonal obstacle; S53, thus, the coordinates of each corner point of the polygonal obstacle in the driving path coordinate system are obtained ( s_i, l_i ),in, i Number the corner points of the polygonal obstacle.

4. The method for predicting and quantifying the risk of collisions ahead on unmanned mining trucks according to claim 3, characterized in that, Step S6 includes the following sub-steps: S61, If ​​an obstacle has a corner point in the absolute risk zone, record the absolute risk zone quantification risk value as 100%, and simultaneously record the s value and l value at the corner point with the smallest lateral distance from the obstacle. S62. If no corner point is in the absolute risk zone, but a corner point is in the relative risk zone, then the quantified risk value is recorded as 1%-99%. The quantified risk value is the ratio of the distance ΔL from the nearest corner point of the obstacle to the relative risk zone to the width of the relative risk zone; at the same time, the s and l values ​​of the obstacle corner point that generates the deepest lateral interference in the relative risk zone are recorded; S63, if an obstacle has no corner points in the absolute or relative risk zone, its quantified risk value is 0%.

Images