Obstacle bypassing path generation method, vehicle and control system

Abstract

The invention provides an obstacle bypassing path generation method, a vehicle and a control system. By repeatedly determining different detection positions, the intelligent trolley collects appearance information of the static obstacles at the detection positions, then the obstacle local model of the static obstacles is perfected, and finally the obstacle local model of the static obstacles bypassing one side is perfected. Therefore, based on the perfect obstacle local model on the bypass side, a bypass path is generated for a subsequent intelligent trolley, and the purpose of effectively avoiding obstacles is achieved.

Description

Technical Field

[0001] This application relates to the field of navigation technology, and more specifically, to a method for generating obstacle detour paths, a vehicle, and a control system. Background Technology

[0002] Intelligent Guided Vehicles (IGVs) are unmanned transport devices used in automated terminals and smart factories. They achieve autonomous navigation and obstacle avoidance through technologies such as the BeiDou Navigation Satellite System, LiDAR, and visual SLAM. They possess high flexibility, intelligent navigation, and superior performance. IGVs typically follow navigation information provided by QR code strips or matrix QR codes.

[0003] Currently, when an IGV encounters a static obstacle on its route, bypassing the obstacle facilitates the IGV's rapid task execution. However, due to the difficulty in obtaining the specific shape and other characteristics of the static obstacle, recklessly bypassing it carries certain risks.

[0004] Therefore, this application provides a method for generating obstacle detour paths to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this application is to provide a method for generating obstacle detour paths, a vehicle, and a control system, which can solve at least one of the aforementioned technical problems. The specific solution is as follows: According to a specific embodiment of this application, in a first aspect, this application provides a method for generating obstacle detour paths, comprising: When a static obstacle is detected on the navigation path of the intelligent vehicle, the appearance information of the static obstacle is collected, and a local obstacle model of the static obstacle is generated in a three-dimensional electronic map based on the appearance information of the static obstacle. The detection location of the intelligent vehicle is determined based on the three-dimensional electronic map; The intelligent vehicle is controlled to move to the detection position to collect new appearance information of the static obstacle; The local model of the obstacle is supplemented based on the new appearance information; A detour path is generated based on the supplemented local obstacle model.

[0006] Optionally, determining the detection location of the intelligent vehicle based on the three-dimensional electronic map includes: Candidate observation positions are determined based on the positional relationship between the current position of the intelligent vehicle and the local model of the obstacle; Generate a candidate path from the current position of the intelligent vehicle to the candidate observation position; The candidate paths are scored to obtain a score value for each candidate path; The candidate observation location corresponding to the candidate path with the highest score is determined as the detection location.

[0007] Optionally, determining candidate observation positions based on the positional relationship between the current position of the intelligent vehicle and the local obstacle model includes: Based on the current position of the intelligent vehicle, determine the farthest exposed position of the local model of the obstacle in the direction that can be bypassed ahead; Determine the farthest plane containing the farthest exposed position, wherein the farthest plane is perpendicular to the driving direction of the intelligent vehicle when it is paused; Determine the farthest candidate observation position outside the farthest exposed position, wherein the farthest candidate observation position is located on the farthest plane and the distance from the local model of the obstacle meets a preset safe distance range; Based on the current position of the intelligent vehicle, determine the most recently exposed position of the local model of the obstacle in the dodgeable direction ahead; Determine the nearest plane containing the most recently exposed location, wherein the nearest plane is parallel to the farthest plane; Determine the nearest candidate observation position outside the nearest exposure position, wherein the nearest candidate observation position is located on the nearest plane and the distance from the local model of the obstacle meets a preset safe distance range; From the nearest candidate observation position to the farthest candidate observation position, including the nearest candidate observation position and the farthest candidate observation position, the candidate observation position outside the local model of the obstacle is determined.

[0008] Optionally, scoring the candidate paths to obtain score values ​​for the candidate paths includes: The candidate paths are evaluated using multiple scoring methods to obtain multiple score values, including: data integrity score, safety gap score, and mobility feasibility score. The data integrity score is used to assess the integrity of the appearance of the local model of the obstacle. The safety gap score is used to assess the safety of the intelligent vehicle passing outside the candidate path. The mobility feasibility score is used to assess the feasibility of other intelligent vehicles hindering the intelligent vehicle from detouring. The total score is obtained based on the aforementioned multiple score values.

[0009] Optionally, the candidate paths are scored using multiple methods to obtain multiple score values, including: Obtain the outer segment information of the candidate path in the local model of the obstacle; The number of local regions is obtained based on the outer segment information; The data integrity score of the candidate path is determined based on the number of local regions.

[0010] Optionally, the candidate paths are scored using multiple methods to obtain multiple score values, including: Obtain the first minimum distance value between the candidate path and the external environment model, and Obtain the first sum of the first half-width value and the preset safety gap value of the intelligent vehicle; When the minimum distance value of the candidate path is less than or equal to the preset first half-width value of the smart car, the safety gap item score value of the candidate path is determined to be zero. When the minimum distance value of the candidate path is greater than the preset first half-width value of the intelligent vehicle, and the minimum distance value of the candidate path is less than or equal to the first sum value, the safety gap item score value of the candidate path is determined to be the first score value. When the minimum distance value of the candidate path is greater than the first sum value, the safety gap item score of the candidate path is determined to be the second score value, wherein the second score value is greater than the first score value.

[0011] Optionally, the candidate paths are scored using multiple methods to obtain multiple score values, including: Obtain the second minimum distance value between the candidate path and the paths traversed by the other intelligent vehicles, and Obtain the second sum of the preset first half-width value of the intelligent vehicle and the preset second half-width values ​​of the other intelligent vehicles, and Obtain the third sum of the second sum and the preset safety gap value; When the minimum distance value of the candidate path is greater than the third sum value, the mobility feasibility score of the candidate path is determined to be the third score value. When the minimum distance value of the candidate path is greater than the second sum value, and the minimum distance value of the candidate path is less than or equal to the third sum value, the mobility feasibility score of the candidate path is determined to be the fourth score value, wherein the fourth score value is less than the third score value; When the minimum distance value of the candidate path is less than or equal to the second sum value and greater than the preset first half-width value of the intelligent vehicle, the borrowed road segment in the candidate path is determined; The travel time of the borrowed road segment is obtained based on the length of the borrowed road segment and the preset driving speed of the intelligent vehicle. When the travel time of the borrowed road segment is less than the average interval time, the mobility feasibility score of the candidate path is determined to be the fifth score, wherein the fifth score is less than the fourth score. When the travel time of the borrowed road segment is greater than or equal to the average transit interval, the mobility feasibility score of the candidate path is determined to be zero.

[0012] Optionally, obtaining the total score based on the multiple score values ​​includes the following formula: A = c1 x S1 + c2 x S2 + c3 x S3; Wherein, A represents the total score of the candidate path, S1 represents the data integrity score of the candidate path, c1 represents the weight value of the data integrity score, S2 represents the safety gap score of the candidate path, c2 represents the weight value of the safety gap score, S3 represents the mobility feasibility score of the candidate path, and c3 represents the weight value of the mobility feasibility score.

[0013] Optionally, the method further includes: The data completeness score, the safety gap score, and the mobility feasibility score are applied to the trained machine learning model to obtain the weight values ​​of the data completeness score, the safety gap score, and the mobility feasibility score. When successfully bypassing the static obstacle, the machine learning model representing the reinforcement learning model is trained by using the data integrity score, the safety gap score, and the mobility feasibility score as input training samples, and the weight values ​​of the data integrity score, the safety gap score, and the mobility feasibility score as output training samples.

[0014] According to a specific embodiment of this application, in a second aspect, this application provides an intelligent vehicle, comprising: Appearance information collector, configured to collect appearance information of static obstacles on the navigation path; The processor is configured to upload appearance information of static obstacles observed at the detection location to the control system, and drive based on the detour path transmitted as described above.

[0015] According to a specific embodiment of this application, in a third aspect, this application provides a control system for controlling the intelligent vehicle described above, comprising: A group control device, which stores a computer program, is configured to execute the computer program to implement the method described above, generate a detour path and transmit it to the processor of the intelligent vehicle.

[0016] Compared with the prior art, the above-described solutions of this application have at least the following beneficial effects: This application provides a method for generating obstacle avoidance paths, a vehicle, and a control system. By repeatedly determining different detection positions, this application enables an intelligent vehicle to collect appearance information of static obstacles at each detection position, thereby improving the local obstacle model of the static obstacle, and finally improving the local obstacle model of the static obstacle on the bypass side. Based on the improved local obstacle model of the bypass side, an obstacle avoidance path is generated for the intelligent vehicle, achieving the goal of effectively avoiding obstacles. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for generating obstacle detour paths according to an embodiment of this application is shown; Figure 2 A schematic diagram of a smart car and a static obstacle according to an embodiment of this application is shown; Figure 3 A schematic diagram of the modules of an intelligent vehicle according to an embodiment of this application is shown; Explanation of reference numerals in the attached figures: 200 - Static obstacle, 201 - Nearest candidate observation position, 202 - Farthest candidate observation position, 203 - Nearest plane, 204 - Farthest plane, 205 - Nearest exposure position, 206 - Farthest exposure position, 207 - Driving direction, 300 - Intelligent vehicle, 301 - Depth information collector, 302 - Processor. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0020] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0021] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0022] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0023] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0024] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0025] The optional embodiments of this application are described in detail below with reference to the accompanying drawings.

[0026] The embodiments provided in this application are embodiments of a method for generating obstacle detour paths.

[0027] The following is combined Figure 1 The embodiments of this application will be described in detail.

[0028] Step S101: When a static obstacle 200 is detected on the navigation path of the intelligent vehicle 300, the appearance information of the static obstacle 200 is collected, and a local obstacle model of the static obstacle 200 is generated in the three-dimensional electronic map based on the appearance information of the static obstacle 200.

[0029] The intelligent vehicle 300 can be either an Automated Guided Vehicle (AGV) or an Infinite Vehicle (IGV).

[0030] Appearance information includes depth information. In the field of computer vision, depth information refers to the distance between an object and the observer in three-dimensional space. In computer vision, depth information allows for an accurate understanding of the positional relationships of objects in three-dimensional space, enabling the fitting of an object's appearance model within that space. In this embodiment, depth information can be acquired using a depth camera, binocular camera, or LiDAR (Light Detection and Ranging) located at the front of the intelligent vehicle 300. By fusing depth information acquired from multiple perspectives using a point cloud registration algorithm, a triangular mesh model of local obstacles is constructed on a three-dimensional electronic map.

[0031] When the intelligent vehicle 300 detects a static obstacle 200 (i.e., an obstacle that has not moved for 3 seconds) on the navigation path, and the static obstacle 200 exceeds the system's preset safety ignore threshold (e.g., a small object with a height < 5cm and a width < 10cm), the detection process for the static obstacle 200 is immediately triggered. At this time, the intelligent vehicle 300 will brake suddenly at a preset distance (e.g., 1m) from the static obstacle 200 and start modeling. A partial obstacle model of the static obstacle 200 is generated in a 3D electronic map using the appearance information of the static obstacle 200. Since the intelligent vehicle 300 can only collect the appearance information of the static obstacle 200 at the point of emergency braking, only partial appearance information of the static obstacle 200 is obtained.

[0032] Step S102: Determine the detection location of the intelligent vehicle 300 based on the three-dimensional electronic map.

[0033] The detection location can be selected near the intelligent vehicle at 300 or near the static obstacle at 200.

[0034] In this embodiment, only local information about the static obstacle 200 can be obtained at the emergency braking point. To bypass the static obstacle 200, the intelligent vehicle 300 needs to determine multiple detection positions based on the detection environment. The appearance information collected from these multiple detection positions is used to continuously improve the local obstacle model, allowing the vehicle to determine a detour path. During this process, the intelligent vehicle 300 needs to continuously determine detection positions based on the current local obstacle model during exploration, allowing it to collect appearance information about unknown parts of the static obstacle 200 from these detection positions.

[0035] In some specific embodiments, determining the detection location of the intelligent vehicle 300 based on the three-dimensional electronic map includes: Step S102-1: Determine candidate observation positions based on the positional relationship between the current position of the intelligent vehicle 300 and the local obstacle model.

[0036] In some specific embodiments, such as Figure 2 As shown, determining candidate observation positions based on the positional relationship between the current position of the intelligent vehicle 300 and the local obstacle model includes: Step S102-11: Determine the farthest exposed position 206 of the local model of the obstacle in the forward detour direction based on the current position of the intelligent vehicle 300.

[0037] Both sides of the static obstacle 200 are directions that can be bypassed.

[0038] The "ahead" refers to the direction 207 of the intelligent vehicle 300's original navigation path.

[0039] The furthest exposed position 206 is the furthest position that can be detected by the device that collects appearance information from the intelligent vehicle 300 on the local model of the obstacle.

[0040] The device that collects appearance information from the intelligent vehicle 300 draws out a cross-section on each of the left and right sides of the local model of the obstacle, and the outermost cross-section is the farthest exposed position 206.

[0041] Step S102-12: Determine the farthest plane 204 containing the farthest exposed position 206.

[0042] The farthest plane 204 is perpendicular to the driving direction 207 of the intelligent vehicle 300 when it is paused.

[0043] Step S102-13: Determine the farthest candidate observation position 202 outside the farthest exposure position 206.

[0044] The farthest candidate observation position 202 is located on the farthest plane 204, and the distance between it and the local model of the obstacle meets the preset safe distance range.

[0045] For example, the preset safe distance range is 30-50cm.

[0046] Step S102-14: Based on the current position of the intelligent vehicle 300, determine the most recently exposed position 205 of the local model of the obstacle in the forward detour direction.

[0047] The most recently exposed position 205 is the position closest to the intelligent vehicle 300 on the local obstacle model.

[0048] Step S102-15: Determine the nearest plane 203 containing the most recently exposed location 205.

[0049] The nearest plane 203 is parallel to the farthest plane 204.

[0050] Step S102-16: Determine the nearest candidate observation location 201 outside the nearest exposure location 205.

[0051] The nearest candidate observation position 201 is located on the nearest plane 203, and the distance between it and the local model of the obstacle meets the preset safe distance range.

[0052] Step S102-17: From the nearest candidate observation position 201 to the farthest candidate observation position 202, including the nearest candidate observation position 201 and the farthest candidate observation position 202, determine the candidate observation position outside the local model of the obstacle.

[0053] In this embodiment, if multiple candidate observation locations are determined, the most suitable candidate observation location can be selected as the detection location. By carefully selecting the detection location, the effectiveness of the observation is improved, avoiding the predicament of being unable to detour due to blind selection of the detection location. At the same time, by selecting the candidate observation location, a safe distance can be maintained between the intelligent vehicle 300 and the static obstacle 200 during the detour process, avoiding damage to the materials transported by the intelligent vehicle 300 from the static obstacle 200.

[0054] Step S102-2: Generate a candidate path from the current position of the intelligent vehicle 300 to the candidate observation position.

[0055] Step S102-3: Score the candidate path to obtain the score value of the candidate path.

[0056] In some specific embodiments, scoring the candidate paths to obtain score values ​​for the candidate paths includes: Steps S102-31: Perform multiple scoring on the candidate paths to obtain multiple score values.

[0057] The multiple scoring values ​​include: data integrity score, safety gap score, and mobility feasibility score. The data integrity score is used to score the integrity of the appearance of the local model of the obstacle. The safety gap score is used to score the safety of the intelligent vehicle 300 passing outside the candidate path. The mobility feasibility score is used to score the feasibility of other intelligent vehicles 300 hindering the intelligent vehicle 300 from detouring.

[0058] In some specific embodiments, the process of performing multiple scoring on the candidate paths to obtain multiple score values ​​includes: Step S102-311a: Generate a virtual surface perpendicular to the horizontal plane based on the virtual navigation path in the three-dimensional electronic map.

[0059] The virtual navigation path is a twin path of the intelligent vehicle 300's navigation path on a 3D electronic map. A virtual surface is created on the 3D electronic map using the virtual navigation path. The virtual surface includes the virtual navigation path.

[0060] Step S102-312a: Divide the local obstacle model into two local sub-models based on the virtual surface.

[0061] Step S102-313a: Determine all local regions based on the local sub-models on the side closest to the candidate path.

[0062] The local sub-model closest to the candidate path is the one closest to the candidate path among the two local sub-models. This indicates that after the intelligent vehicle 300 leaves the original navigation path, the candidate path is closer to this local sub-model.

[0063] Within a local region of the local sub-model, the distance between each data point and its surrounding data points is within a preset distance range. If the distance between a data point and its surrounding data points exceeds the preset distance range, then the data point and its surrounding data points are not in the same local region. The preset distance range is an empirical value obtained through numerous experiments; for example, the preset distance range is a distance greater than 3 cm.

[0064] Step S102-314a: Determine the data integrity score of the candidate path based on the number of all local regions on one side of the candidate path.

[0065] During the observation of static obstacles 200 by the intelligent vehicle 300, discontinuous information inevitably arises due to occlusion, resulting in incomplete surface segment information composed of numerous local regions. The more local regions there are, the lower the integrity of the surface segment information and the lower the data completeness score. In this specific embodiment, the data completeness score of the candidate path is determined by the number of local regions.

[0066] In some specific embodiments, the process of performing multiple scoring on the candidate paths to obtain multiple score values ​​includes: Step S102-311b: Obtain the first minimum distance value between the candidate path and the external environment model, and obtain the first sum of the first half-width value and the preset safety gap value of the intelligent vehicle 300.

[0067] The preset first half-width value of the intelligent vehicle 300 is equal to half of the body width value of the intelligent vehicle 300.

[0068] Step S102-312ba: When the minimum distance value of the candidate path is less than or equal to the preset first half-width value of the intelligent vehicle 300, the safety gap item score value of the candidate path is determined to be zero.

[0069] When the minimum distance value of the candidate path is less than or equal to the preset first half-width value of the intelligent vehicle 300, it indicates that the intelligent vehicle 300 cannot pass.

[0070] The safety gap item score is set to zero so that the first candidate path is disqualified from scoring, and the end point of the candidate path cannot become a detection location.

[0071] Step S102-312bb: When the minimum distance value of the candidate path is greater than the preset first half-width value of the intelligent vehicle 300, and the minimum distance value of the candidate path is less than or equal to the first sum value, the safety gap item score value of the candidate path is determined to be the first score value.

[0072] The first score indicates that the intelligent car 300 can pass, but the width of the path is not wide enough.

[0073] Step S102-312bc: When the minimum distance value of the candidate path is greater than the first sum value, the safety gap item score value of the candidate path is determined to be the second score value.

[0074] The second score is greater than the first score.

[0075] Both the second score and the first score are fixed values.

[0076] The second score indicates that the intelligent car 300 can pass through, but the path is very wide.

[0077] In this specific embodiment, the safety level of the intelligent vehicle 300 using the candidate path is evaluated by the safety gap item score.

[0078] In some specific embodiments, the process of performing multiple scoring on the candidate paths to obtain multiple score values ​​includes: Steps S102-311c: Obtain the second minimum distance value between the candidate path and the paths traversed by the other intelligent vehicles 300; obtain the second sum value of the preset first half-width value of the intelligent vehicle 300 and the preset second half-width value of the other intelligent vehicles 300; and obtain the third sum value of the second sum value and the preset safety gap value.

[0079] Step S102-312ca: When the minimum distance value of the candidate path is greater than the third sum value, the mobility feasibility score of the candidate path is determined to be the third score value.

[0080] The third score indicates that the safe distance between the intelligent vehicle 300 traveling on the candidate path and the intelligent vehicle 300 traveling on other paths is very generous.

[0081] In step S102-312cb, when the minimum distance value of the candidate path is greater than the second sum value and the minimum distance value of the candidate path is less than or equal to the third sum value, the mobility feasibility score of the candidate path is determined to be the fourth score value.

[0082] The fourth score is less than the third score.

[0083] Both the fourth and third scoring values ​​are fixed values.

[0084] The fourth score indicates that the safe distance between the intelligent vehicle 300 traveling on the candidate path and the intelligent vehicle 300 traveling on other paths is barely enough to pass.

[0085] Step S102-312cc: When the minimum distance value of the candidate path is less than or equal to the second sum value and greater than the preset first half-width value of the intelligent vehicle 300, the borrowed road segment in the candidate path is determined.

[0086] The minimum distance value of the candidate path is less than or equal to the second sum value and greater than the preset first half-width value of the intelligent vehicle 300. This can be understood as the intelligent vehicle 300 needing to occupy the driving path of other intelligent vehicles 300 for a short period of time, and bypassing the static obstacle 200 when no other intelligent vehicles 300 are driving.

[0087] Step S102-313c: Obtain the travel time of the borrowed road segment based on the length value of the borrowed road segment and the preset travel speed value of the intelligent vehicle 300.

[0088] Step S102-314ca: When the travel time of the borrowed road segment is less than the average transit interval, the mobility feasibility score of the candidate path is determined to be the fifth score.

[0089] The fifth score is less than the fourth score.

[0090] Both the fourth and fifth scoring values ​​are fixed values.

[0091] The average passing interval refers to the average time interval between the passing of 300 other intelligent vehicles on the adjacent driving path.

[0092] The travel time of the borrowed road section is less than the average interval time, which can be understood as the intelligent vehicle 300 having sufficient time to bypass the static obstacle 200 during the intervals when other intelligent vehicles 300 are traveling.

[0093] Step S102-314cb: When the travel time of the borrowed road segment is greater than or equal to the average transit interval, the mobility feasibility score of the candidate path is determined to be zero.

[0094] If the travel time of the road segment to which the vehicle borrows a pass is greater than or equal to the average interval time, it can be understood that the intelligent vehicle 300 cannot borrow a pass.

[0095] The feasible motion score is set to zero to disqualify the sixth candidate path from being scored, thus making it impossible for the endpoint of the selected path to be detected.

[0096] In this specific embodiment, the ease with which the intelligent vehicle 300 will collide with other intelligent vehicles 300 when it adopts a candidate path is evaluated by the mobility feasibility score.

[0097] Steps S102-32: Obtain the total score based on the multiple score values.

[0098] In some specific embodiments, obtaining the total score based on the multiple score values ​​includes the following formula: A = c1 x S1 + c2 x S2 + c3 x S3; Wherein, A represents the total score of any candidate path, S1 represents the data integrity score of the candidate path, c1 represents the weight value of the data integrity score, S2 represents the safety gap score of the candidate path, c2 represents the weight value of the safety gap score, S3 represents the mobility feasibility score of the candidate path, and c3 represents the weight value of the mobility feasibility score.

[0099] For example, the weight of the data integrity score is 0.4, the weight of the safety gap score is 0.3, and the weight of the mobility feasibility score is 0.3.

[0100] In some specific embodiments, the method further includes: applying the data integrity score, the safety gap score, and the mobility feasibility score to a trained machine learning model to obtain the weight values ​​of the data integrity score, the safety gap score, and the mobility feasibility score.

[0101] Machine learning models are a branch of artificial intelligence that automatically learn from data and make predictions or decisions. They are capable of building models that adapt to new data by analyzing large amounts of data and identifying patterns and regularities, without explicit programming instructions. Examples of machine learning models include reinforcement learning models, Q-learning models, and actor-critic models.

[0102] Reinforcement learning (RL) is a learning method based on reward and punishment feedback mechanisms. An agent learns strategies through interaction with the environment to maximize long-term cumulative rewards. At each step, the agent selects an action, receives reward signals from the environment, and adjusts its strategy accordingly. Reinforcement learning models are suitable for handling continuous decision-making problems.

[0103] Q-Learning is a model-free value function approach that constructs optimal policies by learning the Q-values ​​of state-action pairs. After each interaction, the Q-value is updated based on the immediate reward and the estimated future best outcome until convergence. Q-Learning models are simple and intuitive in structure.

[0104] The Actor-Critic model combines policy optimization (Actor) and value evaluation (Critic). The policy network outputs the action probability distribution, while the value network evaluates the merits of the current policy. The Critic provides optimization direction for the Actor through temporal difference error (TD error), making training more stable.

[0105] In some specific embodiments, the method further includes: When the static obstacle 200 is successfully bypassed, the data integrity score, the safety gap score, and the mobility feasibility score are used as input training samples, and the weights of the data integrity score, the safety gap score, and the mobility feasibility score are used as output training samples to train the machine learning model represented by the reinforcement learning model.

[0106] This specific embodiment provides a reinforcement learning model. The weights c1, c2, and c3 in the overall scoring formula can be dynamically adjusted according to the environment. For example, in scenario one: in a narrow corridor (i.e., the safety gap item score is too low compared to other scores), the weight of c2 (the weight of the full gap item score) is automatically increased; in scenario two: at a busy intersection (i.e., the mobility feasibility item score is too low compared to other scores), the weight of c3 (the weight of the mobility feasibility item score) is automatically increased; in scenario three: in the early stages of exploration, the weight of c1 (the weight of the data completeness item score) is the highest; as the model stabilizes, the weights of c2 and c3 are gradually increased.

[0107] This specific embodiment uses historical successful detour data as training samples and employs a reinforcement learning model to optimize various parameters in the scoring function (including weights and scoring thresholds), enabling the system to evolve itself and find the most efficient observation strategy in a specific scenario.

[0108] Step S102-4: Determine the candidate observation position corresponding to the candidate path with the highest total score as the detection position.

[0109] In this specific embodiment, the detection location is also the safest and most reliable observation location.

[0110] Step S103: Control the intelligent vehicle 300 to move to the detection position to collect new appearance information of the static obstacle 200.

[0111] When the detection location is determined, the intelligent vehicle 300 is instructed to reach the detection location to collect the appearance information of the static obstacle 200, improve the local model of the obstacle, and enhance the continuity of the local model of the obstacle.

[0112] Step S104: Supplement the local model of the obstacle based on the new appearance information.

[0113] Step S105: Generate a detour path based on the supplemented local obstacle model.

[0114] When the intelligent vehicle 300 collects the appearance information of the static obstacle 200, it determines the detection position, collects the appearance information of the static obstacle 200, continuously supplements and improves the local model of the obstacle, and finally generates a detour path through the supplemented local model of the obstacle.

[0115] For methods of generating detour paths based on the supplemented local obstacle model, please refer to existing technologies, which will not be elaborated here.

[0116] This embodiment of the application repeatedly determines various detection positions, enabling the intelligent vehicle 300 to collect appearance information of the static obstacle 200 at each detection position. This process improves the local obstacle model of the static obstacle 200, and ultimately improves the local obstacle model of the static obstacle 200 on the bypass side. Based on the improved local obstacle model on the bypass side, a detour path is generated for the intelligent vehicle 300, achieving the goal of effectively avoiding obstacles.

[0117] like Figure 3 As shown, this application embodiment provides an intelligent vehicle 300, including: Appearance information collector 301 is configured to collect appearance information of static obstacles on the navigation path; The processor 302 is configured to upload the appearance information of the static obstacle observed at the detection location to the control system, and drive based on the detour path transmitted by the method described above.

[0118] This application embodiment repeatedly determines different detection positions, enabling the intelligent vehicle to collect appearance information of static obstacles at each detection position. This process improves the local obstacle model of the static obstacle, ultimately perfecting the local obstacle model of the obstacle on the bypass side. Based on the improved local obstacle model of the bypass side, a detour path is generated for the intelligent vehicle, achieving the goal of effectively avoiding obstacles.

[0119] This application provides a control system for controlling the intelligent vehicle described above, comprising: A group control device, which stores a computer program, is configured to execute the computer program to implement the method described above, generate a detour path and transmit it to the processor of the intelligent vehicle.

[0120] Finally, it should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0121] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating obstacle detour paths, characterized in that, include: When a static obstacle is detected on the navigation path of the intelligent vehicle, the appearance information of the static obstacle is collected, and a local obstacle model of the static obstacle is generated in a three-dimensional electronic map based on the appearance information of the static obstacle. The detection location of the intelligent vehicle is determined based on the three-dimensional electronic map; The intelligent vehicle is controlled to move to the detection position to collect new appearance information of the static obstacle; The local model of the obstacle is supplemented based on the new appearance information; A detour path is generated based on the supplemented local obstacle model.

2. The method according to claim 1, characterized in that, Determining the detection location of the intelligent vehicle based on the three-dimensional electronic map includes: Candidate observation positions are determined based on the positional relationship between the current position of the intelligent vehicle and the local model of the obstacle; Generate a candidate path from the current position of the intelligent vehicle to the candidate observation position; The candidate paths are scored to obtain a score value for each candidate path; The candidate observation location corresponding to the candidate path with the highest score is determined as the detection location.

3. The method according to claim 2, characterized in that, The step of determining candidate observation positions based on the positional relationship between the current position of the intelligent vehicle and the local obstacle model includes: Based on the current position of the intelligent vehicle, determine the farthest exposed position of the local model of the obstacle in the direction that can be bypassed ahead; Determine the farthest plane containing the farthest exposed position, wherein the farthest plane is perpendicular to the driving direction of the intelligent vehicle when it is paused; Determine the farthest candidate observation position outside the farthest exposed position, wherein the farthest candidate observation position is located on the farthest plane and the distance from the local model of the obstacle meets a preset safe distance range; Based on the current position of the intelligent vehicle, determine the most recently exposed position of the local model of the obstacle in the dodgeable direction ahead; Determine the nearest plane containing the most recently exposed location, wherein the nearest plane is parallel to the farthest plane; Determine the nearest candidate observation position outside the nearest exposure position, wherein the nearest candidate observation position is located on the nearest plane and the distance from the local model of the obstacle meets a preset safe distance range; From the nearest candidate observation position to the farthest candidate observation position, including the nearest candidate observation position and the farthest candidate observation position, the candidate observation position outside the local model of the obstacle is determined.

4. The method according to claim 2, characterized in that, The step of scoring the candidate paths to obtain a score value for each candidate path includes: The candidate paths are evaluated using multiple scoring methods to obtain multiple score values, including: data integrity score, safety gap score, and mobility feasibility score. The data integrity score is used to assess the integrity of the appearance of the local model of the obstacle. The safety gap score is used to assess the safety of the intelligent vehicle passing outside the candidate path. The mobility feasibility score is used to assess the feasibility of other intelligent vehicles hindering the intelligent vehicle from detouring. The total score is obtained based on the aforementioned multiple score values.

5. The method according to claim 4, characterized in that, The candidate paths are scored using multiple methods to obtain multiple score values, including: A virtual surface perpendicular to the horizontal plane is generated based on the virtual navigation path in the 3D electronic map; Based on the virtual surface, the local model of the obstacle is divided into two local sub-models; All local regions are determined based on the local sub-models on the side closest to the candidate path; The data integrity score of the candidate path is determined based on the number of all local regions on one side of the candidate path.

6. The method according to claim 4, characterized in that, The candidate paths are scored using multiple methods to obtain multiple score values, including: Obtain the first minimum distance value between the candidate path and the external environment model, and Obtain the first sum of the first half-width value and the preset safety gap value of the intelligent vehicle; When the minimum distance value of the candidate path is less than or equal to the preset first half-width value of the smart car, the safety gap item score value of the candidate path is determined to be zero. When the minimum distance value of the candidate path is greater than the preset first half-width value of the intelligent vehicle, and the minimum distance value of the candidate path is less than or equal to the first sum value, the safety gap item score value of the candidate path is determined to be the first score value. When the minimum distance value of the candidate path is greater than the first sum value, the safety gap item score of the candidate path is determined to be the second score value, wherein the second score value is greater than the first score value.

7. The method according to claim 4, characterized in that, The candidate paths are scored using multiple methods to obtain multiple score values, including: Obtain the second minimum distance value between the candidate path and the paths traversed by the other intelligent vehicles, and Obtain the second sum of the preset first half-width value of the intelligent vehicle and the preset second half-width values ​​of the other intelligent vehicles, and Obtain the third sum of the second sum and the preset safety gap value; When the minimum distance value of the candidate path is greater than the third sum value, the mobility feasibility score of the candidate path is determined to be the third score value. When the minimum distance value of the candidate path is greater than the second sum value, and the minimum distance value of the candidate path is less than or equal to the third sum value, the mobility feasibility score of the candidate path is determined to be the fourth score value, wherein the fourth score value is less than the third score value; When the minimum distance value of the candidate path is less than or equal to the second sum value and greater than the preset first half-width value of the intelligent vehicle, the borrowed road segment in the candidate path is determined; The travel time of the borrowed road segment is obtained based on the length of the borrowed road segment and the preset driving speed of the intelligent vehicle. When the travel time of the borrowed road segment is less than the average interval time, the mobility feasibility score of the candidate path is determined to be the fifth score, wherein the fifth score is less than the fourth score. When the travel time of the borrowed road segment is greater than or equal to the average transit interval, the mobility feasibility score of the candidate path is determined to be zero.

8. The method according to claim 4, characterized in that, The process of obtaining the total score based on the multiple score values ​​includes the following formula: A = c1 x S1 + c2 x S2 + c3 x S3; Wherein, A represents the total score of the candidate path, S1 represents the data integrity score of the candidate path, c1 represents the weight value of the data integrity score, S2 represents the safety gap score of the candidate path, c2 represents the weight value of the safety gap score, S3 represents the mobility feasibility score of the candidate path, and c3 represents the weight value of the mobility feasibility score.

9. The method according to claim 8, characterized in that, The method further includes: The data completeness score, the safety gap score, and the mobility feasibility score are applied to the trained machine learning model to obtain the weight values ​​of the data completeness score, the safety gap score, and the mobility feasibility score. When successfully bypassing the static obstacle, the machine learning model representing the reinforcement learning model is trained by using the data integrity score, the safety gap score, and the mobility feasibility score as input training samples, and the weight values ​​of the data integrity score, the safety gap score, and the mobility feasibility score as output training samples.

10. A smart car, characterized in that, include: Appearance information collector, configured to collect appearance information of static obstacles on the navigation path; The processor is configured to upload appearance information of a static obstacle observed at a detection location to the control system, and to travel based on a detour path transmitted by the method as described in any one of claims 1 to 9.

11. A control system for controlling the intelligent vehicle as described in claim 10, characterized in that, include: A group control device, on which a computer program is stored, is configured to execute the computer program to implement the method as described in any one of claims 1 to 9, generate a detour path and transmit it to the processor of the intelligent vehicle.

Images