A self-adaptive platoon control method for a group of unmanned vehicles on a variable road of an open-pit mine

Abstract

The application discloses a kind of open-pit mine variable road unmanned vehicle group self-adaptive formation control method, and mining truck is fused to perceive external environment by the sensor set by itself, obtain vehicle information in combination with sensor and GPS, obtain road width, curvature information using target detection method based on deep learning;The mining truck with same destination that same direction travels is defined as a vehicle group, and the leader and follower of the mining truck group are determined;Based on the improved A* algorithm, the path of each mining truck in the mining truck group is planned in combination with artificial potential field method;On the basis of 5G network coverage and V2V communication interaction, for different types of variable roads in mine, based on virtual navigation and following method, the formation and transformation of multiple formations of vehicle group, vehicle following, obstacle avoidance action are completed, and the multi-vehicle formation control strategy is designed based on MPC and APC strategy.The application can realize the efficient passing of open-pit mine unmanned mining truck group on variable road, improve transportation efficiency, ensure driving safety, reduce maintenance cost, speed up the scale application of mining truck in open-pit mine, and promote the intelligent process of open-pit mine.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of unmanned vehicle group formation control method, specifically relates to a kind of open-pit mine variable road unmanned vehicle group self-adaptive formation control method. BACKGROUND

[0002] Unmanned technology is a kind of technology that vehicle can perceive surrounding environment by relying on sensor, computer, and plan next time action based on this, and the action is completed by vehicle autonomously, with automatic, efficient, safe, accurate and other outstanding technical advantages.

[0003] In open-pit mine area, mine truck cooperates with excavator, loader and other mine machinery, with efficient transportation, flexible, convenient loading and unloading, strong adaptability, high safety, high reliability, easy maintenance and other advantages, and its number generally reaches dozens of vehicles. With the rapid development of artificial intelligence and other related technologies, mine truck single unmanned driving has gradually matured, and based on effective scheduling, it has basically realized marshalling operation in some open-pit mines. However, the terrain of open-pit mine is complex, and it is easy to appear rough and uneven road. In order to adapt to the terrain and transportation demand of open-pit mine, the road setting is often complex and variable, including curved road with large curvature, narrow road and the like. Under the condition of complex road and variable climate, single intelligent vehicle is difficult to complete efficient operation, and single unmanned driving is obviously insufficient in transportation efficiency and driving safety. Therefore, multi-vehicle cooperative control becomes an important research direction of unmanned driving, and through advanced communication, sensing and control technology, real-time information interaction and cooperative work between multiple mine trucks are realized, so as to improve the efficiency and safety of road transportation system.

[0004] In addition, due to the bad environment, the flow of mine truck on the road is large, and the existence of these variable roads not only affects the driving stability of multiple unmanned mine trucks, but also easily causes problems such as multi-vehicle driving congestion, low transportation efficiency and vehicle collision caused by the decline of sensor function, which increases the risk in transportation process. The current research on mine truck unmanned driving technology mainly focuses on single intelligent vehicle, and the mine truck scheduling is mainly concentrated on the optimal path between loading and unloading points, which cannot effectively meet the requirements of mine truck variable road passing. Therefore, the research on open-pit mine variable road unmanned mine truck multi-vehicle formation control has important significance for enhancing road passing capacity, improving transportation efficiency and ensuring the driving safety of unmanned mine truck. SUMMARY

[0005] In view of the problems existing in the prior art, the present application provides an open-pit mine variable road unmanned vehicle group self-adaptive formation control method, which improves the transportation efficiency of unmanned mine truck vehicle group and ensures the driving safety.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: an adaptive formation control method for unmanned vehicle groups on variable roads in open-pit mines, comprising the following steps:

[0007] Step 1: The unmanned mining truck in the open-pit mine uses multiple sensors to perceive the external environment. It combines state sensors and GPS to obtain state information such as vehicle speed, acceleration, and absolute positioning. It uses a deep learning-based target detection method to identify the changing roads in the mining area and obtain information such as road width and curvature.

[0008] Step 2: Based on the environmental and vehicle information already obtained, define a group of unmanned mining trucks traveling in the same direction and with the same destination as a vehicle group. Trucks can join or leave the group midway. Determine the leader of the unmanned mining truck group, and the rest are followers.

[0009] Step 3: Based on the improved A* algorithm combined with the artificial potential field method, perform path planning for each unmanned mining truck in the unmanned mining truck convoy;

[0010] Step four: Based on the well-developed 5G network coverage and large-scale V2V communication interaction in the mining area, and considering the different types of varied roads in the mine, the virtual navigation and following method is used to complete various formations and changes of the vehicle group, vehicle following, and obstacle avoidance actions. A high-precision multi-vehicle platooning control strategy is designed based on Model Predictive Control (MPC) and Adaptive Control (APC) strategies. The designed control strategy is used to enable the unmanned mining truck group to travel along the predetermined route and formation on varied roads, achieving precise vehicle control.

[0011] Furthermore, the sensors include LiDAR, millimeter-wave radar, depth camera, IMU, GPS, speed sensor, acceleration sensor, pressure sensor, and temperature sensor; the unmanned mining truck is also equipped with an onboard processor. Different sensors perceive the external environment and the truck's own state while it is operating, and record the corresponding data in the onboard processor; the acquisition of absolute positioning information relies on the calculation of the onboard GPS and IMU. Multi-sensor fusion refers to the fusion perception of LiDAR, millimeter-wave radar, depth camera, and IMU: First, the raw data acquired by each sensor is preprocessed. The point clouds of LiDAR and millimeter-wave radar are denoised, compressed, and their coordinates transformed; IMU data is filtered to reduce noise; and depth camera data undergoes depth information correction and image enhancement. Then, data synchronization and joint calibration are performed. Data synchronization involves synchronizing the data from all sensors spatially and temporally so that the fusion algorithm can correctly correlate the data from different sensors. Joint calibration corrects the deviations between sensors, and natural feature points are used to correct the extrinsic and intrinsic parameters between sensors, improving fusion accuracy. Useful features are extracted from the processed data, including the spatial distribution of LiDAR and millimeter-wave radar point clouds, texture information from depth images, and motion trajectories from IMU data. Data fusion directly combines the point clouds from LiDAR and millimeter-wave radar with the depth images at the feature level, using depth information to color the point clouds, increasing semantic information, and also improving the detection accuracy of objects in the depth images. The fused data is used to construct a dataset for road boundary and obstacle detection, real-time map construction, and vehicle localization. In practical applications, the parameters of the fusion algorithm are continuously adjusted based on changes in vehicle behavior and the external environment to optimize the sensor fusion effect. By detecting and identifying the boundaries on both sides of the road, the road width and curvature are determined, and the formation of the unmanned mining truck convoy is adjusted accordingly.

[0012] Furthermore, the deep learning-based target detection method in step one is based on multi-sensor fusion, using the YOLOv5 algorithm to detect unstructured roads and obstacles in open-pit mines, determine road boundaries, and analyze road width and type. The process involves first collecting and labeling a training dataset, including images and corresponding target bounding boxes, category labels, etc.; second, configuring the model's hyperparameters, including learning rate, optimizer type, loss function, confidence threshold, etc., and defining the batch size and number of epochs required during training; then training the model using the training dataset, updating model weights through backpropagation and the Adam optimization algorithm, repeatedly training the model, and calculating the loss function; finally, deploying the trained model to a practical application and performing pruning optimization.

[0013] YOLOv5 algorithm bounding box regression loss:

[0014] L CIOU =λ1·L DIOU +λ2·penalty·(1-α·cos(θ))+λ3·(1-IoU);

[0015] Among them, L DIOU DIOU represents the loss, which measures the distance and overlap area between the predicted and ground truth boxes; α and θ represent the cosine of the aspect ratio of the target box and the predicted box, and the angle difference between the center points of the two boxes, respectively; penalty represents the scale difference between the predicted and ground truth boxes; λ1, λ2, and λ3 represent the loss weights; and IoU represents the result obtained by dividing the overlap of two regions by the sum of the two regions.

[0016] YOLOv5 algorithm's class prediction loss:

[0017] FL(p t )=-α t (1-p t )γlog(p t );

[0018] Where, p t α represents the model's predicted probability for the true label. t γ is the class balance coefficient; γ is a modulating factor used to reduce the weight of easily classified samples.

[0019] Furthermore, in step two, the unmanned mining trucks share their own status information, including destination, pose, absolute positioning, speed, acceleration, etc., through V2V communication technology. Unmanned mining trucks traveling in the same direction and with the same destination send formation requests to each other. Unmanned mining trucks that meet the requirements respond to the requests and prepare to form a formation. In the group of vehicles to be formed, a virtual navigator is used as the lead vehicle of the group through the positioning information shared by each unmanned mining truck, and the others act as followers.

[0020] Furthermore, the path planning in step three is divided into global path planning and local path planning;

[0021] Global path planning is based on an improved A* algorithm. The traditional A* algorithm, after defining the initial and target positions, searches in a directed manner towards the target position, finding the node with the smallest distance F(s) from the current point. Once found, this point is used as the next base point, and the search continues for the point with the smallest distance F(s) from that base point. This process is repeated until the target position is found. The expression for the traditional A* algorithm is:

[0022] F(s) = G(s) + H(s);

[0023] Where F(s) represents the estimated cost from the starting point to the target point; G(s) is the actual cost from the starting point to the current node; and H(s) is the estimated cost from the current node to the target point, also known as the heuristic function.

[0024] The traditional A* algorithm has the problems of many search nodes and small corners and many inflection points in the planned trajectory. By establishing a penalty function and modifying the heuristic function H(s), the number of search nodes can be reduced and the search efficiency can be improved. After obtaining the path planned by the algorithm, cubic spline curves are used to interpolate and fit the turning points of the obtained path to make it smoother, thereby realizing the path planning of a single unmanned mining truck, including the navigator and the follower.

[0025] The modified heuristic function is: H'(s)=(1+α)·H(s);

[0026] Here, α is a penalty coefficient greater than 0, which determines the degree of influence of the penalty function; H(s) is the original heuristic function;

[0027] Therefore, the revised estimated consumption expression is: F(s)=G(s)+(1+α)·H(s);

[0028] Local path planning, or obstacle avoidance, employs an artificial potential field method, treating the target point as a source of attraction and obstacles as sources of repulsion. By constructing gravitational and repulsive potential fields, a potential field function is generated, enabling the unmanned mining truck to successfully avoid obstacles within the potential field space. The gravitational potential field is defined by the distance between the target point and the mining truck; it increases as the truck moves away from the target point and decreases as it approaches. The gravitational potential field function is:

[0029]

[0030] Where, k att d(x,y) is the gravitational gain coefficient; d(x,y) is the distance from the current position (x,y) of the mining truck to the target point.

[0031] When the mining truck approaches an obstacle, the repulsive potential field increases; when the mining truck moves away from the obstacle, the repulsive potential field decreases rapidly. The repulsive potential field function is:

[0032]

[0033] Where, k rep d is the repulsive force gain coefficient; d(x,y) is the distance from the current position (x,y) of the mining card to the obstacle; d0 is the distance affected by the obstacle.

[0034] Combining the gravitational and repulsive potential fields, we obtain the artificial potential field function: U(x,y)=U att (x,y)+Urep (x,y); The force experienced by the unmanned mining truck in the potential field can be expressed as the negative gradient of the potential function:

[0035]

[0036] Based on the calculated potential force, the unmanned mining truck can plan a local route, moving along the direction of the potential force and avoiding obstacles. Furthermore, to address the issue that the artificial potential field method might cause the truck to stop moving at local minima, the potential field function can be adjusted near the local minimum, increasing k. att Or decrease k rep This causes the gravitational potential field to increase or the repulsive potential field to decrease, thereby changing the direction of the potential force, breaking out of the local minimum, and replanning the route to move forward.

[0037] Furthermore, the specific steps of the virtual navigation and following method in step four are as follows:

[0038] S1: First, the shape of the unmanned mining truck convoy is determined to be a straight line formation. The trajectory and speed of the virtual navigator are based on path planning and obtained through pre-defined parameters. Each unmanned mining truck senses its absolute position and attitude information through its own sensors and shares it using V2V technology.

[0039] S2: For each follower mining truck, calculate its expected position, i.e., the relative distance and angle between the virtual navigator and the follower; the positions of the virtual navigator and the follower mining trucks are as follows:

[0040] P leader =[x leader ,y leader ];

[0041] P follower =[x follower ,y follower ];

[0042] Then the follower's expected position P desired for:

[0043] P desired =P leader +T·r;

[0044] Where T is a transformation matrix representing the relative positional relationship between the follower and the virtual navigator; r is a vector representing the relative distance and angle between the follower and the virtual navigator;

[0045] The transformation matrix T is:

[0046]

[0047] Where θ is the angle difference between the follower and the virtual navigator;

[0048] S3: The remaining mining trucks in the vehicle group, i.e. the followers, need to adjust their movement state according to their relative position (distance and angle) with the virtual navigator, and complete the formation of the formation and vehicle following actions based on path planning;

[0049] S4: Calculate the relative position error between each unmanned mining truck and the virtual navigator, and use MPC and APC strategies to adjust its movement to reduce the error and maintain the predetermined formation and spacing.

[0050] The obstacle avoidance system for the convoy first utilizes multi-sensor fusion technology and target detection algorithms on the autonomous mining trucks to perceive the external environment and their own status. These sensors transmit the collected data to the navigator, which uses this data to detect and identify obstacles. Based on information such as the obstacle's position, speed, and size, the navigator formulates an obstacle avoidance strategy using local path planning algorithms. This strategy involves adjusting the truck's speed and trajectory to avoid obstacles. Following the navigator, the trucks receive obstacle avoidance commands and adjust their trajectories and speeds according to their own positions and speeds to avoid collisions. Throughout the obstacle avoidance process, V2V communication and a perception system monitor the situation in real time and make adjustments as needed.

[0051] Furthermore, in step four, the vehicle formation includes single-row and double-row formations. The unmanned mining trucks use sensors to detect and identify the road boundary width and curvature using multi-sensor fusion perception and deep learning-based target detection methods. When the road width is suitable for and exceeds the width of two unmanned mining trucks driving side by side, the unmanned mining truck convoy formation adopts a double-row straight formation with two virtual navigators side by side. When the road width is less than the width of two unmanned mining trucks, a single-row straight formation is adopted with one virtual navigator. The single and double-row straight formations can be interchanged.

[0052] Furthermore, step four involves changing the vehicle formation, including changing from a single-line formation to a double-line formation and vice versa. To achieve this formation change, the target position, speed, and distance between vehicles must first be determined. The navigator formulates a formation change strategy based on the target and conditions. Following vehicles, upon receiving instructions from the navigator, adjust their trajectories and speeds according to their own positions and speeds to adapt to the formation change. During the formation change process, the V2V communication and sensing system monitors the formation change in real time. If any abnormalities are detected, the navigator's speed and instructions should be adjusted promptly to ensure a smooth formation change.

[0053] Transformation from a single-row formation to a double-row formation:

[0054] S1: Number the unmanned mining trucks in the convoy from 0 to N, with 0 as the navigator and 1 to N as the followers; set a new navigator B, and the original navigator A;

[0055] S2: Navigator A sends a formation change command. A and the first N / 2 vehicles (rounded to integer numbers) in the queue gradually increase the distance between themselves and the following vehicles until the required spacing for a double-row formation is reached.

[0056] S3: Navigator B sends a lane change instruction to N / 2 following vehicles. After receiving the instruction, the following vehicles begin to gradually change lanes to the adjacent lane according to the queue.

[0057] S4: After changing lanes, the convoy numbers are updated, with the navigator B as 0 and the followers as new numbers 1 to N. The entire convoy accelerates to a position parallel to convoy A, then decelerates to the same speed as convoy A, forming a double-line formation.

[0058] S5: Navigators A and B adjust their speeds according to the actual situation to maintain the stability of the double-line formation.

[0059] Transformation from a double-row formation to a single-row formation:

[0060] S1: Queue A accelerates, queue B decelerates, and the longitudinal distance between the two queues is gradually increased until the spacing required for a single-line formation is reached.

[0061] S2: Navigator B sends a lane change instruction to its followers, and the followers begin to gradually return to lane A after receiving the instruction;

[0062] S3: After changing lanes, queue B accelerates to queue A, adjusts the distance between the first mining card of queue B and the last mining card of queue A, decelerates to the same speed as queue A, forms a single-line formation, removes the navigator B and updates the numbering, with navigator A as 0 and all followers as the new 1 to N.

[0063] S4: Navigator A adjusts its speed according to the actual situation to maintain formation stability.

[0064] Furthermore, in step four, for different types of roads, including straight roads and curves, formation control strategies for straight roads and formation control strategies for curves are designed based on MPC and APC, respectively. The specific steps are as follows:

[0065] Establish the dynamic differential equations for the unmanned mining truck:

[0066]

[0067] Where m is the mass of the mining truck, and x and y are the positions of the mining truck in a Cartesian coordinate system. It is the speed of the mining truck in the x and y directions. F is the acceleration of the mining truck in the x and y directions, D is the drag coefficient, and δ is the front wheel steering angle.

[0068] The straight-line platooning control employs the Model Predictive Control (MPC) algorithm. The goal of straight-line platooning control is to maintain a fixed distance and speed between mining trucks. MPC uses vehicle dynamics models to predict behavior over a future period and optimizes the control input to minimize the cost function, which is:

[0069]

[0070] Where, x k The state vector at time step k, x ref It is a reference state, u k It is the control input vector at time step k, and Q and R are weight matrices;

[0071] The cornering formation control method employs the Adaptive Path Following Control (APC) algorithm. Cornering formation control needs to consider the vehicle's path following capability. The APC algorithm can adjust the control input based on the vehicle's dynamic characteristics and the geometric characteristics of the path. The path following error is:

[0072] e = y - yd(x);

[0073] Here, yd(x) is the equation of the desired path. The goal of this control strategy is to minimize e and e′ (the first derivative of e). Based on the path following error, a control law is designed to adjust the vehicle's acceleration and steering angle so that the vehicle can accurately follow the desired path.

[0074] Furthermore, during the journey, when one of the unmanned mining trucks in the convoy malfunctions and stops, that unmanned mining truck sends a malfunction information to all the unmanned mining trucks in the convoy. The remaining unmanned mining trucks after that truck will receive the information, re-determine a virtual navigator, generate a new convoy formation, and treat the malfunctioning vehicle as an obstacle. They will then complete the obstacle avoidance actions of the new convoy formation according to the designed obstacle avoidance strategy and continue driving.

[0075] Compared with existing technologies, the beneficial effects of this invention are as follows: It utilizes multi-sensor fusion, including lidar, millimeter-wave radar, depth cameras, and IMUs, to perceive the environment and the autonomous mining truck's own status information; it employs target detection technology to identify the width and type of variable roads in open-pit mines; for different types of mine roads, it adopts a distributed framework, using an improved A* algorithm combined with an artificial potential field method for path planning; and it uses a virtual navigation and following method to complete the formation and transformation of the vehicle convoy, vehicle following, and obstacle avoidance actions, enabling path planning and adaptive convoy driving of autonomous mining truck convoys; utilizing 5G networks and V2V technology, based on MPC and APC, and by designing a high-precision multi-vehicle convoy control strategy, it can achieve precise control of autonomous mining truck convoys traveling along planned trajectories on variable roads, thereby achieving efficient passage of autonomous mining truck convoys on variable roads in open-pit mines, improving transportation efficiency, ensuring driving safety, reducing maintenance costs, accelerating the large-scale application of autonomous mining trucks in open-pit mines, and promoting the intelligentization process of open-pit mines. Attached Figure Description

[0076] Figure 1 This is a flowchart of the formation control method of the present invention.

[0077] Figure 2 This is a flowchart of the multi-sensor fusion process of the present invention.

[0078] Figure 3 This is a flowchart of the object detection method based on deep learning of the present invention.

[0079] Figure 4 This is a flowchart of the virtual navigation and following method of the present invention.

[0080] Figure 5 This is a flowchart of the vehicle formation transformation of the present invention. Detailed Implementation

[0081] The invention will now be further described with reference to the accompanying drawings.

[0082] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0083] like Figure 1 As shown, the present invention provides an adaptive formation control method for unmanned vehicle groups on variable roads in open-pit mines, comprising the following steps.

[0084] Step 1: The unmanned mining truck in the open-pit mine uses a variety of sensors to perceive the external environment. It combines state sensors and GPS to obtain state information such as vehicle speed, acceleration, and absolute positioning. It uses a deep learning-based target detection method to identify the changing roads in the mining area and obtain information such as road width and curvature.

[0085] The sensors used in step one include LiDAR, millimeter-wave radar, depth camera, IMU, GPS, speed sensor, accelerometer, pressure sensor, and temperature sensor; the unmanned mining truck is also equipped with an onboard processor. Different sensors perceive the external environment and the truck's own state while it is operating, and record the corresponding data in the onboard processor; the acquisition of absolute positioning information relies on the calculations performed by the onboard GPS and IMU. Figure 2 As shown, multi-sensor fusion refers to the fusion perception of LiDAR, millimeter-wave radar, depth camera, and IMU. First, the raw data acquired by each sensor is preprocessed: the point clouds from LiDAR and millimeter-wave radar are denoised, compressed, and their coordinates transformed; IMU data is filtered to reduce noise; and depth camera data undergoes depth information correction and image enhancement. Then, data synchronization and joint calibration are performed. Data synchronization involves synchronizing the data from all sensors spatially and temporally to ensure the fusion algorithm can correctly correlate data from different sensors. Joint calibration corrects deviations between sensors, and natural feature points are used to correct the extrinsic and intrinsic parameters between sensors, improving fusion accuracy. The process involves several steps: First, extracting useful features from the processed data, including the spatial distribution of LiDAR and millimeter-wave radar point clouds, texture information from depth images, and motion trajectories from IMU data. Second, data fusion directly combines LiDAR and millimeter-wave radar point clouds with depth images at the feature level. Depth information is used to color the point clouds, increasing semantic information and improving object detection accuracy in depth images. The fused data is then used to construct a dataset for road boundary and obstacle detection, real-time map construction, and vehicle localization. In practical applications, the parameters of the fusion algorithm are continuously adjusted based on changes in vehicle behavior and the external environment to optimize the sensor fusion effect. Finally, by detecting and identifying the boundaries on both sides of the road, the road width and curvature are determined, and the formation of the unmanned mining truck convoy is adjusted accordingly.

[0086] The deep learning-based target detection method in step one is based on multi-sensor fusion, using the YOLOv5 algorithm to detect unstructured roads and obstacles in open-pit mines, determine road boundaries, and analyze road width and type, etc. Figure 3As shown, the process of the deep learning-based object detection method is as follows: First, the training dataset is collected and labeled, including images and corresponding target bounding boxes, category labels, and other information; second, the hyperparameters of the model are configured, including the learning rate, optimizer type, loss function, confidence threshold, etc., and the batch size and number of iterations required during training are defined; then, the model is trained using the training dataset, and the model weights are updated through backpropagation and the Adam optimization algorithm, the model is trained repeatedly, and the loss function is calculated; finally, the trained model is deployed to a practical application, and the model is pruned and optimized.

[0087] YOLOv5 algorithm bounding box regression loss:

[0088] L CIOU =λ1·L DIOU +λ2·penalty·(1-a·cos(θ))+λ3·(1-IoU);

[0089] Among them, L DIOU DIOU represents the loss, which measures the distance and overlap area between the predicted and ground truth boxes; α and θ represent the cosine of the aspect ratio of the target box and the predicted box, and the angle difference between the center points of the two boxes, respectively; penalty represents the scale difference between the predicted and ground truth boxes; λ1, λ2, and λ3 represent the loss weights; and IoU represents the result obtained by dividing the overlap of two regions by the sum of the two regions.

[0090] YOLOv5 algorithm's class prediction loss:

[0091] FL(p t )=-α t (1-p t )γlog(p t );

[0092] Where, p t α represents the model's predicted probability for the true label. t γ is the class balance coefficient; γ is a modulating factor used to reduce the weight of easily classified samples.

[0093] Step two: Based on the environmental and vehicle information already obtained, define a group of unmanned mining trucks traveling in the same direction and with the same destination as a vehicle convoy. Trucks can join or leave the convoy midway. Determine the leader of the unmanned mining truck convoy, and the rest are followers.

[0094] In step two, the unmanned mining trucks share their own status information, including destination, pose, absolute positioning, speed, acceleration, etc., through V2V communication technology. Unmanned mining trucks traveling in the same direction and with the same destination send formation requests to each other. Unmanned mining trucks that meet the requirements respond to the requests and prepare to form a formation. In the group of vehicles to be formed, a virtual navigator is used as the lead vehicle of the group based on the positioning information shared by each unmanned mining truck, and the others act as followers.

[0095] Step 3: Based on the improved A* algorithm combined with the artificial potential field method, perform path planning for each unmanned mining truck in the unmanned mining truck convoy.

[0096] The path planning in step three is divided into global path planning and local path planning;

[0097] Global path planning is based on an improved A* algorithm. The traditional A* algorithm, after defining the initial and target positions, searches in a directed manner towards the target position, finding the node with the smallest distance F(s) from the current point. Once found, this point is used as the next base point, and the search continues for the point with the smallest distance F(s) from that base point. This process is repeated until the target position is found. The expression for the traditional A* algorithm is:

[0098] F(s) = G(s) + H(s);

[0099] Where F(s) represents the estimated cost from the starting point to the target point; G(s) is the actual cost from the starting point to the current node; and H(s) is the estimated cost from the current node to the target point, also known as the heuristic function.

[0100] The traditional A* algorithm has the problems of many search nodes and small corners and many inflection points in the planned trajectory. By establishing a penalty function and modifying the heuristic function H(s), the number of search nodes can be reduced and the search efficiency can be improved. After obtaining the path planned by the algorithm, cubic spline curves are used to interpolate and fit the turning points of the obtained path to make it smoother, thereby realizing the path planning of a single unmanned mining truck, including the navigator and the follower.

[0101] The modified heuristic function is: H'(s)=(1+α)·H(s);

[0102] Here, α is a penalty coefficient greater than 0, which determines the degree of influence of the penalty function; H(s) is the original heuristic function;

[0103] Therefore, the revised estimated consumption expression is: F(s)=G(s)+(1+α)·H(s);

[0104] Local path planning, or obstacle avoidance, employs an artificial potential field method, treating the target point as a source of attraction and obstacles as sources of repulsion. By constructing gravitational and repulsive potential fields, a potential field function is generated, enabling the unmanned mining truck to successfully avoid obstacles within the potential field space. The gravitational potential field is defined by the distance between the target point and the mining truck; it increases as the truck moves away from the target point and decreases as it approaches. The gravitational potential field function is:

[0105]

[0106] Where, k att d(x,y) is the gravitational gain coefficient; d(x,y) is the distance from the current position (x,y) of the mining truck to the target point.

[0107] When the mining truck approaches an obstacle, the repulsive potential field increases; when the mining truck moves away from the obstacle, the repulsive potential field decreases rapidly. The repulsive potential field function is:

[0108]

[0109] Where, k rep d is the repulsive force gain coefficient; d(x,y) is the distance from the current position (x,y) of the mining card to the obstacle; d0 is the distance affected by the obstacle.

[0110] Combining the gravitational and repulsive potential fields, we obtain the artificial potential field function: U(x,y)=U att (x,y)+U rep (x,y); The force experienced by the unmanned mining truck in the potential field can be expressed as the negative gradient of the potential function:

[0111]

[0112] Based on the calculated potential force, the unmanned mining truck can plan a local route, moving along the direction of the potential force and avoiding obstacles. Furthermore, to address the issue that the artificial potential field method might cause the truck to stop moving at local minima, the potential field function can be adjusted near the local minimum, increasing k. att Or decrease k rep This causes the gravitational potential field to increase or the repulsive potential field to decrease, thereby changing the direction of the potential force, breaking out of the local minimum, and replanning the route to move forward.

[0113] Step four: Based on the well-developed 5G network coverage and large-scale V2V communication interaction in the mining area, and considering the different types of varied roads in the mine, the virtual navigation and following method is used to complete various formations and changes of the vehicle group, vehicle following, and obstacle avoidance actions. A high-precision multi-vehicle platooning control strategy is designed based on Model Predictive Control (MPC) and Adaptive Control (APC) strategies. The designed control strategy is used to enable the unmanned mining truck group to travel along the predetermined route and formation on varied roads, achieving precise vehicle control.

[0114] like Figure 4 As shown, the specific steps of the virtual navigation and following method in step four are as follows:

[0115] S1: First, the shape of the unmanned mining truck convoy is determined to be a straight line formation. The trajectory and speed of the virtual navigator are based on path planning and obtained through pre-defined parameters. Each unmanned mining truck senses its absolute position and attitude information through its own sensors and shares it using V2V technology.

[0116] S2: For each follower mining truck, calculate its expected position, i.e., the relative distance and angle between the virtual navigator and the follower; the positions of the virtual navigator and the follower mining trucks are as follows:

[0117] P leader =[x leader ,y leader ];

[0118] P follower =[x follower ,y follower ];

[0119] Then the follower's expected position P desired for:

[0120] P desired =P leader +T·r;

[0121] Where T is a transformation matrix representing the relative positional relationship between the follower and the virtual navigator; r is a vector representing the relative distance and angle between the follower and the virtual navigator;

[0122] The transformation matrix T is:

[0123]

[0124] Where θ is the angle difference between the follower and the virtual navigator;

[0125] S3: The remaining mining trucks in the vehicle group, i.e. the followers, need to adjust their movement state according to their relative position (distance and angle) with the virtual navigator, and complete the formation of the formation and vehicle following actions based on path planning;

[0126] S4: Calculate the relative position error between each unmanned mining truck and the virtual navigator, and use MPC and APC strategies to adjust its movement to reduce the error and maintain the predetermined formation and spacing.

[0127] The obstacle avoidance system for the convoy first utilizes multi-sensor fusion technology and target detection algorithms on the autonomous mining trucks to perceive the external environment and their own status. These sensors transmit the collected data to the navigator, which uses this data to detect and identify obstacles. Based on information such as the obstacle's position, speed, and size, the navigator formulates an obstacle avoidance strategy using local path planning algorithms. This strategy involves adjusting the truck's speed and trajectory to avoid obstacles. Following the navigator, the trucks receive obstacle avoidance commands and adjust their trajectories and speeds according to their own positions and speeds to avoid collisions. Throughout the obstacle avoidance process, V2V communication and a perception system monitor the situation in real time and make adjustments as needed.

[0128] In step four, the vehicle formation includes single-row and double-row formations. The unmanned mining trucks use sensors to detect and identify the road boundary width and curvature using multi-sensor fusion perception and deep learning-based target detection methods. When the road width is suitable for and exceeds the width of two unmanned mining trucks driving side by side, the unmanned mining truck convoy adopts a double-row straight formation with two virtual navigators side by side. When the road width is less than the width of two unmanned mining trucks, a single-row straight formation is adopted with one virtual navigator. The single and double-row straight formations can be interchanged.

[0129] Step four involves changing the vehicle formation, including changing from a single-line formation to a double-line formation and vice versa. To achieve this formation change, the target position, speed, and distance between vehicles must first be determined. The navigator formulates a formation change strategy based on the target and conditions. Following vehicles, upon receiving instructions from the navigator, adjust their trajectories and speeds according to their own positions and speeds to adapt to the formation change. During the formation change, the V2V communication and sensing system monitors the situation in real time. If any abnormalities are detected, the navigator's speed and instructions should be adjusted promptly to ensure a smooth formation change.

[0130] like Figure 5 As shown, a single-row formation is transformed into a double-row formation:

[0131] S1: Number the unmanned mining trucks in the convoy from 0 to N, with 0 as the navigator and 1 to N as the followers; set a new navigator B, and the original navigator A;

[0132] S2: Navigator A sends a formation change command. A and the first N / 2 vehicles (rounded to integer numbers) in the queue gradually increase the distance between themselves and the following vehicles until the required spacing for a double-row formation is reached.

[0133] S3: Navigator B sends a lane change instruction to N / 2 following vehicles. After receiving the instruction, the following vehicles begin to gradually change lanes to the adjacent lane according to the queue.

[0134] S4: After changing lanes, the convoy numbers are updated, with the navigator B as 0 and the followers as new numbers 1 to N. The entire convoy accelerates to a position parallel to convoy A, then decelerates to the same speed as convoy A, forming a double-line formation.

[0135] S5: Navigators A and B adjust their speeds according to the actual situation to maintain the stability of the double-line formation.

[0136] like Figure 5 As shown, the double-column formation is transformed into a single-column formation:

[0137] S1: Queue A accelerates, queue B decelerates, and the longitudinal distance between the two queues is gradually increased until the spacing required for a single-line formation is reached.

[0138] S2: Navigator B sends a lane change instruction to its followers, and the followers begin to gradually return to lane A after receiving the instruction;

[0139] S3: After changing lanes, queue B accelerates to queue A, adjusts the distance between the first mining card of queue B and the last mining card of queue A, decelerates to the same speed as queue A, forms a single-line formation, removes the navigator B and updates the numbering, with navigator A as 0 and all followers as the new 1 to N.

[0140] S4: Navigator A adjusts its speed according to the actual situation to maintain formation stability.

[0141] In step four, different types of roads, including straight roads and curves, are used to design formation control strategies for straight roads and curves based on MPC and APC, respectively. The specific steps are as follows:

[0142] Establish the dynamic differential equations for the unmanned mining truck:

[0143]

[0144] Where m is the mass of the mining truck, and x and y are the positions of the mining truck in a Cartesian coordinate system. It is the speed of the mining truck in the x and y directions. F is the acceleration of the mining truck in the x and y directions, D is the drag coefficient, and δ is the front wheel steering angle.

[0145] The straight-line platooning control employs the Model Predictive Control (MPC) algorithm. The goal of straight-line platooning control is to maintain a fixed distance and speed between mining trucks. MPC uses vehicle dynamics models to predict behavior over a future period and optimizes the control input to minimize the cost function, which is:

[0146]

[0147] Where, x k The state vector at time step k, x ref It is a reference state, u k It is the control input vector at time step k, and Q and R are weight matrices;

[0148] The cornering formation control method employs the Adaptive Path Following Control (APC) algorithm. Cornering formation control needs to consider the vehicle's path following capability. The APC algorithm can adjust the control input based on the vehicle's dynamic characteristics and the geometric characteristics of the path. The path following error is:

[0149] e = y - yd(x);

[0150] Here, yd(x) is the equation of the desired path. The goal of this control strategy is to minimize e and e′ (the first derivative of e). Based on the path following error, a control law is designed to adjust the vehicle's acceleration and steering angle so that the vehicle can accurately follow the desired path.

[0151] Furthermore, during the journey, when one of the unmanned mining trucks in the convoy malfunctions and stops, that unmanned mining truck sends a malfunction information to all the unmanned mining trucks in the convoy. The remaining unmanned mining trucks after that truck will receive the information, re-determine a virtual navigator, generate a new convoy formation, and treat the malfunctioning vehicle as an obstacle. They will then complete the obstacle avoidance actions of the new convoy formation according to the designed obstacle avoidance strategy and continue driving.

[0152] Example:

[0153] The terrain and roads in a certain open-pit mine are complex, so unmanned mining trucks are used for production transportation. These unmanned mining trucks are equipped with lidar, millimeter-wave radar, depth cameras, IMU, GPS, speed sensors, acceleration sensors, pressure sensors and temperature sensors. Through different sensors, the trucks can perceive the external environment and their own status.

[0154] During the unmanned mining truck transportation process, a multi-sensor fusion perception method is used, based on the pruned and optimized YOLOv5 algorithm, to detect unstructured roads in the open-pit mine, unmanned mining trucks, obstacles, etc., and to determine key features such as road boundaries, width, curvature, obstacle size and distance. The vehicle's own speed, acceleration and precise location information are obtained through onboard GPS and IMU. At the same time, each unmanned mining truck uses the improved A* algorithm for global path planning, and on the basis of obstacle identification, it uses the artificial potential field method for local path planning, thereby achieving safe obstacle avoidance.

[0155] First, the unmanned mining trucks in the open-pit mine depart from the parking lot and head to different loading points to perform loading tasks. Unmanned mining trucks located at adjacent loading points within the same loading area, after completing their loading tasks, exit the loading area according to the "first-in, first-out" principle and proceed to the unloading point. Once a target unloading point is given, the unmanned mining trucks share their status information in real time via V2V communication technology and a 5G network, including the target unloading point, pose, absolute positioning, speed, and acceleration. Unmanned mining trucks traveling in the same direction with the same destination send formation requests to each other. Unmanned mining trucks that meet the requirements respond to the requests, and based on the shared positioning information of all unmanned mining trucks, a virtual navigator is used as the lead vehicle in the convoy, with each responding truck acting as a follower, determining the unmanned mining truck convoy formation to be a straight line, ready for formation. The virtual navigator and the follower unmanned mining trucks in the formation-up state complete the formation, vehicle following and obstacle avoidance actions based on the path planning algorithm. The on-board processor calculates the relative position error between each unmanned mining truck and the virtual navigator through the virtual navigator following method, and adopts MPC and APC straight road formation and curve formation control strategies to adjust its own movement, reduce the error, and maintain the predetermined formation and spacing.

[0156] Meanwhile, on the transportation route, a multi-sensor fusion target detection method based on the YOLOv5 algorithm is used to identify the varied roads and obstacles in the mine. For different road types in the mine, such as open sections, narrow sections, and curves, the onboard processor determines in real time whether the road width is suitable for two unmanned mining trucks to travel side-by-side. Then, based on the single / double formation transformation strategy, a virtual navigator-following method is used to adaptively adjust the vehicle formation and ensure vehicles follow each other. For obstacle avoidance, the navigator formulates an obstacle avoidance strategy based on the obstacle's position, speed, and size, combined with a local path planning algorithm. It avoids obstacles by changing vehicle speed and adjusting the mining truck's trajectory. After receiving the navigator's obstacle avoidance instructions, the followers adjust their own trajectory and speed according to their position and speed to avoid collisions with obstacles, enabling multiple unmanned mining trucks to travel along the planned route and formation on varied roads. Upon reaching the vicinity of the unloading point, each follower unmanned mining truck determines the distance between its location coordinates obtained from its onboard GPS and IMU and its respective unloading destination. When this distance equals the set dispersion transport distance, each follower sends and confirms a departure request to the others, allowing them to leave the platoon and proceed to their respective unloading points. After unloading is completed, the above process is repeated to achieve platooning operation of the entire loading, transportation, and unloading process for the unmanned mining truck convoy.

[0157] Furthermore, when a driverless mining truck in the convoy malfunctions and stops, it sends a malfunction information to all other driverless mining trucks in the convoy. The remaining driverless mining trucks after it will receive the information, re-determine a virtual navigator, generate a new convoy formation, and treat the malfunctioning vehicle as an obstacle. They will then complete the obstacle avoidance maneuvers of the new convoy formation according to the designed obstacle avoidance strategy and continue driving.

[0158] Throughout the entire process, supported by the high bandwidth and low latency of the 5G network, the unmanned mining trucks share information in real time via V2V communication technology. Through this process, the unmanned adaptive platooning control method effectively optimizes driving paths and reduces safe following distances, thereby improving road utilization and transportation efficiency in the mining area. Furthermore, dynamic platooning and formation changes enhance the adaptability of unmanned mining truck operations. Simultaneously, based on 5G network and V2V communication, real-time information exchange between the truck groups ensures the transportation safety of unmanned mining trucks in complex and ever-changing environments.

[0159] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0160] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any minor modifications, equivalent substitutions, and improvements made to the above embodiments based on the technical essence of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines, characterized in that, Includes the following steps: Step 1: The unmanned mining truck in the open-pit mine uses its own sensors to perceive the external environment. It combines the sensors with GPS to obtain the vehicle's speed, acceleration, and absolute positioning information. It uses a deep learning-based target detection method to identify the changing roads in the mining area and obtain information on road width and curvature. Step 2: Based on the environmental and vehicle information already obtained, define unmanned mining trucks traveling in the same direction and with the same destination as a vehicle group, determine the leader of the unmanned mining truck group, and the rest are followers. Step 3, based on improvements The algorithm combines the artificial potential field method to perform path planning for each unmanned mining truck in the unmanned mining truck convoy. Step four: Based on 5G network coverage and V2V communication interaction, and considering the different types of varied roads in the mine, the virtual navigation and following method is used to complete various formations and changes of the vehicle group, vehicle following, and obstacle avoidance actions. A high-precision multi-vehicle platooning control strategy is designed based on MPC and APC strategies. The designed control strategy is used to enable the unmanned mining truck group to travel along the predetermined route and formation on varied roads, and to achieve precise vehicle control. The specific steps of the virtual navigation and following method in step four are as follows: S1: The shape of the unmanned mining truck convoy is determined to be a straight line formation. The trajectory and speed of the virtual navigator are based on path planning and obtained through pre-defined parameters. Each unmanned mining truck senses its absolute position and attitude information through its own sensors and shares it using V2V technology. S2: For each follower mining truck, calculate its expected position. The positions of the virtual navigator and follower mining trucks are as follows: ； ； The expected position of the follower for: ； in, It is a transformation matrix that represents the relative positional relationship between the follower and the virtual navigator; It is a vector representing the relative distance and angle between the follower and the virtual navigator; Transformation matrix for: ； in, It is the difference in perspective between the followers and the virtual navigators; S3: The remaining mining trucks in the vehicle group need to adjust their movement state according to their relative position to the virtual navigator, and complete the formation of the formation and vehicle following actions based on path planning; S4: Calculate the relative position error between each unmanned mining truck and the virtual navigator, and use MPC and APC strategies to adjust its movement to reduce the error and maintain the predetermined formation and spacing; Step four involves vehicle formations including single-row and double-row formations. The unmanned mining trucks use sensor fusion and deep learning-based target detection methods to detect and identify the road's boundary width and curvature. When the road width is suitable and exceeds the width of two unmanned mining trucks traveling side-by-side, the unmanned mining truck convoy adopts a double-row straight-line formation with two virtual navigators side-by-side. When the road width is less than the width of two unmanned mining trucks, a single-row straight-line formation is adopted with one virtual navigator. Step four involves different types of roads, including straight roads and curves. Based on MPC and APC, respectively, design formation control strategies for straight roads and curves. The specific steps are as follows: Establish the dynamic differential equations for the unmanned mining truck: ； ； in, It's the quality of the mining card. It is the position of the mining card in a Cartesian coordinate system. It's a mining card. and velocity in the direction, It's a mining card. and acceleration in the direction, It is the driving force. It is the drag coefficient. It is the front wheel steering angle; Straight-line formation control uses the MPC algorithm, and its cost function is: ； in, In time step The state vector, It is a reference state. In time step The control input vector, and It is a weight matrix; The curve formation control uses the APC algorithm, and the path following error is: ； in, It is the equation of the desired path. Based on the path following error, a control law is designed to adjust the vehicle's acceleration and steering angle so that the vehicle can accurately follow the desired path.

2. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 1, characterized in that, The sensors include lidar, millimeter-wave radar, depth camera, IMU, GPS, speed sensor, accelerometer, pressure sensor, and temperature sensor.

3. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 2, characterized in that, The target detection method based on deep learning in step one is based on multi-sensor fusion. It uses the YOLOv5 algorithm to detect unstructured roads and obstacles in open-pit mines, determine road boundaries, and analyze road width and type. The process involves first collecting and labeling the training dataset; then configuring the model's hyperparameters, defining the batch size and number of iterations required during training. Then, the model is trained using the training dataset, and the model weights are updated through backpropagation and the Adam optimization algorithm. The model is trained repeatedly, and the loss function is calculated. Finally, the trained model is deployed to the actual application, and the model is pruned and optimized. YOLOv5 algorithm bounding box regression loss: ； in, This represents the DIOU loss, which measures the distance and overlap area between the predicted bounding box and the ground truth bounding box. , These represent the cosine of the aspect ratio of the target box and the predicted box, and the angle difference between the center points of the two boxes, respectively. This indicates the scale difference between the penalized predicted bounding box and the ground truth bounding box; , , Indicates the loss weight. This represents the result obtained by dividing the overlapping portion of two regions by the combined portion of the two regions. YOLOv5 algorithm's class prediction loss: ； in, This represents the model's predicted probability for the true label; It is the category balance coefficient; It is a moderating factor used to reduce the weight of easily classified samples.

4. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 1, characterized in that, In step two, the unmanned mining trucks share their own status information, including destination, pose, absolute positioning, speed, and acceleration information, through V2V communication technology. Unmanned mining trucks traveling in the same direction and with the same destination send formation requests to each other. Unmanned mining trucks that meet the requirements respond to the requests and prepare to form a formation. In the group of vehicles to be formed, a virtual navigator is used as the lead vehicle of the group based on the positioning information shared by each unmanned mining truck, and the others act as followers.

5. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 1, characterized in that, The path planning in step three is divided into global path planning and local path planning; Global path planning based on improvement The algorithm, and the estimated consumption expression, are as follows: ； in, This represents the estimated travel time from the starting point to the target point. It is the actual consumption from the starting point to the current node. It is the estimated cost from the current node to the target node. It is a penalty coefficient greater than 0, which determines the degree of influence of the penalty function; The local path planning uses the artificial potential field method to construct gravitational and repulsive potential fields at obstacles and build an artificial potential field function so that the unmanned mining truck can successfully avoid obstacles in the potential field space.

6. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 1, characterized in that, The vehicle formation transformation in step four includes transforming a single-line formation into a double-line formation and transforming a double-line formation into a single-line formation. Transformation from a single-row formation to a double-row formation: S1: Number the unmanned mining trucks in the convoy from 0 to N, with the navigator at 0 and the followers at 1 to N; set a new navigator B, and the original navigator A; S2: Navigator A sends a formation change command. Navigator A and the first N / 2 vehicles in the queue gradually increase the distance between themselves and the following vehicles until the required spacing for a double-row formation is reached. S3: Navigator B sends a lane change instruction to N / 2 following vehicles. After receiving the instruction, the following vehicles begin to gradually change lanes to the adjacent lane according to the queue. S4: After changing lanes, the convoy numbers are updated, with the navigator B as 0 and the followers as new numbers 1 to N. The entire convoy accelerates to a position parallel to convoy A, then decelerates to the same speed as convoy A, forming a double-line formation. S5: Navigators A and B adjust their speeds according to the actual situation to maintain the stability of the double-line formation; Transformation from a double-row formation to a single-row formation: S1: Queue A accelerates, queue B decelerates, and the longitudinal distance between the two queues is gradually increased until the spacing required for a single-line formation is reached. S2: Navigator B sends a lane change instruction to its followers, and the followers begin to gradually return to lane A after receiving the instruction; S3: After changing lanes, queue B accelerates to queue A, adjusts the distance between the first mining card of queue B and the last mining card of queue A, decelerates to the same speed as queue A, forms a single-line formation, removes the navigator B and updates the numbering, with navigator A as 0 and all followers as the new 1~N; S4: Navigator A adjusts its speed according to the actual situation to maintain formation stability.

7. The adaptive platooning control method for unmanned vehicle groups on variable roads in open-pit mines according to claim 1, characterized in that, When a driverless mining truck in the vehicle convoy malfunctions and stops during operation, it sends a malfunction notification to all other driverless mining trucks in the convoy. The remaining driverless mining trucks will receive the notification, re-determine a virtual navigator, generate a new vehicle convoy formation, and treat the malfunctioning vehicle as an obstacle. They will then perform obstacle avoidance maneuvers according to the designed obstacle avoidance strategy to continue driving.

Images