Mine transport robot scheduling and path optimization method based on roadway terrain identification

Abstract

The present application relates to the technical field of unmanned coal mine transportation, in particular to a coal mine transportation robot scheduling and path optimization method based on roadway terrain identification, comprising the following steps: S101: acquiring roadway point cloud and vision data in real time; S102: extracting floor terrain feature parameters; S103: constructing and updating a multi-dimensional terrain risk cost map; S104: receiving a transportation task and calculating robot fitness score; S105: performing path search based on improved A* and terrain cost; S106: detecting whether there is a multi-machine path conflict; S107: starting a time-space window-based cooperative avoidance mechanism; S108: issuing path instructions and performing autonomous transportation; S109: making local dynamic adjustments to the path according to real-time feedback. Through the combined perception of the explosion-proof laser radar and the depth camera, the present application realizes the real-time and accurate collection of the roadway terrain in the coal mine underground high-dust and low-light environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned transportation technology in coal mines, and in particular to a method for scheduling and path optimization of coal mine transportation robots based on roadway terrain recognition. Background Technology

[0002] As the coal industry transforms towards intelligent and unmanned operations, unmanned transport robots have become core equipment for underground material and coal transportation in coal mines. Their autonomous scheduling and path planning capabilities directly determine the efficiency, safety, and level of intelligence in underground coal mine transportation.

[0003] The underground environment of coal mine roadways is inherently complex, dynamic, and unique. While existing scheduling and path planning technologies for coal mine transport robots have been applied to some extent, they still have many technical shortcomings and are difficult to adapt to the actual needs of underground operations. For example, traditional path planning technologies rely solely on geometric obstacle avoidance for path searching, and the sensors used are mostly single vision sensors or lidar, which are easily affected by underground dust and uneven lighting, leading to distorted perception data. Furthermore, they fail to accurately extract and quantify core terrain features such as the slope, roughness, and slipperiness of the roadway floor. Path planning focuses solely on minimizing distance, resulting in paths that easily traverse steep slopes, water accumulation, and areas with loose coal deposits. This leads to a high risk of overturning and slipping during robot movement, and unnecessary bumps and climbs significantly increase transport energy consumption. Therefore, it is necessary to design a scheduling and path optimization method for coal mine transport robots based on roadway terrain recognition. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a method for scheduling and path optimization of coal mine transportation robots based on tunnel terrain recognition.

[0005] The technical solution adopted in this invention is: a method for scheduling and path optimization of coal mine transportation robots based on tunnel terrain recognition, comprising the following steps:

[0006] S101: Real-time acquisition of lane point cloud and visual data: The robot uses a sensor array to acquire 3D point cloud data and depth image information of the lane ahead in real time at a preset acquisition frequency. The acquisition range of the sensor array covers the lane area with a preset travel distance in front of the robot.

[0007] S102: Extracting the terrain feature parameters of the floor: The 3D point cloud data and depth image information are cleaned, registered and fused to extract the terrain feature parameters of the tunnel floor;

[0008] S103: Construct and update a multi-dimensional terrain risk cost map: Map the extracted terrain feature parameters to a 2D or 3D grid map according to the actual spatial coordinates of the alleyway, calculate the risk cost of each grid unit in combination with the preset terrain risk cost function, and construct and update a multi-dimensional terrain risk cost map in real time in sync with the sensor acquisition frequency.

[0009] S104: Receive transportation tasks and calculate robot fitness scores: The dispatch center receives underground transportation tasks in the coal mine and parses the task attributes, obtains the real-time status information of all robots in the current work area, calculates the fitness score of each robot based on the preset evaluation model, and selects the robot with the highest fitness score to perform the transportation task.

[0010] S105: Path search based on improved A* and terrain cost: The terrain risk cost in the multi-dimensional terrain risk cost map is introduced into the heuristic function of the improved A* algorithm, and a global path search is performed with distance, stability and energy consumption balance as the objectives to obtain the robot's initial planned path;

[0011] S106: Detect whether there is a multi-robot path conflict: Predict the future travel trajectory of multiple robots within a preset time period through spatiotemporal window technology, and detect whether there is a spatiotemporal overlapping path conflict between the trajectories corresponding to the initial planned paths of each robot. If there is, execute S107; otherwise, execute S108.

[0012] S107: Activate the spatiotemporal window-based cooperative avoidance mechanism: After eliminating path conflicts according to the priority of transportation tasks, re-plan the paths of the relevant robots to obtain the final planned path;

[0013] S108: Issue path instructions and execute autonomous transportation: The dispatch center issues the final planned path instructions to the selected robot. The robot's drive control system parses the control parameters according to the instructions and drives the actuator to perform autonomous transportation operations.

[0014] S109: Local dynamic adjustment of the path based on real-time feedback: During transportation, the robot continuously obtains real-time road condition feedback information through the on-board perception system, and dynamically fine-tunes the driving path through local planning algorithms, forming a closed-loop optimization of terrain perception, path planning, scheduling execution, and dynamic adjustment.

[0015] As a further description of the above technical solution:

[0016] In step S101, the sensor array includes an explosion-proof lidar and a depth camera. The explosion-proof lidar is used to acquire 3D point cloud data of the tunnel; the depth camera is used to acquire depth image information of the tunnel. The preset acquisition frequency is 10-50Hz, and the preset travel distance is 5-20m.

[0017] As a further description of the above technical solution:

[0018] In step S102, a region growing point cloud segmentation algorithm is used to extract the roadway floor area from the fused dataset, excluding interference areas from the roadway sidewalls, top, and fixed equipment. The terrain feature parameters include slope, roughness, and physical properties. The slope is obtained by calculating the angle between the normal vectors of the floor point cloud data using a normal vector analysis algorithm to obtain the floor inclination angle. The roughness is obtained by calculating the root mean square error of the floor point cloud surface to the least squares fitting plane. The physical properties are obtained by analyzing the depth image using multispectral visual recognition technology to determine whether there is water accumulation, loose coal, or floor heave in the roadway floor, and are converted into corresponding wet slip risk parameters.

[0019] As a further description of the above technical solution:

[0020] In step S103, the raster resolution of the raster map is 0.1-0.5m, which can be flexibly adjusted according to the complexity of the roadway terrain; the expression of the terrain risk cost function is: ,in, The terrain risk cost per grid cell, For slope parameters, For roughness parameters, For slippery risk parameters, , , These are the weighting coefficients for each parameter, and The multidimensional terrain risk cost map is dynamically calibrated based on the actual terrain characteristics and robot driving performance in the coal mine. It also includes information on fixed obstacles, temporary obstacles, and risk weights for terrain that is difficult to traverse.

[0021] As a further description of the above technical solution:

[0022] In step S104, the task attributes include the weight of the goods to be transported, the starting and ending points of the transport, and the urgency of the task; the real-time status of the robot includes the remaining battery power, current load, current position, driving speed, and terrain complexity from the starting point of the task; the evaluation model uses the task attribute matching degree and the robot state adaptability as primary evaluation indicators, and each primary indicator has several secondary indicators and is assigned corresponding weights, and the robot adaptability score is calculated by weighted summation.

[0023] As a further description of the above technical solution:

[0024] The heuristic function expression for the improved A* algorithm in step S105 is: ,in, The total cost of searching node n. The actual distance cost from the transportation origin to node n is calculated using grid Euclidean distance. The estimated distance cost from node n to the destination is calculated using the Manhattan distance. The terrain risk cost for the raster cell corresponding to node n is directly extracted from the multidimensional terrain risk cost map; during the path search process, the total cost is selected first. Expand the smallest node.

[0025] As a further description of the above technical solution:

[0026] In step S106, the preset time of the spatiotemporal window is 5-10 seconds. The spatiotemporal coordinates of the future trajectory are calculated based on the robot's current position, initial planned path, and preset travel speed. If the time difference between the trajectories of different robots at the same spatial coordinates is less than a preset threshold, it is determined that there is a path conflict with spatiotemporal overlap. In step S107, the cooperative avoidance mechanism determines the task priority based on the urgency of the transportation task. The low-priority robot finds the nearest refuge chamber in the alleyway through the multi-dimensional terrain risk cost map to temporarily avoid the task. If there is no refuge chamber near the low-priority robot, the robot adjusts its travel speed within the safe speed range to change the trajectory time coordinates until no path conflict is detected.

[0027] As a further description of the above technical solution:

[0028] In step S109, the vehicle-mounted perception system and the sensor array in step S101 are the same set of perception devices. The real-time road condition feedback information includes scattered coal obstacles and equipment failure obstacles on the path, as well as the real-time changes in the slope, roughness, and physical properties of the roadway floor. The local planning algorithm is the DWA dynamic window method, and the local planning algorithm uses the robot's speed, acceleration, and angular velocity as constraints to select the optimal control parameters within the speed window to locally fine-tune the robot's travel path.

[0029] The present invention has the following beneficial effects:

[0030] 1. This invention utilizes a combination of explosion-proof lidar and depth cameras to achieve real-time and accurate acquisition of tunnel terrain in high-dust, low-light environments in coal mines. Combined with algorithms such as point cloud segmentation, normal vector analysis, and multispectral recognition, it accurately extracts core terrain feature parameters such as slope, roughness, and slipperiness. The constructed multi-dimensional terrain risk cost map clearly and quantitatively reflects various high-risk areas within the tunnel. During the path planning stage, the robot automatically avoids terrain prone to overturning, slipping, and jamming, such as deep water accumulation, extremely steep slopes, high roughness, and heaving areas, reducing the robot's driving safety risks from the source. Simultaneously, during transportation, the onboard perception system monitors for sudden obstacles such as scattered coal chunks and equipment malfunctions in real time. The DWA dynamic window method enables rapid local fine-tuning of the path, achieving millisecond-level response to sudden road conditions, further ensuring the robot's driving safety in complex underground environments and reducing the incidence of overturning and slipping failures in coal mine transportation robots.

[0031] 2. This invention incorporates terrain risk cost into the heuristic function of the improved A* algorithm. This allows path planning to prioritize paths with higher flatness, gentler slopes, and less slippery surfaces while considering travel distance. This avoids ineffective travel by the robot on steep slopes and bumpy sections, effectively reducing robot power loss and significantly lowering transportation energy consumption. Simultaneously, the task allocation method based on the fitness evaluation model in step S104 scientifically allocates tasks according to the robot's remaining battery power, current load, terrain adaptability, and other status factors combined with task attributes. This avoids robots with low battery power and high loads performing long-distance, complex terrain transportation tasks, achieving optimal matching between the robot and the transportation task. This further optimizes the overall energy consumption configuration and improves the robot's downhole endurance.

[0032] 3. This invention abandons the rigid scheduling strategy of traditional fixed-path pre-set. The constructed multi-dimensional terrain risk cost map can be updated in real time in sync with the sensor acquisition frequency. When sudden situations such as geological changes, equipment failures, or abrupt terrain changes occur in the tunnel, the scheduling center completes global path replanning within seconds, quickly adapting to the dynamic changes in the tunnel terrain. At the same time, the spatiotemporal window technology accurately predicts the travel trajectories of multiple robots in the next 5-10 seconds, enabling early detection of path conflicts. The collaborative avoidance mechanism based on task priority allows low-priority robots to temporarily avoid conflicts by hiding in avoidance chambers or adjusting their speed, effectively solving the problem of robot congestion in narrow tunnels and realizing dynamic spatiotemporal collaborative scheduling of multiple robots. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0034] The method for scheduling and path optimization of coal mine transportation robots based on tunnel terrain recognition provided by this invention includes the following steps:

[0035] S101: Real-time Acquisition of Tunnel Point Cloud and Visual Data: Data acquisition is completed by a sensor array mounted on the robot's front end. The sensor array consists of an explosion-proof lidar and a depth camera, both meeting the technical standards for explosion-proof, dust-proof, and interference-resistant operation in underground coal mines, and can work stably in high-dust and low-light environments. The sensor array collects data on the tunnel area within a preset travel distance of 5-20m in front of the robot at a preset acquisition frequency of 10-50Hz. The explosion-proof lidar acquires 3D point cloud data of the tunnel in real time, while the depth camera simultaneously acquires depth image information of the tunnel. Both types of data are transmitted in real time to the data fusion processing layer via industrial Ethernet, providing complete and accurate basic perception data for subsequent terrain feature extraction.

[0036] S102: Extracting Floor Terrain Feature Parameters: The data fusion processing layer first preprocesses the received 3D point cloud data and depth image information. A filtering algorithm is used to clean the data, eliminating invalid data caused by dust, noise, and other factors. Then, coordinate registration is used to achieve accurate fusion of the point cloud data and image data. Subsequently, a region growing point cloud segmentation algorithm is used to extract independent regions of the roadway floor from the fused dataset, eliminating the influence of interference areas such as roadway sidewalls, tops, fixed equipment, and supports, retaining only the valid data of the floor. For the extracted roadway floor region, three sets of core terrain feature parameters are calculated sequentially: the angle between the normal vectors of the floor point cloud data is calculated using a normal vector analysis algorithm to obtain the slope parameter reflecting the degree of floor inclination; the least squares method is used to perform plane fitting on the floor point cloud data, calculating the root mean square error from the point cloud surface to the fitted plane to obtain the roughness parameter reflecting the flatness of the floor; and multispectral visual recognition technology is used to perform spectral analysis on the depth image to identify whether there is water accumulation, loose coal, or floor bulging on the roadway floor, and the identification results are converted into a slippery risk parameter reflecting the risk of traction during driving. All extracted terrain feature parameters are transmitted to the dispatch center decision-making level in real time, providing data support for the construction of dynamic terrain risk cost maps.

[0037] S103: Construct and update a multi-dimensional terrain risk cost map: The dispatch center decision-making layer will accurately map the extracted terrain feature parameters such as slope, roughness, and slipperiness risk onto a 2D or 3D raster map according to the actual spatial coordinates of the tunnel. The resolution of the raster map is set to 0.1-0.5m, which can be flexibly adjusted according to the complexity of the tunnel terrain. Define the terrain risk cost function. ,in , , Let be the weight coefficients of each parameter, and satisfy . The terrain risk cost can be dynamically calibrated based on the actual terrain characteristics, robot performance, and load capacity of the coal mine. For key high-risk areas such as water accumulation and steep slopes, the weighting coefficients of corresponding parameters can be appropriately increased. The terrain risk cost for each grid cell is calculated using this cost function. This cost is then integrated with existing information on fixed and temporary obstacles in the grid map to construct a multi-dimensional terrain risk cost map. The map's update frequency is fully synchronized with the data acquisition frequency of the sensor array, enabling real-time dynamic updates of the terrain risk cost and ensuring a high degree of consistency between the map data and changes in the actual terrain of the tunnels.

[0038] S104: Receiving Transportation Tasks and Calculating Robot Fitness Scores: After receiving transportation task instructions from the coal mine, the dispatch center's decision-making layer extracts core task attributes through the task parsing module, including the weight of the goods to be transported, the starting and ending points of the transportation, and the urgency of the task. Simultaneously, through the robot status monitoring module, it obtains real-time status information of all coal mine transportation robots in the current operating area, including remaining battery power, current load, current position, travel speed, and terrain complexity from the task starting point. The dispatch center constructs a fitness evaluation model with task attribute matching degree and robot status adaptability as primary evaluation indicators. Each primary indicator has several secondary indicators, such as cargo weight matching degree and task urgency adaptability under task attribute matching degree, and remaining battery power adaptability, load capacity matching degree, and terrain adaptability adaptability under robot status adaptability, assigning corresponding weights according to the importance of each indicator. A weighted summation method is used to comprehensively score each robot, obtaining the robot's fitness score. A higher fitness score indicates a higher degree of matching between the robot and the current transportation task. The dispatch center selects the robot with the highest fitness score to execute the transportation task.

[0039] S105: Path Search Based on Improved A* and Terrain Cost: The scheduling center decision layer transmits the constructed real-time multi-dimensional terrain risk cost map to the path optimization execution layer in real time. The path optimization execution layer improves the traditional A* algorithm by introducing terrain risk cost into a heuristic function. The improved heuristic function expression is as follows: ,in The total cost of searching node n. The cost is the actual distance from the transportation origin to node n. The estimated distance cost from node n to the destination. This represents the terrain risk cost of the grid cell corresponding to node n. The path optimization execution layer aims to balance travel distance, travel smoothness, and transportation energy consumption. It performs a global path search based on an improved A* algorithm, prioritizing the total cost during the search process. By expanding the smallest node, the robot's initial planned path from the transportation start point to the end point is finally obtained. This path can avoid high-risk terrain to the greatest extent while ensuring a reasonable travel distance, thus achieving a balance between smooth driving and low energy consumption.

[0040] S106: Detecting Multi-Robot Path Conflicts: The path optimization execution layer initiates a multi-robot path conflict detection program, which performs collaborative analysis of the trajectories of multiple robots based on spatiotemporal window technology. First, a spatiotemporal window of 5-10 seconds is set. Based on the current position, initial planned path, and preset driving speed of each robot, the spatiotemporal coordinate information of the robot's trajectory within the preset time is calculated. Then, the spatiotemporal coordinates of all robots' trajectories are compared one by one. If the time difference between the trajectories of different robots at the same spatial coordinate is less than a preset threshold, it is determined that there is a spatiotemporally overlapping path conflict, and the process immediately proceeds to step S107. If there is no spatiotemporal overlap, it is determined that there is no multi-robot path conflict, and the process directly proceeds to step S108.

[0041] S107: Initiating a spatiotemporal window-based collaborative avoidance mechanism: When a multi-machine path conflict is detected, the scheduling center decision layer determines the task priority based on the urgency of each transportation task through the task priority determination module; the higher the urgency of the task, the higher its corresponding priority. The scheduling center sends an avoidance command to the robot corresponding to the low-priority task. The robot quickly searches for and locates the nearest refuge chamber in the alleyway based on its real-time position and multi-dimensional terrain risk cost map, and travels to the chamber for temporary refuge. If there is no refuge chamber nearby, the robot adjusts its speed within its safe speed range and changes the time coordinate of its trajectory to achieve trajectory staggering. During the avoidance process, the path optimization execution layer continuously detects conflicts in the multi-machine trajectories until no spatiotemporally overlapping path conflicts are detected. After the conflict is eliminated, the path optimization execution layer re-plans the path of the low-priority robot globally to obtain the final planned path.

[0042] S108: Issuing Path Instructions and Executing Autonomous Transportation: The dispatch center decision-making level issues the final planned path instructions to the selected robot's drive control system via industrial Ethernet. Upon receiving the path instructions, the drive control system uses the instruction parsing module to convert the path information into specific control parameters such as the robot's speed, steering angle, and braking timing. These control parameters are then transmitted in real-time to the robot's actuators, including wheels, steering system, and braking system, driving the robot to smoothly and efficiently execute autonomous transportation operations according to the final planned path. During transportation, the robot maintains real-time data communication with the dispatch center, continuously reporting its driving status, current position, load information, and sensor data, ensuring the dispatch center can monitor the robot in real-time throughout the entire process.

[0043] S109: Local Dynamic Adjustment of the Path Based on Real-Time Feedback: During autonomous transportation, the onboard perception system (the same device as the sensor array in S101) continuously monitors road conditions ahead at a preset frequency. This real-time feedback data is rapidly transmitted to the path optimization execution layer. The monitored road conditions include sudden obstacles such as scattered coal blocks, equipment malfunctions, and personnel intrusions, as well as real-time dynamic changes in the slope, roughness, and physical properties of the tunnel floor. The path optimization execution layer uses the DWA dynamic window method as a local planning algorithm. This algorithm uses the robot's maximum speed, maximum acceleration, and maximum angular velocity as constraints. Within the robot's speed window, it selects the optimal speed and steering combination through an evaluation function to perform local dynamic fine-tuning of the robot's travel path. The drive control system adjusts the robot's actuator control parameters in real time based on the fine-tuned path instructions to ensure that the robot always travels on a safe and optimal path, forming a closed-loop optimization process from terrain perception, path planning, scheduling execution to dynamic adjustment.

[0044] Specifically:

[0045] S101: Real-time acquisition of tunnel point cloud and visual data:

[0046] This step is accomplished by a sensor array in the perception system layer. The sensor array is mounted at the front end of the coal mine transport robot, at a height of 0.5-1.0m above the ground. Its field of view covers a 180° horizontal and 90° vertical tunnel area in front of the robot. It includes an explosion-proof 32-line LiDAR and an industrial-grade depth camera, both of which have obtained underground explosion-proof certification, achieving an IP67 protection level. They operate stably in underground environments with dust concentrations ≤1000mg / m³ and light intensity of 0-200lx. In this embodiment, the preset acquisition frequency of the sensor array is set to 30Hz, and the preset travel distance is set to 10m. Specifically, the explosion-proof LiDAR acquires 3D point cloud data of the tunnel within a 10m range in front of the robot at a frequency of 30Hz, with a point cloud resolution of 0.05m. Simultaneously, the depth camera acquires depth image information of the same area at a frequency of 30Hz, with an image resolution of 1280×720. The acquired 3D point cloud data and depth image information are transmitted in real time to the edge computing node of the data fusion processing layer via industrial Ethernet at a transmission rate of 100Mbps, providing basic perception data for subsequent terrain feature extraction.

[0047] S102: Extracting the base terrain feature parameters:

[0048] The edge computing nodes of the data fusion processing layer preprocess the received 3D point cloud data and depth image information. First, a statistical filtering algorithm is used to clean the 3D point cloud data, removing noise points that deviate from the main body of the point cloud. A median filtering algorithm is then used to denoise the depth image, eliminating image blur caused by dust. Subsequently, by calibrating the camera and LiDAR extrinsic parameters, coordinate registration of the point cloud data and image data is completed, achieving accurate fusion of the two types of data. A region growing point cloud segmentation algorithm is used, with the geometric features of the tunnel floor as the growth criterion. Independent regions of the tunnel floor are extracted from the fused dataset. By setting height and normal vector thresholds, the influence of interfering areas such as tunnel sidewalls, roof, hydraulic supports, and transportation equipment is eliminated, retaining only the effective point cloud data and image data of the floor. For the extracted tunnel floor region, three sets of core terrain feature parameters are calculated sequentially:

[0049] Slope: Principal component analysis was used to perform normal vector analysis on the base point cloud data. The angle between the normal vector of each point cloud unit and the horizontal plane was calculated, and the average value was taken as the slope parameter for that area. The unit is °; the steeper the slope, the higher the terrain risk.

[0050] Roughness: The least squares method is used to fit a plane to the point cloud data of the base plate, and the equation of the fitted plane is obtained. The perpendicular distance from each point cloud point to the fitted plane is calculated using the formula. Calculate the root mean square error as a roughness parameter, where, For the number of points in the point cloud, The distance from a single point in the point cloud to the fitted plane. The roughness value represents the average distance; a higher roughness value indicates a more uneven base plate.

[0051] Physical characteristics: Multispectral visual recognition technology is used to perform spectral analysis of depth images in four bands: red, green, blue, and near-infrared. Based on the differences in spectral characteristics of water accumulation, loose coal, and hard rock, the presence of water accumulation, loose coal, or floor heave in the roadway floor is identified, and the identification results are converted into a 0-1 wet slip risk parameter. 0 indicates no risk of slippery conditions, and 1 indicates an extremely high risk of slippery conditions.

[0052] All extracted terrain feature parameters are transmitted in real time to the server at the decision-making level of the dispatch center via industrial Ethernet, providing data support for the construction of dynamic terrain risk cost maps.

[0053] S103: Construct and update a multi-dimensional terrain risk cost map:

[0054] The server at the dispatch center's decision-making level extracts terrain feature parameters such as slope, roughness, and slipperiness risk, and accurately maps them onto a 3D raster map according to the actual spatial coordinates (underground Gaussian coordinates) of the underground coal mine roadways. In this embodiment, the raster map's resolution is set to 0.2m, and each raster cell corresponds to an actual underground spatial area of ​​0.2m × 0.2m × 0.2m. A terrain risk cost function is defined. Based on the terrain characteristics of the coal mine in this embodiment (mainly gentle slopes and localized water accumulation), the weight coefficients of each parameter are determined using the analytic hierarchy process (AHP). , , And satisfy The terrain risk cost for each grid cell is calculated based on this cost function. The cost value ranges from 0 to 10, with a higher value indicating a higher terrain risk for the grid cell. The terrain risk cost is fused with existing information on fixed obstacles (such as supports and walls) and temporary obstacles (such as construction equipment and scattered materials) in the grid map to construct a multi-dimensional terrain risk cost map. This map simultaneously includes spatial location, obstacle type, terrain features, and risk cost information. In this embodiment, the update frequency of the multi-dimensional terrain risk cost map is consistent with the data acquisition frequency of the sensor array, i.e., 30Hz, to achieve real-time dynamic updates of the terrain risk cost and ensure that the map data remains highly consistent with changes in the actual terrain of the alleyway.

[0055] S104: Receive the transport task and calculate the robot's fitness score:

[0056] The server at the dispatch center's decision-making level receives transportation task instructions from underground coal mines through the smart mine management system. It then extracts core task attributes, including the weight of the goods to be transported, through the task parsing module. (In this embodiment, the figure is 5t), the starting point (fully mechanized mining face) and the ending point (bottom of the shaft), and the urgency of the task. (In this embodiment, it is an emergency measure, so the value is assigned to 0.9). Simultaneously, the robot status monitoring module obtains real-time status information of the five coal mine transport robots in the current work area, including remaining battery power. Current load Current location Driving speed and the terrain complexity from the mission starting point The scheduling center constructs an fitness evaluation model based on task attribute matching degree. Adaptability to robot state These are primary evaluation indicators, with weights of 0.6 and 0.4 respectively; The following is a description of cargo weight matching. Task urgency suitability The weights are 0.7 and 0.3, respectively; The remaining power level is set below. Load capacity matching degree Terrain adaptability The weights are 0.4, 0.3, and 0.3, respectively. The index score is calculated by the ratio of the actual value to the ideal value of each secondary index, and then the fitness score for each robot is calculated using a weighted summation method. In this embodiment, robot number 3 was calculated to have the highest fitness score (0.92), and the scheduling center selected robot number 3 to perform the transportation task.

[0057] S105: Path search based on improved A* and terrain cost:

[0058] The server at the dispatch center's decision-making level transmits the constructed real-time multi-dimensional terrain risk cost map to the algorithm server at the path optimization execution level via industrial Ethernet. The algorithm server improves the traditional A* algorithm by incorporating terrain risk cost into a heuristic function. The improved heuristic function expression is as follows: In this embodiment, The actual distance cost from the transportation starting point (fully mechanized mining face) to node n is calculated using grid Euclidean distance. The estimated distance cost from node n to the transportation endpoint (bottom yard) is calculated using Manhattan distance to avoid excessively high estimated costs that could reduce the algorithm's search efficiency. The terrain risk cost for the raster cell corresponding to node n is directly extracted from the multi-dimensional terrain risk cost map. The algorithm server, aiming to balance driving distance, driving smoothness, and transportation energy consumption, performs a global path search based on an improved A* algorithm. During the search, a priority queue is used to store nodes to be expanded, prioritizing the selection of nodes with the highest total cost. The smallest node was expanded, and the search step size was set to 0.2m to match the grid resolution. The search yielded the initial planned path for Robot No. 3 from the fully mechanized mining face to the bottom of the shaft. This path avoided high-risk areas with slopes greater than 15° and water depths greater than 0.1m, balancing travel distance and terrain safety.

[0059] S106: Detect whether there are multi-machine path conflicts:

[0060] The algorithm server of the path optimization execution layer initiates a multi-machine path conflict detection program, which performs collaborative analysis of the travel trajectories of five robots within the work area based on spatiotemporal window technology. In this embodiment, the preset time of the spatiotemporal window is set to 8 seconds. Based on the current position, initial planned path, and preset travel speed (1.5 m / s) of each robot, the spatiotemporal coordinate information of each robot's travel trajectory within the next 8 seconds is calculated using a trajectory prediction algorithm, i.e., the arrival time of the robot corresponding to each spatial grid. Subsequently, the spatiotemporal coordinates of all robots' trajectories are compared one by one, with a time threshold of 1 second. If the arrival time difference of different robots' trajectories at the same spatial coordinate is less than 1 second, it is determined that there is a spatiotemporally overlapping path conflict. After detection, robot No. 3 and robot No. 5 will have a path conflict in a narrow section of the underground transport tunnel within the next 5 seconds, and immediately proceed to step S107.

[0061] S107: Initiate a spatiotemporal window-based cooperative avoidance mechanism:

[0062] When a path conflict is detected between robots 3 and 5, the server in the dispatch center's decision-making layer determines the task priority based on the urgency of each transportation task using the task priority determination module. Robot 3, performing an urgent cargo transportation task, has a higher priority than robot 5, which is performing a regular material transportation task. The dispatch center sends an avoidance command to robot 5. Based on its real-time location and a multi-dimensional terrain risk cost map, robot 5 quickly searches for and locates a shelter chamber 8 meters away from its current position. It then adjusts its speed to 0.8 m / s and moves to the shelter chamber. During robot 5's avoidance process, the algorithm server in the path optimization execution layer continuously detects conflicts between the two robots' trajectories at a frequency of 10 Hz. When no spatiotemporally overlapping path conflicts are detected, the algorithm server re-plans the global path for robot 5 to obtain the final planned path, while keeping robot 3's initial planned path unchanged.

[0063] S108: Issue route instructions and execute autonomous transportation:

[0064] The server at the dispatch center's decision-making level sends the initial planned path (final planned path) instruction for Robot No. 3 to the robot's drive control system via industrial Ethernet. After receiving the path instruction, the embedded controller in the drive control system converts the path information into specific control parameters such as the robot's travel speed (1.5 m / s), steering angle, and braking timing through the instruction parsing module. These control parameters are then transmitted in real-time to the robot's actuators, including the hub motors, electric steering system, and hydraulic braking system, via the CAN bus. Based on the control parameters, the actuators drive the robot to travel smoothly and efficiently from the fully mechanized mining face to the bottom yard according to the final planned path, performing autonomous transportation operations. During transportation, the robot maintains real-time data communication with the dispatch center via industrial Ethernet, feeding back its driving status, current position, load information, and sensor-collected data at a frequency of 5 Hz, ensuring the dispatch center's real-time monitoring of the robot throughout the entire process.

[0065] S109: Perform local dynamic adjustments to the path based on real-time feedback:

[0066] During autonomous transportation, the onboard perception system (the same system as the sensor array in S101) continuously monitors road conditions within a 10m radius ahead at a frequency of 30Hz. This real-time feedback data is rapidly transmitted to the algorithm server of the path optimization execution layer via industrial Ethernet. In this embodiment, when the robot reaches the middle section of the transport tunnel, the onboard perception system detects scattered coal blocks (sudden obstacles) 3m ahead, and the area of ​​water accumulation in the local area has expanded compared to the initial map. This road condition information is fed back to the algorithm server in real time. The algorithm server uses the DWA dynamic window method as a local planning algorithm, with constraints including the robot's maximum speed of 2m / s, maximum acceleration of 0.5m / s², and maximum angular velocity of 30° / s. Within the robot's speed window, an evaluation function (including distance cost, speed cost, and obstacle avoidance cost) is constructed. The optimal combination of speed (1.0m / s) and steering angle (10° left deviation) is selected through the evaluation function to perform local dynamic fine-tuning of the robot's path. Based on the fine-tuned path instructions, the drive control system adjusts the control parameters of the robot's hub motors and electric steering system in real time, enabling the robot to smoothly avoid scattered coal blocks and expanding water accumulation areas before resuming its original planned path. Through this local dynamic adjustment, the robot is ensured to always travel on a safe and optimal path, forming a closed-loop optimization process from terrain perception, path planning, scheduling execution to dynamic adjustment.

[0067] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for scheduling and path optimization of coal mine transport robots based on tunnel terrain recognition, characterized in that... This includes the following steps: S101: Real-time acquisition of lane point cloud and visual data: The robot uses a sensor array to acquire 3D point cloud data and depth image information of the lane ahead in real time at a preset acquisition frequency. The acquisition range of the sensor array covers the lane area with a preset travel distance in front of the robot. S102: Extracting the terrain feature parameters of the floor: The 3D point cloud data and depth image information are cleaned, registered and fused to extract the terrain feature parameters of the tunnel floor; S103: Construct and update a multi-dimensional terrain risk cost map: Map the extracted terrain feature parameters to a 2D or 3D grid map according to the actual spatial coordinates of the alleyway, calculate the risk cost of each grid unit in combination with the preset terrain risk cost function, and construct and update a multi-dimensional terrain risk cost map in real time in sync with the sensor acquisition frequency. S104: Receive transportation tasks and calculate robot fitness scores: The dispatch center receives underground transportation tasks in the coal mine and parses the task attributes, obtains the real-time status information of all robots in the current work area, calculates the fitness score of each robot based on the preset evaluation model, and selects the robot with the highest fitness score to perform the transportation task. S105: Path search based on improved A* and terrain cost: The terrain risk cost in the multi-dimensional terrain risk cost map is introduced into the heuristic function of the improved A* algorithm, and a global path search is performed with distance, stability and energy consumption balance as the objectives to obtain the robot's initial planned path; S106: Detect whether there is a multi-robot path conflict: Predict the future travel trajectory of multiple robots within a preset time period through spatiotemporal window technology, and detect whether there is a spatiotemporal overlapping path conflict between the trajectories corresponding to the initial planned paths of each robot. If there is, execute S107; otherwise, execute S108. S107: Activate the spatiotemporal window-based cooperative avoidance mechanism: After eliminating path conflicts according to the priority of transportation tasks, re-plan the paths of the relevant robots to obtain the final planned path; S108: Issue path instructions and execute autonomous transportation: The dispatch center issues the final planned path instructions to the selected robot. The robot's drive control system parses the control parameters according to the instructions and drives the actuator to perform autonomous transportation operations. S109: Local dynamic adjustment of the path based on real-time feedback: During transportation, the robot continuously obtains real-time road condition feedback information through the on-board perception system, and dynamically fine-tunes the driving path through local planning algorithms, forming a closed-loop optimization of terrain perception, path planning, scheduling execution, and dynamic adjustment.

2. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S101, the sensor array includes an explosion-proof lidar and a depth camera. The explosion-proof lidar is used to acquire 3D point cloud data of the tunnel; the depth camera is used to acquire depth image information of the tunnel. The preset acquisition frequency is 10-50Hz, and the preset travel distance is 5-20m.

3. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S102, a region growing point cloud segmentation algorithm is used to extract the roadway floor area from the fused dataset, excluding interference areas from the roadway sidewalls, top, and fixed equipment. The terrain feature parameters include slope, roughness, and physical properties. The slope is obtained by calculating the angle between the normal vectors of the floor point cloud data using a normal vector analysis algorithm to obtain the floor inclination angle. The roughness is obtained by calculating the root mean square error of the floor point cloud surface to the least squares fitting plane. The physical properties are obtained by analyzing the depth image using multispectral visual recognition technology to determine whether there is water accumulation, loose coal, or floor heave in the roadway floor, and converting it into corresponding slippery risk parameters.

4. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S103, the raster resolution of the raster map is 0.1-0.5m, which can be flexibly adjusted according to the complexity of the roadway terrain; the expression of the terrain risk cost function is: ,in, The terrain risk cost per grid cell, For slope parameters, For roughness parameters, For slippery risk parameters, , , These are the weighting coefficients for each parameter, and The multidimensional terrain risk cost map is dynamically calibrated based on the actual terrain characteristics and robot driving performance in the coal mine. It also includes information on fixed obstacles, temporary obstacles, and risk weights for terrain that is difficult to traverse.

5. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S104, the task attributes include the weight of the goods to be transported, the starting point and the destination of the transport, and the urgency of the task; the real-time status of the robot includes the remaining battery power, current load, current position, driving speed, and terrain complexity from the starting point of the task; the evaluation model uses the task attribute matching degree and the robot state adaptability as the primary evaluation indicators, and each primary indicator has several secondary indicators and is assigned corresponding weights. The robot fitness score is calculated by weighted summation.

6. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... The heuristic function expression for the improved A* algorithm in step S105 is as follows: ,in, The total cost of searching node n. The actual distance cost from the transportation origin to node n is calculated using grid Euclidean distance. The estimated distance cost from node n to the destination is calculated using the Manhattan distance. The terrain risk cost for the raster cell corresponding to node n is directly extracted from the multidimensional terrain risk cost map; during the path search process, the total cost is selected first. Expand the smallest node.

7. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S106, the preset time of the spatiotemporal window is 5-10 seconds. The spatiotemporal coordinates of the future trajectory are calculated based on the robot's current position, initial planned path, and preset travel speed. If the time difference between the trajectories of different robots at the same spatial coordinates is less than a preset threshold, it is determined that there is a path conflict with spatiotemporal overlap. In step S107, the cooperative avoidance mechanism determines the task priority based on the urgency of the transportation task. The low-priority robot finds the nearest refuge chamber in the alleyway through the multi-dimensional terrain risk cost map to temporarily avoid the task. If there is no refuge chamber near the low-priority robot, the robot adjusts its travel speed within the safe speed range to change the trajectory time coordinates until no path conflict is detected.

8. The method for scheduling and path optimization of coal mine transport robots based on roadway terrain recognition according to claim 1, characterized in that... In step S109, the vehicle-mounted perception system and the sensor array in step S101 are the same set of perception devices. The real-time road condition feedback information includes scattered coal obstacles and equipment failure obstacles on the path, as well as the real-time changes in the slope, roughness, and physical properties of the roadway floor. The local planning algorithm is the DWA dynamic window method, and the local planning algorithm uses the robot's speed, acceleration, and angular velocity as constraints to select the optimal control parameters within the speed window to locally fine-tune the robot's travel path.

Images