Intelligent control method and system for mine underground unmanned equipment

Abstract

The invention discloses a mine underground unmanned equipment intelligent control method and system, and relates to the technical field of intelligent mine equipment automatic control. Comprising the steps of constructing a regulation and control logic map according to a mining roadway center line; the method comprises the following steps: disassembling a long-flow operation task into a plurality of independent subtask units, and training an independent end-to-end task control model; when each task is started, a task navigation map is generated based on the regulation and control logic map, the executed task and the current position; and the front-end equipment triggers the corresponding sub-task control model based on the task navigation graph, the sensing information and the position information, and completes autonomous control of the current operation task through the real-time sensing information in the sub-task stage. Through operation area division, operation task decomposition, action model decoupling, reinforcement learning training, regulation and control of a logic map and an event-driven regulation and control mechanism, stable, reliable and efficient control of unmanned equipment operation under a coarse positioning condition is realized, and the method has good robustness, expandability and generalization.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control of intelligent mining equipment, and in particular to an intelligent control method and system for unmanned equipment in underground mines. Background Art

[0002] The current unmanned equipment in underground mines (such as unmanned shovel loaders, unmanned transport trucks, inspection robots, etc.) mainly relies on the "high-precision mapping - high-precision positioning - path planning" technical framework for automated control. Usually, the mine production control center formulates task processes based on operational requirements and constructs a static three-dimensional environmental map through high-precision sensors (such as lidar, visual SLAM, inertial measurement units, etc.), and then combines path planning algorithms (such as hybrid A*, Dijkstra, RRT, etc.) to generate the equipment's driving path and operation plan, which is ultimately executed by a rule-based controller. This type of technical solution has a certain adaptability in structured and geometrically stable ground scenes, but faces significant constraints in underground mining environments.

[0003] Specifically, underground operating environments are characterized by being enclosed, narrow, dusty, and humid, with drastic changes in lighting. In addition, there is no GNSS signal coverage. Even with the integration of multi-source sensing methods such as UWB and visual SLAM, it is still difficult to achieve sustained, stable, high-precision positioning over long distances and in multiple areas. Especially during operational progress, tunnel structures frequently expand, deform, or become disturbed, leading to a mismatch between existing maps and the real environment, further exacerbating the problems of positioning drift and map failure. Therefore, existing control solutions that rely on high-precision maps and high-precision positioning face challenges such as high-frequency updates, high maintenance costs, poor stability, and poor generalization, making them difficult to support the long-term robust operation of underground operations. Furthermore, current systems often utilize a control approach based on preset paths and rules, lacking sufficient perception-driven and intelligent decision-making capabilities, making them difficult to fully integrate with higher-level scheduling systems. At the operational execution level, key tasks such as loading and unloading still require manual intervention or semi-automatic control, failing to achieve truly autonomous, full-process operation.

[0004] To sum up, the existing unmanned control technology in mines still has obvious shortcomings in the following aspects: First, it relies too much on high-precision maps and high-precision positioning systems, resulting in low stability, high system deployment costs, and heavy maintenance burdens; second, the existing model structure for long tasks is complex, highly coupled, with high training overhead and lack of cross-scenario generalization capabilities, making it difficult to support efficient migration and deployment between multiple tasks and multiple scenarios; third, the control method based on preset paths and rules and the existence of the above problems make it difficult to integrate with the upper-level scheduling system to achieve flexible scheduling.

[0005] Therefore, it is an urgent problem for those skilled in the art to propose an intelligent control method and system for unmanned equipment in underground mines to solve the difficulties existing in the prior art. Summary of the Invention

[0006] In view of this, the present invention provides an intelligent control method and system for unmanned equipment in underground mines. Through operation area division, operation task decomposition, action model decoupling, reinforcement learning training and control logic map and event-driven control mechanism, it realizes stable, reliable and efficient control and scheduling of unmanned equipment operations under coarse positioning conditions, and has good robustness, scalability and generalization.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for intelligently controlling unmanned equipment in underground mines, comprising the following steps:

[0009] S1. Construction of control logic map: Construct a control logic map based on the centerline of the mine roadway;

[0010] S2. Subtask Division and Model Decoupling: Decompose long-process tasks into multiple independent subtask units, and train independent end-to-end subtask control models for each subtask and subtask operation scenario.

[0011] S3. Central task control: When each task is started or an event is triggered, a task navigation map is generated based on the control logic map, the executed task, and the current location;

[0012] S4. Device Perception and Execution: The front-end device triggers the corresponding subtask control model based on the task navigation map, perception information, and location information, and completes the autonomous control of the current stage of the operation task through real-time perception information.

[0013] In the above method, optionally, the control logic map constructed in S1 is a topological map based on the centerline of the lane, and the topological map includes a number of nodes, edges and lengths; wherein the nodes represent the switches or other key scene control points, the edges represent the lanes, and the length is the measured length or the designed length of the lanes.

[0014] In the above method, optionally, the task navigation graph generated in S3 includes an operation path and subtask boundaries, which are used for device operation navigation and subtask switching triggering logic.

[0015] The above method is optional. In S4, a training scenario is constructed in a simulation environment, and a reinforcement learning method is used to train the end-to-end task control model. The input of the subtask control model is the device body perception information, and the output is continuous or discrete control instructions. After the training is completed, it is deployed on the front-end device and fine-tuned according to the real scenario.

[0016] In the above method, optionally, the device body perception information includes images, point clouds, positioning and device working conditions.

[0017] An intelligent control system for unmanned equipment in underground mines, applying any of the above-mentioned intelligent control methods for unmanned equipment in underground mines, comprising a central control module, a perception module, a control execution module and a communication module connected in sequence;

[0018] The central control module is used to generate a task navigation map based on the control logic map and the executed tasks, and send the task navigation map to the front-end device;

[0019] The perception module includes a positioning submodule and a scene recognition submodule, which is used to correct positioning errors or serve as position redundancy by identifying key scene positions;

[0020] The control execution module is used to deploy multiple subtask control models and call and execute corresponding actions based on the perception information and task navigation graph;

[0021] The communication module is used for status synchronization and information exchange between various modules and devices.

[0022] In the above-mentioned system, optionally, the perception module realizes positioning based on SLAM, UWB or wheel odometry, and identifies key scenes, corrects the errors existing in positioning based on SLAM, UWB or wheel odometry through control points on the mission navigation map, or uses the identified position as position information when positioning fails completely.

[0023] In the above system, optional key scene control points include ore exit points, turnout points, and ore unloading points.

[0024] In the above system, optionally, the communication module is used to transmit positioning information, status feedback and mission navigation map.

[0025] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides an intelligent control method and system for unmanned equipment in underground mines, which has the following beneficial effects:

[0026] (1) The present invention constructs a control logic map and a task navigation map on the central side, and implements a subtask control mechanism based on this, thereby constructing a closed-loop scheduling control framework of "task planning on the central side - behavior switching on the device side - model execution on the device side - information feedback on the device side", so that the system can respond to the scheduling control of the upper-level scheduling system in a timely manner.

[0027] (2) Without relying on high-precision maps and high-precision positioning, the present invention realizes stable, reliable and efficient control of unmanned equipment operations under coarse positioning conditions through work area division, work task decomposition, action model decoupling, reinforcement learning training and control logic map and event-driven control mechanism, and has good robustness, scalability and generalization.

[0028] (3) The present invention effectively realizes the decoupling of action models by dividing the complex underground task scenes into multiple simple scenes with a single structure, which is conducive to reinforcement learning training. At the same time, it supports the rapid adaptation of changing scenes and the flexible configuration of operation processes. It has significant engineering deployability and system expansion potential, can be widely used in unmanned operation tasks in various underground scenes, and has good robustness, adaptability and generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0030] Figure 1 This is a flow chart of an intelligent control method for unmanned equipment in underground mines provided by the present invention;

[0031] Figure 2 This is a structural block diagram of an intelligent control system for unmanned equipment in underground mines provided by the present invention;

[0032] Figure 3 A schematic diagram of a control logic map for underground unmanned equipment provided by the present invention;

[0033] Figure 4 A schematic diagram of the underground unmanned equipment mission navigation map provided by the present invention;

[0034] Figure 5 This is a schematic diagram of the intelligent operation of unmanned underground equipment provided by the present invention;

[0035] Figure 6 A schematic diagram of the composition of the underground unmanned equipment provided by the present invention;

[0036] Figure 7 A schematic diagram of the intelligent control process of unmanned underground equipment provided by an embodiment of the present invention;

[0037] Among them, 301-control logic map, 302-control point, 401-sub-map path, 402-walking task area, 403-fork decision area, 404-loading operation area, 501-underground intelligent equipment, 502-perception information, 503-task navigation map. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Reference Figure 1 As shown, the present invention discloses an intelligent control method for unmanned equipment in a mine, comprising the following steps:

[0040] S1. Construction of control logic map: Construct a control logic map based on the centerline of the mine roadway;

[0041] S2. Subtask Division and Model Decoupling: Decompose long-process tasks into multiple independent subtask units, and train independent end-to-end subtask control models for each subtask and subtask operation scenario.

[0042] S3. Central task control: When each task is started or an event is triggered, a task navigation map is generated based on the control logic map, the executed task, and the current location;

[0043] S4. Device Perception and Execution: The front-end device triggers the corresponding subtask control model based on the task navigation map, perception information, and location information, and completes the autonomous control of the current stage of the operation task through real-time perception information.

[0044] Furthermore, the control logic map constructed in S1 is a topological map based on the centerline of the tunnel, which contains several nodes, edges and lengths; among them, the nodes represent the switches or other key scene control points, the edges represent the tunnels, and the length is the measured length or the designed length of the tunnel.

[0045] Furthermore, the task navigation graph generated in S3 includes the running path and subtask boundaries, which are used for device running navigation and subtask switching triggering logic.

[0046] Furthermore, S4 constructs training scenarios in a simulation environment and uses reinforcement learning methods to train end-to-end subtask control models. The input of the subtask control model is the device body perception information, and the output is continuous or discrete control instructions. After training is completed, it is deployed on the front-end device and fine-tuned according to the real scenario.

[0047] Furthermore, the device body perception information includes images, point clouds, positioning and device working conditions.

[0048] In a specific embodiment, Figure 7As shown, the system consists of upper-layer central equipment and front-end equipment. The upper-layer central equipment is responsible for planning production tasks and managing equipment scheduling in the mine area. It includes production task generation, control logic map construction, real-time equipment information, and task navigation maps. Front-end intelligent equipment is deployed in the underground environment and has a certain degree of perception and operation control capabilities.

[0049] (1) Production task generation sets the mining task process according to the mine operation requirements and clarifies the operation sequence and stage goals of each device. The real-time information of the equipment is used to continuously receive the positioning information, operation status and scene perception results uploaded by each front-end device, and combined with the control logic map, it determines the path node and task stage where the device is currently located, forming a global perception of the spatial distribution of the equipment and the progress of task execution on the central side.

[0050] A control logic map is constructed based on the centerlines of the mine tunnels, and the tunnel structure is abstractly modeled to generate a topological structure diagram (control logic map) based on the tunnel centerlines. Unlike urban road navigation maps, underground tunnels are inherently confined spaces, and the vehicle driving area is strictly limited by physical boundaries. The topological map composed of tunnel centerlines is expressed using a graph structure, containing several nodes and edges: nodes are used to represent key locations (such as forks, subtask areas, etc.), and edges represent feasible channels between nodes. The length, as an attribute of the edge, represents the tunnel length to support path planning and dynamic scheduling. This structure ensures that the map remains lightweight while still having complete path connectivity and control expression capabilities. More importantly, using an abstract expression method centered on topology, the construction of the scheduling logic map does not rely on high-precision geometric models, but is based only on tunnel centerlines and task semantic information. When the mine structure expands (such as tunnel advancement) or undergoes local deformation, map reconstruction and scheduling adaptation can be completed by simply updating the node and area definitions. It has good reconfigurability, scalability and generalization capabilities, and is suitable for underground working environments with frequent scene deformation.

[0051] During the actual operation process, the central control module plans the execution path in real time from the global control logic map according to the equipment type, task objectives and current status, and forms a topological subgraph. On the basis of the subgraph, combined with the operation requirements, the path segments are quickly divided into regions according to the principles of "structural simplicity" and "semantic consistency". This division includes task areas such as "walking area", "avoidance area", "loading area", and "unloading area", and each area is bound to a corresponding subtask control model. Ultimately, the subgraph structure and regional division together form a complete "task navigation map" to guide the path execution and model scheduling of the current equipment in the current task cycle. The task subgraph only contains the area information required for the current equipment operation, which not only ensures the scheduling accuracy during system operation, but also significantly reduces the computational burden and data loading pressure of the online system.

[0052] (2) Subtask division and model decoupling: Decompose long-process tasks into multiple independent subtask units, and train independent end-to-end subtask control models for each subtask and subtask operation scenario;

[0053] Specifically, as an example, for shoveling tasks, the task sequence of equipment is formulated according to the functional definition of each shaft and tunnel project in the mining area, and the target areas and operation logic of each stage such as shoveling, transportation, and unloading are clearly defined.

[0054] The long-process operation task is divided into several subtask units with clear functions and limited state space. Each subtask corresponds to an independent subtask control model.

[0055] Each subtask control model is trained end-to-end using reinforcement learning, enabling autonomous decision-making for similar equipment in specific operational scenarios. This avoids the overfitting, policy failure, and generalization deficiencies that can often occur when a single, large model handles tasks across multiple scenarios and states, and structurally enhances the modularity and adaptability of the control system.

[0056] (3) Central task control: When each task is started or an event occurs, a task navigation map is generated based on the control logic map, the executed task, and the current location;

[0057] Specifically, the central control module determines the task stage of the equipment in real time based on the control logic map, equipment positioning information and operation status feedback, and extracts the corresponding topological sub-graph from the global map in combination with the task objectives and equipment types, completes the regional division, and generates a task navigation graph containing path structure and task semantics. This navigation graph has the path information, regional function division and sub-model activation rules required for the current task of the equipment, and is sent to the device end through the communication link. Based on this, the equipment autonomously parses the task stage and loads the corresponding sub-model to perform operations. This mechanism realizes the structured distribution of task control logic by the center, reduces communication load, and improves scheduling efficiency and system robustness.

[0058] (4) Device perception and execution: The front-end device triggers the corresponding subtask control model based on the task navigation map, perception information and location information to complete the autonomous control of the current stage of the task.

[0059] Specifically, the subtask control model trained by reinforcement learning is deployed, and the corresponding subtask control model is activated according to the current task navigation map and the location of the scene, and action reasoning and control execution are performed to achieve specific work tasks such as shoveling, walking, and unloading.

[0060] During the perception process, models based on YOLO V12 and GAnet can be integrated to detect key structures or objects in the work area (such as intersections, avoidance chambers, loading points, unloading platforms, beacons, etc.), thereby determining the type of the current work scene and coarsely positioning the semantic area. This visual recognition process not only improves the sub-model's ability to understand the semantics of the environment, but also can be used to compensate and correct for positioning errors, achieving a fusion of visual perception and motion estimation.

[0061] The perception results can be expressed in the Frenet coordinate system based on the centerline of the topological logical map. The device's current position is projected onto an arc segment using the nearest point, with the longitudinal position denoted as s and the lateral offset d ideally set to d = 0, assuming the device always moves along the centerline.

[0062] In terms of control execution, a sub-model library is deployed locally on the device, integrating multiple end-to-end task control models trained for front-end devices using reinforcement learning. These models correspond to typical operational tasks such as walking, loading, and unloading. After receiving the task navigation map from the central processing unit, the front-end device switches to the corresponding sub-task control model based on its own task area. It then performs real-time action reasoning based on local perception information (such as environmental point clouds, images, and coarse positioning results), thereby autonomously completing the current stage of operational control.

[0063] (5) Subtask control model training. The training method mainly includes the following two stages:

[0064] 1. Simulation training

[0065] During the initial training phase, a typical operating environment is constructed using high-fidelity simulation platforms (such as IsaacLab and Gazebo) to pre-train the control models for different subtasks. Based on the semantic characteristics and operational requirements of each task, corresponding training scenarios are designed (such as narrow channel driving, turning, avoidance, shoveling, and unloading). Appropriate dynamic constraints and interaction rules are set based on the physical parameters of the equipment and the kinematic model. Reinforcement learning methods (such as PPO and DDPG) are used for model training, with a reward function guiding the policy learning to achieve objectives such as task completion efficiency, path smoothness, and safety.

[0066] In response to the operating requirements under the coarse positioning conditions of the present invention, all training scenarios are designed with appropriate boundary expansion to cover the range of positioning errors that may occur in actual deployment, and simulate various postures and initial states of the device entering the area under different positioning deviation conditions, so that the training model has a certain degree of fault tolerance and adaptability. It should be noted that in order to maintain the principle of singleness of scene structure during the training of the subtask control model, the design of the expansion range is not arbitrarily expanded, but is based on the error upper limit of the coarse positioning method adopted, and is controlled within an acceptable accuracy range. Specifically, the expansion boundary only covers the maximum error band that may be generated by the positioning system in actual operation, ensuring that the model has the necessary robustness and generalization ability without destroying the semantic unity and structure of the scene. This design can effectively avoid problems such as unclear model learning objectives, increased scene interference, and decreased strategy generalization ability caused by excessive fuzziness of regional boundaries, thereby ensuring the stability and transferability of the training effect.

[0067] At the same time, to improve the generalization ability and cross-scenario adaptability of the subtask control model in actual deployment, a unified training framework in a diverse semantic environment was constructed during the training process to meet the control requirements of the same type of equipment in different task areas. This invention does not rely on the unique modeling of a specific physical space. Instead, it abstractly extracts the functional structural characteristics (such as path morphology, boundary constraints, etc.) of various operation scenarios based on the consistency of task semantics. In the simulation, multiple scene sample sets with similar structures but diverse environmental differences are constructed to train the same subtask control model.

[0068] This method introduces scene variability and layout disturbances during the strategy training process, prompting the model to learn a robust control strategy. When facing the same type of task objectives, it can still achieve reliable behavioral output in the presence of different topological structures, diverse perception inputs, or positioning error disturbances, thereby significantly enhancing the generalization ability and robustness of the model when deployed in a mining environment.

[0069] 2. Model Deployment

[0070] After training is completed, the model needs to be migrated to the actual mining area for testing and optimization to ensure that it adapts to the real environment and can perform tasks stably.

[0071] First, we conducted multiple rounds of testing in a simulation environment to verify the accuracy of subtask switching and evaluate the stability and performance of the subtask control model under different operating conditions. By comparing the data differences between simulation and real-world environments, we adjusted the normalization method of the model input to reduce domain shift issues.

[0072] During the trial operation phase at the mine, the trained subtask control model was deployed on underground shovel-loaders. Actual operational data, including navigation accuracy and operational efficiency, was collected and evaluated. Because the real-world environment may not match the simulated one, offline reinforcement learning (RL) methods were used to fine-tune the model to better suit the mine's operating conditions and improve robustness.

[0073] In another specific embodiment, a scheduling logic map embodiment and a typical operation process description are provided below:

[0074] like Figure 3 As shown in Figure 3, a typical control logic map includes multiple key structural elements, primarily consisting of a control logic map 301 and control points 302. The tunnel center topology 301 describes the direction, length, and spatial connectivity of the main paths in the mining area, forming the global foundation for task scheduling and path planning. Control points 302, on the other hand, mark key locations involved in task execution, such as forks, loading areas, and unloading areas. They serve as an important reference for subsequently generating task navigation maps and dividing work areas.

[0075] like Figure 4 The figure shows how the central control module dynamically generates a task navigation map based on the control logic map 301, combined with the current equipment type and the production tasks it needs to perform. Subgraph path 401, extracted from the control logic map, is used to define the operating range within the current task cycle; 402 is the corresponding walking task area, 403 is the fork decision area, and 404 is the loading operation area. These task areas are semantically divided by the central control module based on equipment status and task processes, and are bound to corresponding subtask control models to guide the equipment in executing the corresponding control strategies at each stage.

[0076] Combine Figure 5 As shown, the following is an application example of a scheduling logic map in an actual operation process.

[0077] At time T0, the underground intelligent device 501 uploads its current position, perception information 502 and equipment operating status to the central control module, and receives the task navigation map 503 generated by the central control module based on its device model and the current production task. The navigation map clearly plans the path structure, sub-task area division and corresponding control model of the equipment within the current task cycle. According to the guidance of the task navigation map 503, the device starts to execute the "walking model" from time T0, and approaches the fork decision area 403 at time T1. According to the model calling rules embedded in the task navigation map, it selects and switches to the "fork path decision model" from the local model library, completes the left turn operation, correctly enters the designated channel, and realizes path selection and action control.

[0078] like Figure 6 As shown, to implement the method of an embodiment of the present invention, an embodiment of the present invention also provides an intelligent control device for underground unmanned equipment, comprising: an upper-layer hub device and a front-end device, wherein each module includes at least one device power supply, a computing unit, a storage unit, and a wireless network transceiver unit. The various components of the intelligent control device for underground unmanned equipment are coupled together via a bus system. It will be understood that the bus system is used to enable connectivity and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0079] The upper-layer hub device control method disclosed in this embodiment is implemented by a computing unit, which includes multiple processing modules: an information transceiver module, a control map construction module, a task database generation module, and a task scheduling module. The software modules and database reside in a storage unit. The computing unit reads the information from the storage unit and, in conjunction with its hardware, implements the upper-layer hub device control method.

[0080] The front-end device control method disclosed in this embodiment is implemented by a computing unit, which includes multiple processing modules: an information transceiver module, a scene perception module, a control model selection module, a subtask control model, and an action execution module. The software modules and database are located in the storage unit. The computing unit reads the information in the storage unit and, in conjunction with its hardware, implements the front-end device control method.

[0081] and Figure 1 Corresponding to the method described above, the embodiment of the present invention also provides an intelligent control system for unmanned equipment in a mine, which is used to Figure 1 The specific implementation of the method is shown in the following diagram: Figure 2 As shown, it includes a central control module, a perception module, a control execution module and a communication module connected in sequence;

[0082] The central control module is used to generate a task navigation map based on the control logic map and the executed tasks, and send the task navigation map to the front-end device;

[0083] The perception module includes a positioning submodule and a scene recognition submodule, which is used to correct positioning errors or serve as position redundancy by identifying key scene positions;

[0084] The control execution module is used to deploy multiple subtask control models and call and execute corresponding actions based on the perception information and task navigation graph;

[0085] The communication module is used for status synchronization and information exchange between various modules and devices.

[0086] Specifically, front-end intelligent equipment is deployed in the underground environment and has a certain degree of perception and operation control capabilities.

[0087] Furthermore, the perception module realizes positioning based on SLAM, UWB or wheel odometry, and identifies key scenes. It corrects the errors in positioning based on SLAM, UWB or wheel odometry through control points on the mission navigation map, or uses the identified position as position information when positioning fails completely.

[0088] Specifically, the perception module is mainly used to identify the current working environment and task area of the equipment, providing the necessary scene semantic information for the action decision-making of the equipment terminal task control model. As an independent perception system, it relies on multi-source sensors to perceive and process the environmental status in real time. The perception module consists of three parts: positioning submodule, scene recognition submodule and subtask execution perception:

[0089] Positioning submodule: For example, a lightweight, low-cost "positioning" solution based on SLAM or a fusion of UWB and wheel odometry is constructed. This solution does not rely on a high-precision map positioning system. As long as the positioning error is within the maximum error band mentioned above, it can meet the area determination and task switching requirements of the present invention. Coarse positioning information is used to assist in determining the device's location node and its corresponding task area, serving as an important reference for subsequent action model invocation.

[0090] Scene Recognition Submodule: Based on visual perception models (such as YOLO V12 and GAnet) or other lightweight visual recognition networks, the scene recognition submodule identifies the scene type and key areas of the device's current environment. Its main functions include: first, extracting semantic features of the current operating scene to support the identification and determination of the task area; second, serving as a redundancy mechanism, irregularly correcting the position information obtained through positioning.

[0091] Subtask execution perception: As the input of the subtask control model, it includes images, point clouds, positioning, and equipment conditions.

[0092] Furthermore, key scene control points include mining exit points, road junctions, and unloading points.

[0093] Furthermore, the communication module is used to transmit positioning information, status feedback and mission navigation map.

[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0095] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent control method for unmanned equipment in underground mines, characterized in that: The following steps are involved: S1. Construction of control logic map: Construct a control logic map based on the centerline of the mine roadway; S2. Subtask Division and Model Decoupling: Decompose long-process tasks into multiple independent subtask units, and train independent end-to-end subtask control models for each subtask and subtask operation scenario. S3. Central task control: When each task is started or an event is triggered, a task navigation map is generated based on the control logic map, the executed task, and the current location; S4. Device Perception and Execution: The front-end device triggers the corresponding subtask control model based on the task navigation map, perception information, and location information, and completes the autonomous control of the current stage of the operation task through real-time perception information.

2. The intelligent control method for unmanned equipment in underground mines according to claim 1, characterized in that: The control logic map constructed in S1 is a topological map based on the centerline of the lane. The topological map contains several nodes, edges, and lengths. Among them, the nodes represent the switches or other key scene control points, the edges represent the lanes, and the length is the measured length or the designed length of the lanes.

3. The intelligent control method for unmanned equipment in underground mines according to claim 1, characterized in that: The task navigation graph generated in S3 includes the running path and subtask boundaries, which are used for device operation navigation and subtask switching trigger logic.

4. The intelligent control method for unmanned equipment in underground mines according to claim 1, characterized in that: In S4, training scenarios are constructed in a simulation environment, and reinforcement learning methods are used to train end-to-end subtask control models. The input of the subtask control model is the device body perception information, and the output is continuous or discrete control instructions. After training is completed, it is deployed on the front-end device and fine-tuned according to the real scenario.

5. The intelligent control method for unmanned equipment in underground mines according to claim 4, characterized in that: The device body perception information includes images, point clouds, positioning and device working conditions.

6. An intelligent control system for unmanned equipment in underground mines, characterized in that: An intelligent control method for unmanned equipment in underground mines according to any one of claims 1 to 5 is applied, comprising a central control module, a perception module, a control execution module and a communication module connected in sequence; The central control module is used to generate a task navigation map based on the control logic map and the executed tasks, and send the task navigation map to the front-end device; The perception module includes a positioning submodule and a scene recognition submodule, which is used to correct positioning errors or serve as position redundancy by identifying key scene positions; The control execution module is used to deploy multiple subtask control models and call and execute corresponding actions based on the perception information and task navigation graph; The communication module is used for status synchronization and information exchange between various modules and devices.

7. The intelligent control system for unmanned equipment in underground mines according to claim 6, characterized in that: The perception module achieves positioning based on SLAM, UWB or wheel odometry, and identifies key scenes. It corrects the errors in positioning based on SLAM, UWB or wheel odometry through control points on the mission navigation map, or uses the identified position as position information when positioning fails completely.

8. The intelligent control system for unmanned equipment in underground mines according to claim 6, characterized in that: Key scene control points include mining exit points, road junctions, and unloading points.

9. The intelligent control system for unmanned equipment in underground mines according to claim 6, characterized in that: The communication module is used to transmit positioning information, status feedback and mission navigation maps.

Images