Dynamic path planning system for underground coal mine unmanned aerial vehicle

Abstract

The invention provides an underground coal mine unmanned aerial vehicle dynamic path planning system. Comprising a map construction unit, a risk prediction unit, a cost calculation unit, a path planning unit and an execution feedback unit. The system collects underground environment information through a multi-wire-harness laser radar, a depth camera, a gas sensor and an inertial measurement unit, constructs a real-time three-dimensional voxel map through point cloud splicing and inertial navigation fusion, and records gas concentration, temperature, dust density and SLAM uncertainty; according to the system, gas concentration, temperature, dust density and SLAM uncertainty are continuously recorded by using a three-dimensional voxel map, and a future delta t-oriented space-time risk field is constructed through a gas convection diffusion model, a temperature time sequence model and a micro-seismic threshold model, so that the unmanned aerial vehicle can know a risk evolution trend in advance in a path planning stage; and the problem that in a traditional scheme, a person enters a future high-risk area by mistake while dodging local obstacles is avoided.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety inspection technology, and in particular to a dynamic path planning system for underground unmanned aerial vehicles (UAVs) in coal mines. Background Technology

[0002] The underground environment of coal mines is characterized by narrow roadways, complex structures, insufficient lighting, high dust concentration, and multiple dangerous factors such as gas, temperature, and micro-vibrations. When UAVs perform inspection, monitoring, or mapping tasks underground, the requirements for the real-time performance, safety, and environmental adaptability of the path planning system are extremely high. Most existing underground UAVs adopt static global path planning, which generates a one-time global path through a pre-built map and relies on lidar or depth cameras to perform local obstacle avoidance operations during flight.

[0003] However, this model generally lacks the ability to comprehensively model the dynamic environment downhole. It relies only on the environmental information at the current moment to make local or short-term decisions. It cannot simultaneously assess the overall impact of multiple dynamic factors on path safety, which makes the planned path unable to predict future risk changes or avoid communication blind spots. This may cause the UAV to enter a more risky area in the future after bypassing obstacles locally, or even cause safety hazards such as flight control link interruption.

[0004] To address this, a dynamic path planning system for unmanned aerial vehicles (UAVs) in coal mines is proposed. Summary of the Invention

[0005] In view of this, the present invention provides a dynamic path planning system for unmanned aerial vehicles (UAVs) in coal mines to solve or alleviate the technical problems existing in the prior art, and at least provides a beneficial alternative.

[0006] The technical solution of this invention is implemented as follows: A dynamic path planning system for underground unmanned aerial vehicles in coal mines, comprising:

[0007] The map building unit is used to perform time synchronization on the data collected by LiDAR, depth camera, gas sensor and inertial measurement unit. It generates a real-time updated 3D voxel map through point cloud stitching and inertial navigation fusion algorithm, and writes the ambient gas concentration, temperature and dust information into the corresponding voxels.

[0008] The lidar in the map building unit adopts a multi-beam rotating structure, and the depth camera adopts structured light or TOF imaging. The map building unit generates a dense voxel map through point cloud stitching algorithm and IMU inertial navigation fusion algorithm, and uses the pose covariance output by SLAM to label the uncertainty of the voxel map.

[0009] The risk prediction unit is used to receive a three-dimensional voxel map from the map building unit, and make risk predictions for the voxels at future time Δt based on the gas diffusion equation, temperature time series model and microseismic event triggering model, and generate a spatiotemporal risk field R(x, y, z, t) containing the time dimension.

[0010] The risk prediction unit predicts the risk value of each voxel in the future time Δt by running a gas diffusion model based on convection diffusion partial differential equations, a temperature prediction model based on time series analysis, and a roof fracture model based on microseismic event threshold triggering, and writes the prediction results into the spatiotemporal risk field.

[0011] The cost calculation unit is used to receive the spatiotemporal risk field from the risk prediction unit, input the three-dimensional map, risk field, dynamic obstacle location and communication attenuation matrix into the preset cost function C=α·D+β·R+γ·O+δ·Q, and output the cost matrix for planning.

[0012] The cost function C in the cost calculation unit consists of four matrix terms: α·D, β·R, γ·O, and δ·Q.

[0013] Where D is the three-dimensional geometric distance matrix between voxels, R is the risk prediction matrix, O is the dynamic obstacle occupancy matrix, and Q is the communication signal attenuation matrix. The cost calculation unit automatically adjusts the values ​​of weights α, β, γ, and δ according to the UAV's remaining battery power, mission priority, and environmental risk level through an adaptive weight adjustment algorithm.

[0014] The path planning unit is used to obtain the cost matrix from the cost calculation unit, execute the improved A* or improved RRT* search algorithm within a set time window interval, generate the current optimal flight sub-path of the UAV, and repeat the planning when the time window moves forward to achieve real-time path updates.

[0015] The path planning unit uses FPGA or GPU as a hardware acceleration module. It performs improved A* or improved RRT* search through block graph search and parallel node expansion technology to achieve fast path solving within a set time window. After each rolling cycle, it removes the executed path segments to generate continuous sub-paths.

[0016] The execution feedback unit is used to send the optimal flight sub-path to the UAV flight control system, and receive the pose, IMU data and radar echo matching results uploaded by the UAV. The feedback information is then transmitted to the map building unit to trigger the map update and planning recalculation in the next cycle.

[0017] The execution feedback unit includes a path offset detection module. The module calculates the actual trajectory of the UAV through the echo matching results of the IMU, optical flow sensor and lidar, and determines the trajectory offset based on the trajectory fitting algorithm. When the offset exceeds a set threshold, it sends an instant replanning trigger signal to the path planning unit.

[0018] More preferably, the strength of the communication signal is measured in real time by the wireless communication module, and the measurement result is mapped to the communication signal attenuation matrix Q after being fitted by the fading model. This matrix is ​​updated synchronously with the cost function to penalize path nodes in the communication blind zone.

[0019] More preferably, the system includes a SLAM uncertainty compensation unit, which dynamically adjusts the obstacle expansion radius of map voxels using a Gaussian dilation algorithm based on the pose covariance matrix output by SLAM, so that the UAV automatically increases the safe distance from the lane wall and obstacles in the uncertain positioning area.

[0020] More preferably, the system includes a task coordination unit, which runs an allocation algorithm based on multi-objective optimization, reallocates tasks among multiple UAVs based on UAV remaining battery power, path risk score, task priority matrix and communication quality score, and uses a cooperative avoidance algorithm to generate a set of non-conflicting multi-UAV paths.

[0021] More preferably, the execution feedback unit has a built-in return-to-home safety model. By calculating the current battery level E of the UAV, the return-to-home path length Lr, and the energy consumption per unit distance k, it determines whether Ek·Lr is less than the safety threshold. When the condition is met, the path planning unit is automatically invoked to generate a return-to-home path or switch to the nearest safe landing point path.

[0022] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions:

[0023] I. The system of this invention uses a three-dimensional voxel map to continuously record gas concentration, temperature, dust density and SLAM uncertainty. It constructs a spatiotemporal risk field for the future Δt through a gas convection and diffusion model, a temperature time series model and a microseismic threshold model. This enables UAVs to know the risk evolution trend in advance during the path planning stage, avoiding the problem of "avoiding local obstacles but accidentally entering a high-risk area in the future" in traditional solutions.

[0024] Second, the system of this invention uses a communication attenuation matrix in the cost function to penalize weak and blind communication areas, thereby enabling proactive constraints on the quality of the flight control link during path planning. This reduces the risk of flight loss of control caused by communication interruptions underground. Through rolling time window planning and hardware-accelerated search, it achieves millisecond-level dynamic replanning capability. Combined with the execution feedback unit, it realizes track deviation detection and rapid replanning triggering, enabling the UAV to adjust its path in real time under sudden situations such as gas surges, roof breakage, traffic interference, or communication fading.

[0025] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a system module connection diagram of the present invention. Detailed Implementation

[0028] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0029] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] like Figure 1 As shown, this embodiment of the invention provides a dynamic path planning system for unmanned aerial vehicles (UAVs) in coal mines. The system consists of a map building unit, a risk prediction unit, a cost calculation unit, a path planning unit, an execution feedback unit, and optional SLAM uncertainty compensation unit, task coordination unit, and return-to-home safety model.

[0031] The map building unit in this embodiment includes:

[0032] A multi-beam lidar is used to acquire the 3D structure of the tunnel; a depth camera is used to acquire spatial depth maps and environmental textures; a gas sensor is used to acquire parameters such as gas concentration, temperature, and dust density; an inertial measurement unit (IMU) is used to provide attitude, acceleration, and angular velocity data; a data synchronization module is used to time-align data from different sensors; a point cloud processing module is used to perform point cloud registration, stitching, and dense voxelization; and a SLAM module is used to output the UAV pose and covariance.

[0033] The system achieves continuous positioning through the inertial navigation fusion of IMU and lidar, and then supplements texture and local geometric information through a depth camera, ultimately generating a three-dimensional voxel map that expands in real time with flight.

[0034] Each voxel cell records: gas concentration, temperature, dust density, space occupancy status, and SLAM uncertainty markers, ensuring that the map can be used as input for risk prediction and route planning.

[0035] In this embodiment, the risk prediction unit receives a real-time updated three-dimensional voxel map from the map building unit. The voxels contain environmental information such as gas concentration, temperature, dust density, and micro-seismic monitoring.

[0036] After acquiring this data, the system inputs it as state variables into the risk prediction model:

[0037] (1) For gas risk, the system executes the gas diffusion solution process based on the convection diffusion partial differential equation, and uses the current voxel gas gradient, local wind speed field and gas source term to calculate the change trend of gas concentration with time in the future Δt time, thereby generating the gas risk prediction sequence for each voxel.

[0038] (2) For temperature risk, the system uses ARIMA, LSTM or exponential smoothing time series models to generate short-term temperature predictions by fitting historical temperature sequences and writes the predicted temperatures into the corresponding voxel thermal risk labels.

[0039] (3) For the risk of roof breakage, the system calculates the unstable area that may break the roof in the future based on the real-time triggering record of microseismic events, event energy level, distribution location of microseismic points and energy accumulation changes in the recent time period using the set microseismic threshold model, and maps the prediction results to the voxel structure.

[0040] The three risk prediction processes mentioned above are uniformly stacked in the time dimension, thereby generating a risk value sequence with time extension characteristics for each voxel. The risk prediction unit integrates gas risk, temperature risk, roof breakage risk and other environmental risks into a four-dimensional spatiotemporal risk field R(x, y, z, t) according to a preset fusion method. This spatiotemporal risk field contains the estimated curves of the risks at each location within the future Δt, which can be continuously used as input for cost calculation and dynamic path planning.

[0041] In this embodiment, the cost calculation unit is used to construct the cost matrix required for path planning under multi-source dynamic environmental factors. It includes four main stages: data acquisition, cost function item calculation, adaptive weight adjustment, and matrix generation.

[0042] After receiving the spatiotemporal risk field output by the risk prediction unit, the system obtains the three-dimensional geometric distance D between each voxel from the three-dimensional voxel map, and forms a distance matrix by calculating the relationship between the node connection side length and the spatial structure distance. Then, based on R(x, y, z, t) output by the risk prediction unit, the system extracts the risk slice corresponding to the planning time and writes it into the risk matrix R, which is used to reflect the potential risk level of each node in the planning time window. At the same time, the system also obtains the dynamic obstacle positions from the execution feedback unit and the map building unit, and discretizes the occupancy state of these obstacles in space into an occupancy matrix O, so that it assigns a higher penalty value to the obstacle cluster area in the cost function. Real-time RSSI, RTT and SINR data are obtained from communication, and the communication attenuation matrix Q is obtained by fitting the wireless channel fading model, so that the path planning can suppress the entry into weak communication areas, blind areas and areas where communication may be lost.

[0043] After the above four types of matrices are prepared, the adaptive weight adjustment stage begins. The cost calculation unit dynamically adjusts the four weights α, β, γ, and δ in the cost function based on the drone's current remaining battery power, mission urgency, environmental risk level, and communication quality, to ensure that the priority of path planning under different mission scenarios can correctly reflect the system requirements.

[0044] For example, when the task is urgent and the environmental risk is low, the system can increase the distance weight α to shorten the total task path; when the risk of gas or temperature increases, the risk weight β is increased to prioritize avoiding high-risk areas; when there are many obstacles or dense underground traffic, γ is increased to suppress obstacle nodes; when communication attenuation is severe, δ is increased to prioritize areas with stable communication; according to the cost function C=α·D+β·R+γ·O+δ·Q, the cost value of each node is calculated point by point, forming a cost value matrix and outputting it to the path planning unit to achieve a comprehensive evaluation of risk, distance, obstacles and communication factors.

[0045] In this embodiment, the path planning unit generates real-time executable optimal flight sub-paths under the constraints of dynamic environment and time-varying costs. It comprises three main parts: hardware acceleration, rolling time window updates, and optimal sub-path stitching. First, it receives the cost matrix generated by the cost calculation unit and loads it into the planning memory area. To meet the stringent time response requirements of the downhole environment, this embodiment uses an FPGA or GPU as the planning acceleration hardware. A block graph search strategy divides the global map into multiple independently computable subgraphs, and parallelization of node expansion is implemented on the hardware platform, enabling the A* or RRT* algorithm to complete the heuristic expansion of tens of thousands of nodes within milliseconds. Simultaneously, by introducing spatiotemporal risk estimation into the heuristic function, the search direction is made more inclined towards low-risk areas, improving planning accuracy and stability.

[0046] During path planning, the system uses a rolling time window mechanism for continuous planning based on the spatiotemporal risk field and cost matrix of the future Δt. When the time window begins, the system searches for currently feasible paths and generates the optimal flight sub-path from the UAV's current position to the end of the time window. Subsequently, the system advances the time window periodically, deletes executed path segments, and recalculates the next path segment, enabling the UAV's trajectory to continuously follow dynamic environmental changes. At the same time, when the yaw amount provided by the execution feedback unit exceeds the threshold, communication fading reaches the danger zone, or the local risk level suddenly increases, the system immediately triggers an instant replanning command, causing the path planning unit to re-search for the optimal sub-path within the current time window.

[0047] In this embodiment, the execution feedback unit is mainly used to realize the entire feedback link of UAV path execution, track deviation monitoring and map closed-loop update. By receiving the optimal flight sub-path generated by the path planning unit, it encodes it according to the instruction format required by the UAV flight control system (such as waypoint sequence, speed constraint, heading angle, etc.) and sends it to the UAV in real time through the flight control communication interface, so that the UAV can perform autonomous flight according to the planned path. During the UAV's path execution, the execution feedback unit simultaneously obtains the acceleration and angular velocity data output by the inertial measurement unit (IMU) from the UAV's host system, obtains the motion vector information of the optical flow sensor, and receives the matching results of the LiDAR echo to describe the UAV's real motion trajectory relative to the environment in the underground space.

[0048] After completing the above data collection, the execution feedback unit performs fitting calculations on the actual flight trajectory of the UAV based on the trajectory fitting algorithm. By comparing the difference between the actual heading of the UAV and the heading of the planned path, the yaw is calculated. When the yaw is less than a set threshold, the system considers the UAV to be still in a path-controllable state and only sends its pose and environmental observation data to the map building unit for map updates. When the yaw exceeds the set threshold, the execution feedback unit immediately sends an instant replanning trigger signal to the path planning unit, so that the path planning unit can perform a re-path search based on the latest spatiotemporal environment and attitude information, ensuring that the UAV will not enter a dangerous area due to inertial deviation or dynamic changes in the environment.

[0049] In this embodiment, the strength of the communication signal is measured in real time by the wireless communication module. The measurement result is fitted by the fading model and mapped to the communication signal attenuation matrix Q, which is updated synchronously with the cost function to penalize path nodes in the communication blind zone.

[0050] In this embodiment, the system includes a SLAM uncertainty compensation unit. The uncertainty compensation unit dynamically adjusts the obstacle expansion radius of the map voxels according to the pose covariance matrix output by SLAM through a Gaussian dilation algorithm, so that the UAV automatically increases the safe distance between itself and the lane wall and obstacles in the uncertain positioning area.

[0051] In this embodiment, the system includes a task coordination unit, which runs an allocation algorithm based on multi-objective optimization. The task coordination unit reallocates tasks to multiple drones based on the drone's remaining battery power, path risk score, task priority matrix, and communication quality score, and uses a cooperative avoidance algorithm to generate a set of non-conflicting multi-drone paths.

[0052] In this embodiment, the execution feedback unit has a built-in return-to-home safety model. By calculating the current battery power E of the drone, the return-to-home path length Lr, and the energy consumption per unit distance k, it determines whether Ek·Lr is less than the safety threshold. When the condition is met, the path planning unit is automatically invoked to generate a return-to-home path or switch to the nearest safe landing point path.

[0053] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic path planning system for unmanned aerial vehicles in underground coal mines, characterized in that, The application relates to a method for real-time path planning of unmanned aerial vehicles (UAVs) in a mine environment, which comprises the following steps: a map construction unit is used to perform time synchronization on data collected by a laser radar, a depth camera, a gas sensor and an inertial measurement unit, a real-time updated three-dimensional voxel map is generated by point cloud splicing and inertial navigation fusion algorithm, and environment gas concentration, temperature and dust information are written into corresponding voxels; a risk prediction unit is used to receive the three-dimensional voxel map from the map construction unit, perform risk prediction on voxels in future Delta t time according to a gas diffusion equation, a temperature time series model and a microseismic event triggering model, and generate a space-time risk field R(x, y, z, t) containing a time dimension; a cost calculation unit is used to receive the space-time risk field from the risk prediction unit, input a three-dimensional map, a risk field, a dynamic obstacle position and a communication attenuation matrix into a preset cost function C = alpha * D + beta * R + gamma * O + delta * Q, and output a cost value matrix for planning; a path planning unit is used to obtain the cost value matrix from the cost calculation unit, perform an improved A* or improved RRT* search algorithm in a set time window interval, generate a current optimal flight sub-path of the unmanned aerial vehicle, and repeatedly perform planning when the time window is moved forward to realize real-time updating of the path; and an execution feedback unit is used to issue the optimal flight sub-path to a flight control system of the unmanned aerial vehicle, receive pose, IMU data and radar echo matching results uploaded by the unmanned aerial vehicle, and transmit the feedback information to the map construction unit to trigger map updating and planning recalculation in the next cycle. The laser radar in the map construction unit adopts a multi-line beam rotating structure, the depth camera adopts a structured light or TOF imaging mode, the map construction unit generates a dense voxel map through a point cloud splicing algorithm and an IMU inertial navigation fusion algorithm, and the uncertainty of the voxel map is marked by using a pose covariance output by SLAM. The risk prediction unit predicts the risk value of each voxel in future Delta t time by running a gas diffusion model based on a convection diffusion partial differential equation, a temperature prediction model based on time series analysis and a roof breaking model based on a microseismic event threshold triggering, and writes the prediction result into a space-time risk field. The cost function C in the cost calculation unit is composed of four matrix items alpha * D, beta * R, gamma * O and delta * Q; D is a three-dimensional geometric distance matrix between voxels, R is a risk prediction matrix, O is a dynamic obstacle occupancy matrix, and Q is a communication signal attenuation matrix, the cost calculation unit automatically adjusts the values of the weights alpha, beta, gamma and delta according to the residual power of the unmanned aerial vehicle, the task priority and the environment risk level by using an adaptive weight adjustment algorithm. The path planning unit adopts FPGA or GPU as a hardware acceleration module, performs improved A* or improved RRT* search by using block graph search and parallel node expansion technology, realizes fast path solving in a set time window, and removes the executed path segment after each rolling period to generate a continuous sub-path.

2. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: ​ 3. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: ​ 4. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: ​ ​ 5. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: ​ 6. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: The intensity of the communication signal is measured in real time by the wireless communication module, and the measurement result is mapped into a communication signal attenuation matrix Q after fitting by a fading model, and is updated synchronously with the cost function to punish the path nodes in the communication blind area.

7. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: The execution feedback unit includes a path offset detection module, which calculates the actual flight path of the unmanned aerial vehicle through the IMU, the optical flow sensor and the laser radar echo matching result, and determines the flight path offset based on a flight path fitting algorithm, and sends an immediate re-planning trigger signal to the path planning unit when the offset exceeds a set threshold.

8. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: The system includes a SLAM uncertainty compensation unit, which dynamically adjusts the obstacle inflation radius of the map voxel according to the pose covariance matrix output by the SLAM through a Gaussian inflation algorithm, so that the unmanned aerial vehicle automatically increases the safety distance from the lane wall and obstacles in the positioning uncertainty area.

9. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: The system includes a task coordination unit, which runs a multi-objective optimization-based allocation algorithm to perform task re-allocation for multiple unmanned aerial vehicles based on the remaining power of the unmanned aerial vehicles, path risk integral, task priority matrix and communication quality score, and generates a set of multiple unmanned aerial vehicle paths that do not conflict with each other using a cooperative avoidance algorithm.

10. The dynamic path planning system for unmanned aerial vehicle in underground coal mine of claim 1, wherein: The execution feedback unit is built-in with a return safety model, which determines whether E-k·Lr is less than a safety threshold by calculating the current power E of the unmanned aerial vehicle, the return path length Lr and the unit distance energy consumption k, and automatically calls the path planning unit to generate a return path or switch to the nearest safe landing point path when the condition is met.

Images