Machine dog transportation route analysis system and method based on multi-source data fusion

Abstract

This invention discloses a robot dog transportation route analysis system and method based on multi-source data fusion, belonging to the field of transportation analysis technology. The invention loads the material to be transported onto the robot dog's load platform and collects the material load characteristics; receives mining transportation task instructions, collects basic mining information, and generates an adapted transportation path using a path planning algorithm based on the load characteristics. The robot dog is controlled to move along the path. When any single foot enters the lifting cycle, the target landing position area of ​​that foot is determined, three-dimensional point cloud data is acquired, and a pit and crater feature dataset of the landing area is extracted. The pit and crater feature dataset is spatially registered with the spatial coordinates of the movable area of ​​the single foot's sole and a mapping relationship is established. Based on this, a set of pre-control parameters for the push rod is calculated, and pre-control is executed. After a single foot enters the climbing cycle, the target driving force parameters and the driving force distribution coefficients of each area are output using a climbing driving force prediction model to generate real-time dynamic adjustment control parameters for the electric push rod, completing real-time regulation.

Description

Technical Field

[0001] This invention relates to the field of route analysis technology, specifically to a robot dog transportation route analysis system and method based on multi-source data fusion. Background Technology

[0002] Mining transportation scenarios are mostly unstructured and complex. Whether it's the slope access roads and uneven gravel roads in open-pit mines, or the narrow tunnels and pitted floors in underground mines, the terrain is generally rugged, with varying slopes and localized road surface damage. Traditional wheeled and tracked transport equipment suffers from limited passage, large turning radii, and poor adaptability to narrow passages, making it difficult to perform precise transport operations for loose materials, emergency supplies, and small maintenance parts. Legged robotic dogs, with their strong terrain obstacle-crossing ability, flexible body, and highly controllable gait, are gradually becoming a new type of equipment for transporting materials in complex mine terrains and an important direction for the development of intelligent mining transportation.

[0003] Current technology cannot predict the terrain of the landing area in advance during the single-foot lift cycle. It only passively adjusts after the foot lands based on the machine's posture or pressure feedback. The foot's electric actuator lacks a pre-adjustment mechanism, making it difficult for the foot to quickly conform to the terrain when facing localized potholes in mines, resulting in severe single-point stress. Current technology cannot establish a correspondence between pothole locations and foot zones. The electric actuator can only make overall coarse adjustments and cannot perform fine-grained zone control for localized potholes. Its terrain adaptability is weak and it is difficult to adapt to the complex and ever-changing local surfaces in mines. Current climbing drive forces mostly use fixed preset parameters, which cannot accurately match the driving force required for climbing, nor can they achieve reasonable distribution of driving force among the movable areas of the foot. Summary of the Invention

[0004] The purpose of this invention is to provide a robot dog transportation route analysis system and method based on multi-source data fusion, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a method for analyzing the transportation routes of robot dogs based on multi-source data fusion, including the following steps:

[0007] The robot dog loads the materials to be transported onto its load platform and collects the load characteristics of the materials; it receives mine transportation task instructions, collects basic mine information, and uses a path planning algorithm to generate a transportation route based on the load characteristics.

[0008] The robot dog is controlled to move along the transport path. When any one of the robot dog's legs enters the lifting cycle, the target landing position area of ​​the leg is determined. The target landing position area is then scanned in three dimensions to obtain three-dimensional point cloud data and extract the pit and depression feature dataset of the landing area.

[0009] The data set of pit features in the landing area is registered with the coordinate row of the movable area of ​​the sole of a single foot to establish a mapping relationship. Based on the mapping relationship, the target extension and extension speed of the push rod corresponding to each movable area are calculated to generate a set of push rod pre-control parameters. During the landing transition cycle of a single foot, control is performed based on the set of push rod pre-control parameters.

[0010] During the single-foot climbing cycle, combined with the push rod pre-control parameter set, a pre-trained climbing driving force prediction model is used to output the target driving force parameters and driving force distribution coefficients. Based on the target driving force parameters and driving force distribution coefficients, real-time dynamic adjustment control parameters are generated for the electric push rods corresponding to each movable area of ​​the sole of the single foot, and real-time adjustments are made.

[0011] In conjunction with the first aspect, in a first embodiment of the first aspect of this application, the step of loading the material to be transported onto the robot dog's load platform and collecting the load characteristics of the material to be transported includes:

[0012] The material to be transported is placed in the preset effective bearing area of ​​the load platform, triggering the platform locking mechanism to perform a locking action, thus completing the rigid constraint between the material to be transported and the load platform.

[0013] The weighing sensor unit collects the total weight of the material to be transported and the spatial distribution data of the weight within the bearing surface; the attitude sensor unit collects the real-time attitude data of the load platform after loading and the spatial coordinate data of the center of gravity of the material; the contour sensor unit collects the three-dimensional outer contour dimension data of the material to be transported and the projection boundary data of the material within the bearing surface; the inertial measurement unit built into the robot dog body collects the static attitude data of the whole machine after loading the material and the offset data of the whole machine's center of gravity; the data is preprocessed and integrated to form the load characteristics of the material to be transported.

[0014] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the step of receiving the mine transportation task instruction, collecting basic mine information, and generating a transportation route using a path planning algorithm based on load characteristics includes:

[0015] Upon receiving a mine transportation task instruction, the system performs structured parsing on the instruction to extract the spatial coordinates of the transportation start point, the spatial coordinates of the transportation end point, the effective time limit of the task, the spatial range of the restricted area, the operation avoidance priority rules, and the material transportation safety constraint parameters, thereby generating a task constraint parameter set. Using the spatial line connecting the spatial coordinates of the transportation start point to the spatial coordinates of the transportation end point as a reference, the system extends outward by a preset buffer distance to delineate the effective collection boundary of basic mine information, determine the collection spatial range, and collect basic mine information.

[0016] Load characteristics are retrieved and matched with the task constraint parameter set and basic mine information to extract the total weight constraint, center of gravity offset constraint, machine height and width limit constraint, road surface bearing capacity threshold constraint, and slope passage limit constraint corresponding to the load. Task time limit constraint, no-entry and obstacle avoidance constraint, and mine transportation safety standard constraint are superimposed to construct a path planning constraint set. A preset path planning algorithm model is retrieved, with the spatial coordinates of the transportation start point as the algorithm start node and the spatial coordinates of the transportation end point as the algorithm end node. The path planning constraint set is input into the algorithm model, and iterative search of path nodes and passage cost calculation are performed. The candidate path with the lowest passage cost is selected as the transportation path. The path planning algorithm selected is the A* algorithm.

[0017] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the controlled robot dog travels along the transport path, and when any single foot of the robot dog enters a lifting cycle, the target landing position area of ​​that single foot is determined, including:

[0018] Using the time-series benchmark of the robot dog's gait cycle as a reference, for each single foot, the real-time dataset of the single foot's gait state is matched in real time to determine the stage of the gait cycle that the single foot is currently in. The stages of the gait cycle are divided into the support cycle, the lift-off cycle, the foot-landing transition cycle, and the climbing cycle.

[0019] Using the start time stamp of the single-foot lift-off cycle as a reference, the theoretical stride length, stride length, stride height, and landing time of the single foot in this gait cycle are calculated to determine the predicted fuselage pose. Based on the predicted fuselage pose, theoretical stride length, and stride length, the theoretical landing point reference spatial coordinates of the single foot in this landing cycle are calculated through the transformation relationship between the fuselage global coordinate system and the foot local coordinate system. Centered on the theoretical landing point reference spatial coordinates, the physical contour size parameters of the sole of the single foot, the total distribution range parameters of the movable area of ​​the sole, and the preset gait landing deviation threshold are retrieved, combined with the single foot movement... The kinematic reachability constraint defines the planar boundary of the target landing position area on the horizontal plane and generates a planar two-dimensional boundary coordinate set for the area. Based on the planar two-dimensional boundary coordinate set, the corresponding three-dimensional terrain benchmark data of the mine is retrieved to determine the upper and lower limits of the terrain elevation of the area. Combined with the preset allowable step height range of a single foot, the elevation boundary of the target landing position area is defined, and a three-dimensional spatial boundary parameter set for the area is generated. The target landing position area of ​​the single foot is determined, and the effective timing window and landing action execution time node of the single foot lifting cycle are bound.

[0020] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of performing a three-dimensional scan of the target landing location area to obtain three-dimensional point cloud data and extracting a dataset of pit and depression features of the landing area includes:

[0021] Within the effective scanning window of the single-leg gait lift-off cycle, a full-range scan of the target area is performed to acquire 3D point cloud data, which is then preprocessed. A topographic reference plane for the target area is generated by fitting the 3D point cloud data. Using the topographic reference plane as a reference, the elevation deviation of each point cloud is calculated. Point cloud sets with elevation deviations below a preset pothole depth threshold are selected. Invalid isolated points are removed through connectivity analysis, and the spatial boundaries of the effective pothole areas are delineated. For all effective pothole areas, the spatial coordinate range, pothole depth, elevation distribution, and edge slope are extracted to generate a pothole feature dataset for the footing area.

[0022] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the step of registering the dataset of pit features in the foot-landing area with the spatial distribution coordinates of the movable area of ​​the sole of a single foot to establish a mapping relationship includes:

[0023] Based on the pit and crater feature dataset of the foot landing area, the topological features of the pit and crater distribution are extracted to obtain the overall contour features of the foot landing area formed by the combination of all movable areas of the foot landing area and the topological features of the movable area distribution. Using the topological features of the pit and crater distribution and the movable area distribution as the matching benchmark, the feature point matching algorithm is used to initially align the overall contour features of the foot landing area with the topographic reference plane of the foot landing area, determine the initial registration transformation relationship, and make the overall coverage of the movable area of ​​the foot landing area completely coincide with the effective range of the target foot landing location area.

[0024] Based on the initial registration transformation relationship, an iterative spatial optimization matching algorithm is used for spatial registration. Based on the spatial reference completed by spatial registration, an equal-scale spatial grid division algorithm is used to define each independent movable area of ​​the foot as an independent matching grid unit. The topographic reference plane of the foot landing area is divided into the same number of corresponding grid units according to the same grid scale and topological characteristics as the movable area of ​​the foot. A spatial overlap matching algorithm is used to calculate the spatial overlap ratio between each grid unit of the movable area of ​​the foot and each grid unit of the foot landing area. The grid unit of the foot landing area with the highest overlap ratio is determined as the unique corresponding matching unit of the movable area of ​​the foot.

[0025] For each set of grid cells that has been matched, the pit feature data in the corresponding grid of the foot landing area is bound to the spatial distribution coordinates of the corresponding movable area of ​​the foot and the push rod stroke boundary parameters to establish a one-to-one mapping relationship between the spatial distribution location of pits, pit feature parameters and a single movable area of ​​the foot.

[0026] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of calculating the target extension and extension speed of the push rod corresponding to each movable area based on the mapping relationship, and generating a set of push rod pre-control parameters, includes:

[0027] Based on the full terrain elevation data of the target landing area and the overall load center of gravity distribution parameters, a spatial plane fitting algorithm is used to fit and generate the optimal support reference plane corresponding to the landing of a single foot. For each independent movable area of ​​the foot, the terrain elevation and pit depth of its corresponding mapped position are used as inputs. Combined with the optimal support reference plane, the initial target extension amount of the push rod corresponding to the movable area is generated through elevation difference matching calculation. An iterative multi-constraint optimization algorithm is used to superimpose push rod stroke boundary constraints, smooth transition constraints of extension amount of adjacent movable areas, load center of gravity distribution compensation constraints, and fuselage attitude correction constraints. Multiple rounds of iterative optimization and convergence verification are performed on the initial target extension amount to determine the target extension amount of each push rod.

[0028] For each electric actuator, the effective execution time window of the actuator action is determined by taking its current real-time extension / retraction position as the starting point, the position corresponding to the determined target extension / retraction amount as the ending point, and the critical time point before the landing action is executed. A smooth acceleration / deceleration trajectory planning algorithm is used, combined with the actuator's dynamic response characteristics and acceleration / deceleration limit thresholds, to plan and generate the velocity timing curve of the actuator's entire stroke. Through a multi-axis synchronous matching algorithm, the velocity timing curves of all single-foot electric actuators are optimized in a coordinated manner, adjusting the acceleration / deceleration timing nodes of each actuator to ensure that all actuators reach the target extension / retraction position synchronously, and that the speed and acceleration / deceleration of the entire stroke are within the preset thresholds, thus determining the target extension / retraction speed of each actuator.

[0029] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of using a pre-trained climbing drive force prediction model, in conjunction with a push rod pre-control parameter set, to output target drive force parameters and drive force distribution coefficients during the single-leg climbing cycle, includes:

[0030] The system collects the following datasets for this boarding cycle: the footing area pothole feature dataset, the push rod pre-control parameter set, the current plantar pressure distribution data, the load feature dataset, the robot dog's current walking speed parameters, and the fuselage attitude data. These datasets are then time-aligned and normalized to generate a standardized feature vector set. The system retrieves the boarding driving force prediction model, which has undergone offline training and online validation optimization. The model is trained using a bidirectional gated recurrent unit deep learning algorithm based on an attention mechanism. Forward inference and feature weighting calculations are performed to output the target driving force parameters and driving force allocation coefficients.

[0031] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the step of generating real-time dynamic adjustment control parameters for the electric actuator corresponding to each movable area of ​​the sole of a single foot based on the target driving force parameters and the driving force distribution coefficient, and performing real-time adjustment, includes:

[0032] Based on the target driving force parameters, and combined with the driving force distribution coefficients corresponding to each movable area of ​​the foot, the single target driving force that the push rod corresponding to each independent movable area needs to output is calculated, and the correspondence between the single target driving force and the push rod is established. For each push rod, based on its single target driving force, a mapping transformation algorithm is used to calculate the push rod's real-time target extension and contraction amount, extension and contraction speed, and acceleration and deceleration timing parameters.

[0033] Secondly, this application provides a robot dog transportation route analysis system based on multi-source data fusion, including:

[0034] The transportation route generation module includes: a load feature acquisition unit that loads the material to be transported onto the robot dog's load platform and collects the load features of the material to be transported; and a transportation route generation unit that receives mine transportation task instructions, collects basic mine information, and uses a route planning algorithm to generate a transportation route based on the load features.

[0035] The landing area pothole feature extraction module includes: a landing position area calculation unit that controls the robot dog to move along the transportation path, and when any single foot of the robot dog enters the lifting cycle, it determines the target landing position area of ​​the single foot; and a landing area pothole feature extraction unit that performs a 3D scan of the target landing position area, obtains 3D point cloud data, and extracts the landing area pothole feature dataset.

[0036] The push rod control module includes: a mapping relationship establishment unit that registers the data set of pit features in the landing area with the spatial distribution coordinate row of the movable area of ​​the sole of a single foot to establish a mapping relationship; a push rod pre-control parameter calculation unit that calculates the target extension and extension speed of the push rod corresponding to each movable area based on the mapping relationship, generating a push rod pre-control parameter set; and a push rod control unit that performs control based on the push rod pre-control parameter set during the landing transition cycle of a single foot.

[0037] The push rod real-time adjustment module includes: a driving force calculation unit that, during the single-foot climbing cycle, combines the push rod pre-control parameter set with a pre-trained climbing driving force prediction model to output target driving force parameters and driving force distribution coefficients; and a push rod real-time adjustment unit that, based on the target driving force parameters and driving force distribution coefficients, generates real-time dynamic adjustment control parameters for the electric push rod corresponding to each movable area of ​​the sole of the single foot, and performs real-time adjustments.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] 1. This invention locks the landing area in advance during the single-foot lifting cycle, extracts the pit features through three-dimensional scanning, and completes the pre-adjustment of the electric push rod in advance. Before landing, the shape of the foot is adapted to the local pit terrain, which greatly improves the landing fit and the stability of the machine body. It is perfectly adapted to the pit and rugged road surface of the mine and eliminates the hidden dangers of single-point force and machine body shaking.

[0040] 2. This invention establishes a mapping relationship between the distribution of pits and depressions and the movable area of ​​the foot through spatial registration, and calculates the push rod control parameters separately for each foot partition to achieve accurate adaptation of local pits and depressions, which greatly improves the adaptability of the robot dog's foot to complex mine terrain.

[0041] 3. This invention employs a pre-trained artificial intelligence model for predicting climbing driving force, integrates multi-dimensional real-time data, accurately outputs the target driving force and zonal allocation coefficients, and simultaneously generates real-time dynamic adjustment parameters to achieve intelligent matching of driving force and reasonable allocation of foot zonals, ensuring a continuous climbing gait, reducing equipment power consumption, and adapting to heavy-load climbing requirements. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the steps of the robot dog transportation route analysis method based on multi-source data fusion of the present invention;

[0043] Figure 2 This is a system structure diagram of the robot dog transportation route analysis system based on multi-source data fusion of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example: Figures 1-2 As shown, the present invention provides a technical solution:

[0046] like Figure 1 As shown, this application provides a method for analyzing the transportation routes of robot dogs based on multi-source data fusion, including the following steps:

[0047] Step S100: Load the material to be transported onto the robot dog's load platform and collect the load characteristics of the material to be transported; receive the mine transportation task instruction, collect basic mine information, and combine the load characteristics to generate a transportation route using a path planning algorithm;

[0048] Specifically, the material to be transported is placed in the preset effective bearing area of ​​the load platform, triggering the platform locking mechanism to perform a locking action, thus completing the rigid constraint between the material to be transported and the load platform;

[0049] The weighing sensor unit collects the total weight of the material to be transported and the spatial distribution data of the weight within the bearing surface; the attitude sensor unit collects the real-time attitude data of the load platform after loading and the spatial coordinate data of the center of gravity of the material; the contour sensor unit collects the three-dimensional outer contour dimension data of the material to be transported and the projection boundary data of the material within the bearing surface; the inertial measurement unit built into the robot dog body collects the static attitude data of the whole machine after loading the material and the offset data of the whole machine's center of gravity; the data is preprocessed and integrated to form the load characteristics of the material to be transported.

[0050] Furthermore, upon receiving mine transportation task instructions, the system performs structured parsing on these instructions to extract the spatial coordinates of the transportation start point, the spatial coordinates of the transportation end point, the effective time limit of the task, the spatial range of the restricted area, the operation avoidance priority rules, and the material transportation safety constraint parameters corresponding to the task, thereby generating a task constraint parameter set. Using the spatial line connecting the spatial coordinates of the transportation start point to the spatial coordinates of the transportation end point as a reference, the system extends outward by a preset buffer distance to delineate the effective collection boundary of basic mine information, determine the collection spatial range, and collect basic mine information.

[0051] Load characteristics are retrieved and matched with the task constraint parameter set and basic mine information to extract the total weight constraint, center of gravity offset constraint, machine height and width limit constraint, road surface bearing capacity threshold constraint, and slope passage limit constraint corresponding to the load. Task time limit constraint, no-entry and obstacle avoidance constraint, and mine transportation safety standard constraint are superimposed to construct a path planning constraint set. A preset path planning algorithm model is retrieved, with the spatial coordinates of the transportation start point as the algorithm start node and the spatial coordinates of the transportation end point as the algorithm end node. The path planning constraint set is input into the algorithm model, and iterative search of path nodes and passage cost calculation are performed. The candidate path with the lowest passage cost is selected as the transportation path. The path planning algorithm selected is the A* algorithm.

[0052] In one specific embodiment, for the scenario of transporting maintenance materials in a northern open-pit iron mine, a combination of mine bearing maintenance parts and 5L hydraulic oil is placed in the 0.8m×0.6m effective load-bearing area of ​​the robot dog load platform, triggering the electromagnetic locking mechanism to engage with a locking force of 800N, thus completing the rigid constraint between the materials and the platform.

[0053] Synchronous acquisition of load characteristics: The weighing sensor unit detected a total material weight of 42.8 kg, with the weights on the left and right sides of the bearing surface being 19.2 kg and 23.6 kg respectively; the attitude sensor unit measured a platform horizontal tilt angle of 0.2°, and the material's center of gravity coordinates as X+2.1 cm, Y-1.2 cm, and Z+15.3 cm; the contour sensor unit acquired the material's outer contour of 62 cm × 45 cm × 38 cm, with a projected boundary of 0.78 m × 0.58 m; the inertial measurement unit measured the overall machine's static pitch angle of 0.3°, roll angle of 0.1°, and center of gravity offset of 2.3 cm. After data preprocessing, the data was integrated into standard load characteristics.

[0054] Upon receiving the mine transportation task instruction, the core parameters are analyzed: starting point coordinates X:3286.42, Y:1572.18, Z:128.65; ending point coordinates X:3302.15, Y:1590.33, Z:116.42; task time limit 45 minutes; the restricted area is a steep slope section with a radius of 12m; the passage constraints are slope ≤15°, height limit 1.2m, and width limit 0.9m; the data collection boundary is defined by extending 8m outward from the line connecting the starting point and the ending point; the average slope of the area is 8°, the maximum slope is 12°, and the road surface load capacity is ≥50kg / ㎡.

[0055] Combining load characteristics, task constraints, and basic mine information, a path planning constraint set was constructed. The A* algorithm was used to iteratively search 128 path nodes, generating 3 candidate paths. The optimal path with a travel cost of 126.8 was selected. This path is 218.6m long and is expected to take 32 minutes to travel, meeting the heavy-load transportation constraints throughout.

[0056] Step S200: Control the robot dog to move along the transport path. When any one of the robot dog's legs enters the lifting cycle, determine the target landing position area of ​​the leg. Perform a three-dimensional scan on the target landing position area to obtain three-dimensional point cloud data and extract the pit and depression feature dataset of the landing area.

[0057] Specifically, using the time-series benchmark of the robot dog's gait cycle as a reference, for each single foot, the real-time dataset of the single foot's gait state is matched in real time to determine the stage of the gait cycle that the single foot is currently in. The stages of the gait cycle are divided into the support cycle, the lift cycle, the foot-landing transition cycle, and the climbing cycle.

[0058] Using the start time stamp of the single-foot lift-off cycle as a reference, the theoretical stride length, stride length, stride height, and landing time of the single foot in this gait cycle are calculated to determine the predicted fuselage pose. Based on the predicted fuselage pose, theoretical stride length, and stride length, the theoretical landing point reference spatial coordinates of the single foot in this landing cycle are calculated through the transformation relationship between the fuselage global coordinate system and the foot local coordinate system. Centered on the theoretical landing point reference spatial coordinates, the physical contour size parameters of the sole of the single foot, the total distribution range parameters of the movable area of ​​the sole, and the preset gait landing deviation threshold are retrieved, combined with the single foot movement... The kinematic reachability constraint defines the planar boundary of the target landing position area on the horizontal plane and generates a planar two-dimensional boundary coordinate set for the area. Based on the planar two-dimensional boundary coordinate set, the corresponding three-dimensional terrain benchmark data of the mine is retrieved to determine the upper and lower limits of the terrain elevation of the area. Combined with the preset allowable step height range of a single foot, the elevation boundary of the target landing position area is defined, and a three-dimensional spatial boundary parameter set for the area is generated. The target landing position area of ​​the single foot is determined, and the effective timing window and landing action execution time node of the single foot lifting cycle are bound.

[0059] Furthermore, within the effective scanning window of the single-leg gait lift-off cycle, a full-range scan of the target area is performed to acquire 3D point cloud data, which is then preprocessed. For the 3D point cloud data, a topographic reference plane for the target area is generated. Using the topographic reference plane as a reference, the elevation deviation of each point cloud is calculated. Point cloud sets with elevation deviations lower than a preset pothole depth threshold are selected, and invalid isolated points are removed through connectivity analysis to delineate the spatial boundaries of the effective pothole areas. For all effective pothole areas, the spatial coordinate range, pothole depth, elevation distribution, and edge slope are extracted to generate a pothole feature dataset for the footing area.

[0060] In one specific embodiment, the robot dog travels stably along the optimal transport path generated in S100, maintaining a constant overall speed of 0.6 m / s. The robot maintains a real-time pitch angle of 0.4° and a roll angle of 0.2°, strictly adhering to step S200 throughout the entire process to perform gait control and terrain perception operations. The specific implementation process and experimental data are as follows:

[0061] The robot dog traveled along the transport path to the 112.4m mark of the 218.6m total length. The main control system monitored the gait status of the four legs in real time. Using the preset gait cycle timing benchmark as a reference, it simultaneously collected data on the sole contact surface pressure, joint rotation angle, and foot spatial position of each leg, defining four major cycle stages: single-leg support, lifting, landing transition, and climbing. Among them, the sole contact surface pressure of the right forefoot dropped sharply from 126N in steady-state support to 18N, which was lower than the preset support disengagement threshold of 25N. The corresponding hip and knee joint rotation angles reached 38° and 52° respectively, matching the joint angle range at the start of the lifting cycle. After consistency verification for 5 consecutive sampling cycles (sampling frequency 100Hz), it was determined that the right forefoot officially entered the lifting cycle. The start timestamp of the lifting cycle was marked as 1682451236.72, the effective timing window duration was set to 0.8s, and the landing action execution node was locked at timestamp 1682451237.52.

[0062] Based on the start time stamp of the right forefoot lift cycle, combined with the current overall speed, load characteristics and predicted posture of the aircraft, the theoretical stride length of this gait cycle is calculated to be 32cm, stride length to be 28cm, stride height to be 15cm, and the predicted coordinates of the aircraft body at the foot landing node are X:3294.26, Y:1581.35, Z:122.78. Through rigid transformation between the global coordinate system of the fuselage and the local coordinate system of the foot, the theoretical landing point reference space coordinates of the right forefoot are calculated as X:3294.31, Y:1581.42, Z:122.65. Centered on these reference coordinates, and combined with the physical contour of the right forefoot sole of 18cm×12cm, the total range of the movable area of ​​the sole of 16cm×10cm, the preset landing deviation threshold of ±3cm, and the kinematic reach constraint, a target landing area plane boundary of 40cm×30cm is defined on the horizontal plane, corresponding to the two-dimensional coordinate interval X:3294.11-3294.51, Y:1581.22-1581.62.

[0063] Based on the aforementioned planar boundary, the corresponding three-dimensional terrain benchmark data of the mine is retrieved to determine the lower limit of the terrain elevation of the area as 122.3m and the upper limit as 123.1m. Combined with the allowable step height range of 12cm-18cm, the three-dimensional boundary of the target landing area is delineated, a complete three-dimensional spatial boundary parameter set is generated, and the right forefoot number, the timing window of this lifting cycle and the landing execution node are bound simultaneously to lock the final target landing area.

[0064] Within the effective scanning window of 0.8s (right foreleg lift cycle), the laser 3D radar scanning module on the robot dog is triggered. The scanning range is the locked 3D target area, with a scanning resolution of 1mm and a sampling point density of 200 points / cm². After completing the full-area scan, a total of 12,640 raw 3D point cloud data points are collected. Preprocessing of the raw point cloud data removes 317 noise points and outliers. Coordinate normalization and terrain coordinate system registration are then performed, resulting in 12,323 standardized effective point clouds.

[0065] Based on standardized point cloud data, a terrain reference plane for the target area was generated using the least squares method. A pothole depth threshold of -1.2 cm was set (potholes 1.2 cm below the reference plane were considered valid). The elevation deviation of each point cloud was calculated, and 426 point cloud sets meeting the threshold were selected. After connectivity analysis, 79 isolated points were removed, and two continuous valid pothole areas were identified. The first pothole has spatial coordinates of X: 3294.22-3294.38 and Y: 1581.31-1581.47, with a maximum pothole depth of 1.8 cm, an average elevation of 122.42 m, and an edge slope of 12°. The second pothole has spatial coordinates of X: 3294.35-3294.49 and Y: 1581.44-1581.58, with a maximum pothole depth of 1.5 cm, an average elevation of 122.47 m, and an edge slope of 9°.

[0066] Step S300: Register the data set of pit features in the landing area with the coordinate row of the spatial distribution of the movable area of ​​the sole of a single foot to establish a mapping relationship; based on the mapping relationship, calculate the target extension and extension speed of the push rod corresponding to each movable area to generate a set of push rod pre-control parameters; during the landing transition cycle of a single foot, control is performed based on the set of push rod pre-control parameters.

[0067] Specifically, based on the pit and crater feature dataset of the foot landing area, the topological features of the pit and crater distribution are extracted to obtain the overall contour features of the foot landing area formed by the combination of all movable areas of the foot landing area and the topological features of the movable area distribution. Using the topological features of the pit and crater distribution and the movable area distribution as the matching benchmark, the feature point matching algorithm is used to initially align the overall contour features of the foot landing area with the topographic reference plane of the foot landing area, determine the initial registration transformation relationship, and make the overall coverage of the movable area of ​​the foot landing area completely coincide with the effective range of the target foot landing location area.

[0068] Based on the initial registration transformation relationship, an iterative spatial optimization matching algorithm is used for spatial registration. Based on the spatial reference completed by spatial registration, an equal-scale spatial grid division algorithm is used to define each independent movable area of ​​the foot as an independent matching grid unit. The topographic reference plane of the foot landing area is divided into the same number of corresponding grid units according to the same grid scale and topological characteristics as the movable area of ​​the foot. A spatial overlap matching algorithm is used to calculate the spatial overlap ratio between each grid unit of the movable area of ​​the foot and each grid unit of the foot landing area. The grid unit of the foot landing area with the highest overlap ratio is determined as the unique corresponding matching unit of the movable area of ​​the foot.

[0069] For each set of grid cells that has been matched, the pit feature data in the corresponding grid of the foot landing area is bound to the spatial distribution coordinates of the corresponding movable area of ​​the foot and the push rod stroke boundary parameters to establish a one-to-one mapping relationship between the spatial distribution location of pits, pit feature parameters and a single movable area of ​​the foot.

[0070] Furthermore, based on the full terrain elevation data of the target landing area and the overall load center of gravity distribution parameters, a spatial plane fitting algorithm is used to fit and generate the optimal support reference plane corresponding to the landing of a single foot. For each independent movable area of ​​the foot, the terrain elevation and pit depth of its corresponding mapped position are used as inputs, and combined with the optimal support reference plane, the initial target extension amount of the push rod corresponding to the movable area is generated through elevation difference matching calculation. An iterative multi-constraint optimization algorithm is used, superimposed with push rod stroke boundary constraints, smooth transition constraints of extension amount of adjacent movable areas, load center of gravity distribution compensation constraints, and fuselage attitude correction constraints, to perform multiple rounds of iterative optimization and convergence verification on the initial target extension amount, and determine the target extension amount of each push rod.

[0071] For each electric actuator, the effective execution time window of the actuator action is determined by taking its current real-time extension / retraction position as the starting point, the position corresponding to the determined target extension / retraction amount as the ending point, and the critical time point before the landing action is executed. A smooth acceleration / deceleration trajectory planning algorithm is used, combined with the actuator's dynamic response characteristics and acceleration / deceleration limit thresholds, to plan and generate the velocity timing curve of the actuator's entire stroke. Through a multi-axis synchronous matching algorithm, the velocity timing curves of all single-foot electric actuators are optimized in a coordinated manner, adjusting the acceleration / deceleration timing nodes of each actuator to ensure that all actuators reach the target extension / retraction position synchronously, and that the speed and acceleration / deceleration of the entire stroke are within the preset thresholds, thus determining the target extension / retraction speed of each actuator.

[0072] In one specific embodiment, the dataset of pit features in the right forefoot landing area and the topographic reference plane parameters of the target landing area generated by S200 are retrieved simultaneously, along with the core parameters of the movable area of ​​the right forefoot sole: the sole of this single foot is divided into 5 independent movable areas, numbered from front to back and left to right as movable areas 1 to 5. The size of each individual movable area is 3.2cm × 2cm, and the overall movable area distribution size is 16cm × 10cm. The corresponding local spatial distribution coordinates have been pre-calibrated. Using the theoretical landing point reference coordinates determined by S200 as the unified registration origin, the mine global coordinate system corresponding to the pit features and the foot local coordinate system corresponding to the movable area of ​​the sole are transformed to the same landing reference plane through a rigid body coordinate transformation algorithm, thus completely eliminating coordinate system deviation.

[0073] An initial coarse registration operation was performed, extracting the topological distribution features and edge inflection point features of two effective pits in the pit feature dataset. At the same time, the overall contour topological features and partitioned arrangement features of the combination of five movable areas on the sole of the foot were extracted. Using the two types of topological features as the matching benchmark, the feature point matching algorithm was called to perform initial alignment, quickly completing the overlap matching between the overall coverage of the movable area on the sole of the foot and the target foot landing area. After coarse registration, the overall spatial deviation was controlled within 0.6mm, which met the initial conditions for subsequent fine registration, and the initial registration transformation relationship was determined.

[0074] Based on the coarse registration results, an iterative spatial optimization matching algorithm was initiated for fine registration. Using feature points at the edges of potholes and corner points of the movable area on the sole as matching objects, an iteration termination deviation threshold of 0.1 mm was set. After six rounds of iteration optimization, the overall registration deviation was reduced to 0.08 mm, and the deviations of all local matching points were below the preset threshold, achieving high-precision spatial registration. Subsequently, an equal-scale spatial grid division algorithm was used to divide the five independent movable areas of the sole into five independent matching grid units. Simultaneously, the topographic reference plane of the foot landing area was divided into five corresponding grid units with identical grid scale and topology, ensuring complete correspondence in the number, size, and position of the grids.

[0075] The spatial overlap matching algorithm was invoked to calculate the spatial overlap ratio between each movable area grid on the sole and the terrain grid. The calculation showed that movable area 1 had an overlap ratio of 98.7% with the first pit grid, movable area 2 had an overlap ratio of 97.3% with the edge grid of the first pit, movable area 3 had an overlap ratio of 100% with the grid of the flat area, movable area 4 had an overlap ratio of 98.1% with the grid of the second pit, and movable area 5 had an overlap ratio of 96.9% with the edge grid of the second pit. The grid with the highest overlap ratio was determined as the unique matching unit, and a one-to-one mapping relationship was formally established between the spatial location and features of the pit and the five movable areas on the sole, binding all parameters without overlap or omission.

[0076] Entering the push rod pre-control parameter calculation stage, based on the full terrain elevation of the target area of ​​S200 and the center of gravity distribution parameters of the 42.8kg material load, a spatial plane fitting algorithm is used to generate the optimal support reference plane for the right forefoot, with the reference plane elevation locked at 122.62m. For each movable area, combined with the elevation and depth data of the corresponding pits, the initial target expansion and contraction amount is calculated through elevation difference matching: movable area 1 corresponds to a maximum pit depth of 1.8cm, with an initial expansion and contraction amount of -18mm; movable area 2 corresponds to a pit edge depth of 1.2cm, with an initial expansion and contraction amount of -12mm; movable area 3 corresponds to flat terrain, with an initial expansion and contraction amount of 0mm; movable area 4 corresponds to a pit depth of 1.5cm, with an initial expansion and contraction amount of -15mm; movable area 5 corresponds to a pit edge depth of 0.9cm, with an initial expansion and contraction amount of -9mm.

[0077] An iterative multi-constraint optimization algorithm was invoked, superimposed with boundary constraints of ±20mm push rod stroke, ≤3mm smoothness difference of extension / retraction between adjacent regions, center of gravity offset compensation constraint, and fuselage attitude correction constraint. The initial extension / retraction was optimized and converged in 4 rounds, and the target extension / retraction of the push rod corresponding to each movable region was finally determined: movable region 1 is -17.6mm, movable region 2 is -11.8mm, movable region 3 is 0mm, movable region 4 is -14.7mm, and movable region 5 is -8.9mm. All parameters are within the allowable range of push rod stroke.

[0078] Subsequently, the target extension and retraction speed of the push rods was calculated. Taking the critical timestamp of the landing action locked by S200 (1682451237.52) as the cutoff node, the effective execution time of the landing transition cycle was defined as 0.3s. Combining the target extension and retraction of each push rod, a smooth acceleration and deceleration trajectory planning algorithm was called, and the push rod acceleration and deceleration limit threshold was set to 5mm / s². The full stroke speed timing curve was planned. Then, the timing of each push rod was optimized through a multi-axis synchronous matching algorithm to ensure that the five sets of push rods reached the target position synchronously. Finally, the target extension and retraction speeds were determined as follows: movable area 1: 58.7mm / s, movable area 2: 39.3mm / s, movable area 3: 0mm / s, movable area 4: 49mm / s, and movable area 5: 29.7mm / s. The speed and acceleration / deceleration throughout the stroke all met the hardware threshold requirements.

[0079] Step S400: During the single-foot climbing cycle, using the pre-trained climbing driving force prediction model in conjunction with the push rod pre-control parameter set, output the target driving force parameters and driving force distribution coefficients; based on the target driving force parameters and driving force distribution coefficients, generate real-time dynamic adjustment control parameters for the electric push rod corresponding to each movable area of ​​the sole of the single foot, and perform real-time adjustment.

[0080] Specifically, the system collects the following datasets for the current boarding cycle: the footing area pothole feature dataset, the push rod pre-control parameter set, the current plantar pressure distribution data, the load feature dataset, the robot dog's current travel speed parameters, and the fuselage attitude data. These datasets are then time-aligned and normalized to generate a standardized feature vector set. The system retrieves the boarding driving force prediction model, which has undergone offline training and online validation optimization. The model is trained using a bidirectional gated recurrent unit deep learning algorithm based on an attention mechanism. Forward inference and feature weighting calculations are performed to output the target driving force parameters and driving force allocation coefficients.

[0081] Furthermore, based on the target driving force parameters and combined with the driving force distribution coefficients corresponding to each movable area of ​​the foot, the single target driving force that the push rod corresponding to each independent movable area needs to output is calculated, and the correspondence between the single target driving force and the push rod is established. For each push rod, based on its single target driving force, a mapping transformation algorithm is used to calculate the push rod's real-time target extension and contraction amount, extension and contraction speed, and acceleration and deceleration timing parameters.

[0082] In one specific embodiment, multi-source data of the climbing cycle is collected: feature data of two potholes in the landing area, the set of push rod pre-control parameters generated by S300, real-time plantar pressure distribution data (pressure of 22N in movable area 1, 18N in area 2, 35N in area 3, 20N in area 4, and 16N in area 5), ​​feature data of 42.8kg load, overall machine travel speed of 0.6m / s, and fuselage pitch angle of 0.4° and roll angle of 0.2°; all data are synchronized and normalized to generate a standardized feature vector set.

[0083] The pre-trained climbing driving force prediction model is retrieved, and standardized feature vectors are imported to complete forward inference. After feature weighting calculation, the total target driving force of the right forefoot is output as 132N. At the same time, the driving force distribution coefficients of the five movable areas of the foot are output: 0.22 for area 1, 0.18 for area 2, 0.30 for area 3, 0.20 for area 4, and 0.10 for area 5. The sum of the coefficients is 1, which meets the constraint requirements.

[0084] Based on the total target driving force, the individual target driving force of each region is calculated using the allocation coefficient: Region 1 is 29.04N, Region 2 is 23.76N, Region 3 is 39.6N, Region 4 is 26.4N, and Region 5 is 13.2N. Through a mapping transformation algorithm, combined with real-time terrain feedback and push rod status, real-time dynamic adjustment parameters for each electric push rod are calculated: real-time target extension / retraction amounts are -17.8mm, -12.0mm, 0mm, -14.9mm, and -9.0mm, respectively, and real-time extension / retraction speeds are 56mm / s, 38mm / s, 0mm / s, 47mm / s, and 28mm / s, respectively, with acceleration / deceleration timings synchronized to the ascent rhythm.

[0085] The main control system issues real-time adjustment commands, and all push rods synchronously complete dynamic fine-tuning, conforming to the rhythm of the climbing cycle throughout the process. The foot support force is evenly distributed, and the body attitude fluctuation is less than 0.1°, ensuring stability and no shaking during the heavy-load climbing process, and successfully completing this single-leg climbing operation.

[0086] like Figure 2 As shown, this application provides a robot dog transportation route analysis system based on multi-source data fusion, including:

[0087] The transportation route generation module includes: a load feature acquisition unit that loads the material to be transported onto the robot dog's load platform and collects the load features of the material to be transported; and a transportation route generation unit that receives mine transportation task instructions, collects basic mine information, and uses a route planning algorithm to generate a transportation route based on the load features.

[0088] The landing area pothole feature extraction module includes: a landing position area calculation unit that controls the robot dog to move along the transportation path, and when any single foot of the robot dog enters the lifting cycle, it determines the target landing position area of ​​the single foot; and a landing area pothole feature extraction unit that performs a 3D scan of the target landing position area, obtains 3D point cloud data, and extracts the landing area pothole feature dataset.

[0089] The push rod control module includes: a mapping relationship establishment unit that registers the data set of pit features in the landing area with the spatial distribution coordinate row of the movable area of ​​the sole of a single foot to establish a mapping relationship; a push rod pre-control parameter calculation unit that calculates the target extension and extension speed of the push rod corresponding to each movable area based on the mapping relationship, generating a push rod pre-control parameter set; and a push rod control unit that performs control based on the push rod pre-control parameter set during the landing transition cycle of a single foot.

[0090] The push rod real-time adjustment module includes: a driving force calculation unit that, during the single-foot climbing cycle, combines the push rod pre-control parameter set with a pre-trained climbing driving force prediction model to output target driving force parameters and driving force distribution coefficients; and a push rod real-time adjustment unit that, based on the target driving force parameters and driving force distribution coefficients, generates real-time dynamic adjustment control parameters for the electric push rod corresponding to each movable area of ​​the sole of the single foot, and performs real-time adjustments.

[0091] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for analyzing the transportation routes of robot dogs based on multi-source data fusion, characterized in that, Includes the following steps: The robot dog loads the materials to be transported onto its load platform and collects the load characteristics of the materials; it receives mine transportation task instructions, collects basic mine information, and uses a path planning algorithm to generate a transportation route based on the load characteristics. Control the robot dog to move along the transport path. When any one of the robot dog's legs enters the lifting cycle, determine the target landing position area of ​​that leg. Perform a 3D scan of the target landing area to obtain 3D point cloud data and extract the pit and depression feature dataset of the landing area; The dataset of pit and crater features of the foot landing area is registered with the spatial distribution coordinate row of the movable area of ​​the sole of a single foot to establish a mapping relationship; Based on the mapping relationship, the target extension and extension speed of the push rod corresponding to each movable area are calculated to generate a set of push rod pre-control parameters; during the foot landing transition cycle of a single foot, control is performed based on the set of push rod pre-control parameters. During the single-leg climbing cycle, the target driving force parameters and driving force distribution coefficients are output by combining the push rod pre-control parameter set and using the pre-trained climbing driving force prediction model. Based on the target driving force parameters and driving force distribution coefficient, real-time dynamic adjustment control parameters are generated for the electric actuator corresponding to each movable area of ​​the sole of a single foot, and real-time adjustments are made.

2. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The process of loading the material to be transported onto the robot dog's load platform and collecting the load characteristics of the material to be transported includes: The material to be transported is placed in the preset effective bearing area of ​​the load platform, triggering the platform locking mechanism to perform a locking action, thus completing the rigid constraint between the material to be transported and the load platform. The weighing sensor unit collects the total weight of the material to be transported and the spatial distribution data of the weight within the bearing surface; the attitude sensor unit collects the real-time attitude data of the load platform after loading and the spatial coordinate data of the center of gravity of the material; the contour sensor unit collects the three-dimensional outer contour dimension data of the material to be transported and the projection boundary data of the material within the bearing surface; the inertial measurement unit built into the robot dog body collects the static attitude data of the whole machine after loading the material and the offset data of the whole machine's center of gravity; the data is preprocessed and integrated to form the load characteristics of the material to be transported.

3. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The process of receiving mine transportation task instructions, collecting basic mine information, combining load characteristics, and generating transportation routes using path planning algorithms includes: Upon receiving a mine transportation task instruction, the system performs structured parsing on the instruction to extract the spatial coordinates of the transportation start point, the spatial coordinates of the transportation end point, the effective time limit of the task, the spatial range of the restricted area, the operation avoidance priority rules, and the material transportation safety constraint parameters, thereby generating a task constraint parameter set. Using the spatial line connecting the spatial coordinates of the transportation start point to the spatial coordinates of the transportation end point as a reference, the system extends outward by a preset buffer distance to delineate the effective collection boundary of basic mine information, determine the collection spatial range, and collect basic mine information. Load characteristics are retrieved and matched with the task constraint parameter set and basic mine information to extract the total weight constraint, center of gravity offset constraint, machine height and width limit constraint, road surface bearing capacity threshold constraint, and slope passage limit constraint corresponding to the load. Task time limit constraint, no-entry and obstacle avoidance constraint, and mine transportation safety standard constraint are superimposed to construct a path planning constraint set. A preset path planning algorithm model is retrieved, with the spatial coordinates of the transportation start point as the algorithm start node and the spatial coordinates of the transportation end point as the algorithm end node. The path planning constraint set is input into the algorithm model, and iterative search of path nodes and passage cost calculation are performed. The candidate path with the lowest passage cost is selected as the transportation path. The path planning algorithm selected is the A* algorithm.

4. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The controlled robot dog travels along the transport path. When any one of the robot dog's legs enters a lifting cycle, the target landing position area of ​​that leg is determined, including: Using the time-series benchmark of the robot dog's gait cycle as a reference, for each single foot, the real-time dataset of the single foot's gait state is matched in real time to determine the stage of the gait cycle that the single foot is currently in. The stages of the gait cycle are divided into the support cycle, the lift-off cycle, the foot-landing transition cycle, and the climbing cycle. Using the start time stamp of the single-foot lift-off cycle as a reference, the theoretical stride length, stride length, stride height, and landing time of the single foot in this gait cycle are calculated to determine the predicted fuselage pose. Based on the predicted fuselage pose, theoretical stride length, and stride length, the theoretical landing point reference spatial coordinates of the single foot in this landing cycle are calculated through the transformation relationship between the fuselage global coordinate system and the foot local coordinate system. Centered on the theoretical landing point reference spatial coordinates, the physical contour size parameters of the sole of the single foot, the total distribution range parameters of the movable area of ​​the sole, and the preset gait landing deviation threshold are retrieved, combined with the single foot movement... The kinematic reachability constraint defines the planar boundary of the target landing position area on the horizontal plane and generates a planar two-dimensional boundary coordinate set for the area. Based on the planar two-dimensional boundary coordinate set, the corresponding three-dimensional terrain benchmark data of the mine is retrieved to determine the upper and lower limits of the terrain elevation of the area. Combined with the preset allowable step height range of a single foot, the elevation boundary of the target landing position area is defined, and a three-dimensional spatial boundary parameter set for the area is generated. The target landing position area of ​​the single foot is determined, and the effective timing window and landing action execution time node of the single foot lifting cycle are bound.

5. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The step of performing a 3D scan of the target landing location area to obtain 3D point cloud data and extracting the pit and depression feature dataset of the landing area includes: Within the effective scanning window of the single-leg gait lift-off cycle, a full-range scan of the target area is performed to acquire 3D point cloud data, which is then preprocessed. A topographic reference plane for the target area is generated by fitting the 3D point cloud data. Using the topographic reference plane as a reference, the elevation deviation of each point cloud is calculated. Point cloud sets with elevation deviations below a preset pothole depth threshold are selected. Invalid isolated points are removed through connectivity analysis, and the spatial boundaries of the effective pothole areas are delineated. For all effective pothole areas, the spatial coordinate range, pothole depth, elevation distribution, and edge slope are extracted to generate a pothole feature dataset for the footing area.

6. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The step of registering the data set of pit and depression features of the foot-landing area with the spatial distribution coordinates of the movable area of ​​the sole of a single foot to establish a mapping relationship includes: Based on the pit and crater feature dataset of the foot landing area, the topological features of the pit and crater distribution are extracted to obtain the overall contour features of the foot landing area formed by the combination of all movable areas of the foot landing area and the topological features of the movable area distribution. Using the topological features of the pit and crater distribution and the movable area distribution as the matching benchmark, the feature point matching algorithm is used to initially align the overall contour features of the foot landing area with the topographic reference plane of the foot landing area, determine the initial registration transformation relationship, and make the overall coverage of the movable area of ​​the foot landing area completely coincide with the effective range of the target foot landing location area. Based on the initial registration transformation relationship, an iterative spatial optimization matching algorithm is used for spatial registration. Based on the spatial reference completed by spatial registration, an equal-scale spatial grid division algorithm is used to define each independent movable area of ​​the foot as an independent matching grid unit. The topographic reference plane of the foot landing area is divided into the same number of corresponding grid units according to the same grid scale and topological characteristics as the movable area of ​​the foot. A spatial overlap matching algorithm is used to calculate the spatial overlap ratio between each grid unit of the movable area of ​​the foot and each grid unit of the foot landing area. The grid unit of the foot landing area with the highest overlap ratio is determined as the unique corresponding matching unit of the movable area of ​​the foot. For each set of grid cells that has been matched, the pit feature data in the corresponding grid of the foot landing area is bound to the spatial distribution coordinates of the corresponding movable area of ​​the foot and the push rod stroke boundary parameters to establish a one-to-one mapping relationship between the spatial distribution location of pits, pit feature parameters and a single movable area of ​​the foot.

7. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, Based on the mapping relationship, the target extension and extension speed of the push rod corresponding to each movable area are calculated to generate a set of push rod pre-control parameters, including: Based on the full terrain elevation data of the target landing area and the overall load center of gravity distribution parameters, a spatial plane fitting algorithm is used to fit and generate the optimal support reference plane corresponding to the landing of a single foot. For each independent movable area of ​​the foot, the terrain elevation and pit depth of its corresponding mapped position are used as inputs. Combined with the optimal support reference plane, the initial target extension amount of the push rod corresponding to the movable area is generated through elevation difference matching calculation. An iterative multi-constraint optimization algorithm is used to superimpose push rod stroke boundary constraints, smooth transition constraints of extension amount of adjacent movable areas, load center of gravity distribution compensation constraints, and fuselage attitude correction constraints. Multiple rounds of iterative optimization and convergence verification are performed on the initial target extension amount to determine the target extension amount of each push rod. For each electric actuator, the effective execution time window of the actuator action is determined by taking its current real-time extension / retraction position as the starting point, the position corresponding to the determined target extension / retraction amount as the ending point, and the critical time point before the landing action is executed. A smooth acceleration / deceleration trajectory planning algorithm is used, combined with the actuator's dynamic response characteristics and acceleration / deceleration limit thresholds, to plan and generate the velocity timing curve of the actuator's entire stroke. Through a multi-axis synchronous matching algorithm, the velocity timing curves of all single-foot electric actuators are optimized in a coordinated manner, adjusting the acceleration / deceleration timing nodes of each actuator to ensure that all actuators reach the target extension / retraction position synchronously, and that the speed and acceleration / deceleration of the entire stroke are within the preset thresholds, thus determining the target extension / retraction speed of each actuator.

8. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, During the single-leg climbing cycle, combined with the push rod pre-control parameter set, a pre-trained climbing driving force prediction model is used to output target driving force parameters and driving force distribution coefficients, including: The system collects the following datasets for this boarding cycle: the footing area pothole feature dataset, the push rod pre-control parameter set, the current plantar pressure distribution data, the load feature dataset, the robot dog's current walking speed parameters, and the fuselage attitude data. These datasets are then time-aligned and normalized to generate a standardized feature vector set. The system retrieves the boarding driving force prediction model, which has undergone offline training and online validation optimization. The model is trained using a bidirectional gated recurrent unit deep learning algorithm based on an attention mechanism. Forward inference and feature weighting calculations are performed to output the target driving force parameters and driving force allocation coefficients.

9. The method for analyzing the transportation route of a robot dog based on multi-source data fusion according to claim 1, characterized in that, The method involves generating real-time dynamic adjustment control parameters for the electric actuator corresponding to each movable area of ​​the sole of a single foot, based on the target driving force parameters and driving force distribution coefficient, and performing real-time adjustments, including: Based on the target driving force parameters, and combined with the driving force distribution coefficients corresponding to each movable area of ​​the foot, the single target driving force that the push rod corresponding to each independent movable area needs to output is calculated, and the correspondence between the single target driving force and the push rod is established. For each push rod, based on its single target driving force, a mapping transformation algorithm is used to calculate the push rod's real-time target extension and contraction amount, extension and contraction speed, and acceleration and deceleration timing parameters.

10. A robot dog transportation route analysis system based on multi-source data fusion, using the robot dog transportation route analysis method based on multi-source data fusion as described in any one of claims 1-9, characterized in that, include: The transportation route generation module includes: a load feature acquisition unit that loads the material to be transported onto the robot dog's load platform and collects the load features of the material to be transported; and a transportation route generation unit that receives mine transportation task instructions, collects basic mine information, and uses a route planning algorithm to generate a transportation route based on the load features. The landing area pothole feature extraction module includes: a landing position area calculation unit that controls the robot dog to move along the transportation path, and when any single foot of the robot dog enters the lifting cycle, it determines the target landing position area of ​​the single foot; and a landing area pothole feature extraction unit that performs a 3D scan of the target landing position area, obtains 3D point cloud data, and extracts the landing area pothole feature dataset. The push rod control module includes: a mapping relationship establishment unit that registers the data set of pit features in the landing area with the spatial distribution coordinate row of the movable area of ​​the sole of a single foot to establish a mapping relationship; a push rod pre-control parameter calculation unit that calculates the target extension and extension speed of the push rod corresponding to each movable area based on the mapping relationship, generating a push rod pre-control parameter set; and a push rod control unit that performs control based on the push rod pre-control parameter set during the landing transition cycle of a single foot. The push rod real-time adjustment module includes: a driving force calculation unit that, during the single-foot climbing cycle, combines the push rod pre-control parameter set with a pre-trained climbing driving force prediction model to output target driving force parameters and driving force distribution coefficients; and a push rod real-time adjustment unit that, based on the target driving force parameters and driving force distribution coefficients, generates real-time dynamic adjustment control parameters for the electric push rod corresponding to each movable area of ​​the sole of the single foot, and performs real-time adjustments.

Images