Fire-fighting robot adaptive obstacle crossing control method based on terrain feature learning

Abstract

This invention discloses an adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning, belonging to the field of firefighting robot technology. The method includes: collecting ground geometric information and ground bearing pressure information to form a spatially correlated terrain dataset; performing terrain feature learning based on the spatially correlated terrain dataset to construct a joint terrain feature matrix; dynamically calculating the bearing margin and collapse probability of each area of ​​the ground based on the joint terrain feature matrix to generate a terrain bearing risk map with risk levels marked; calculating an adaptive obstacle-crossing control parameter set based on the terrain bearing risk map and the joint terrain feature matrix; planning and optimizing the obstacle-crossing path based on the adaptive obstacle-crossing control parameter set, and controlling the firefighting robot to move along the optimized obstacle-crossing path according to the adaptive obstacle-crossing control parameter set. This invention improves the obstacle-crossing safety and mission continuity of firefighting robots in complex firefighting environments.

Description

Technical Field

[0001] This invention relates to the field of firefighting robot technology, and more specifically, to an adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning. Background Technology

[0002] In fire rescue and complex environment operations, such as warehouse mezzanines, equipment platforms, underground garages, and ruined buildings, accurate assessment of terrain accessibility and bearing capacity is crucial for enabling firefighting robots to move and overcome obstacles in complex terrain. Based on this assessment, ground bearing risk can be evaluated and obstacle-crossing control parameters adjusted. In these scenarios, the ground geometry is complex and variable, and firefighting robots need to cross thresholds, climb slopes, traverse steps, and navigate through partially collapsed areas, resulting in a dynamic state of contact between the robot's wheels and tracks and the ground.

[0003] To assess terrain accessibility, existing technologies typically employ terrain geometry feature acquisition and analysis. The core principle of these technologies is as follows: before the firefighting robot is deployed, sensors collect terrain geometry features, including obstacle height, slope, surface roughness, step shape, and wheel-track contact angle. A mapping relationship is established between these geometric features and terrain accessibility, enabling the classification and determination of terrain accessibility. During subsequent operations, the controller directly calls preset terrain accessibility determination rules to process the real-time collected terrain geometry information, obtaining the terrain accessibility determination result. Based on this result, the controller selects drive torque, travel speed, center of gravity adjustment, and attitude control strategies to achieve obstacle crossing and path planning.

[0004] However, existing technologies only determine terrain accessibility rules based on terrain geometry, without considering the impact of terrain bearing capacity on accessibility. In real fire rescue scenarios, due to the effects of fire, high temperatures, localized burning, water immersion, and secondary collapse, the ground surface remains flat and meets geometric passability conditions, while the internal bearing capacity of the ground decreases. After a fire, steel gratings, carbonized wood panels, burned laminated floors, and partially collapsed suspended ceilings appear as flat ground in visual and point cloud information. Existing terrain accessibility determination methods based on terrain geometry classify the terrain corresponding to these areas as low-risk accessible terrain. However, the wheels and tracks of firefighting robots apply loads to the ground during movement, causing these areas to collapse. The disconnect between terrain geometry and carrying capacity directly leads to a mismatch between terrain accessibility assessment results and actual operational risks. Pre-defined terrain accessibility assessment rules based solely on terrain geometry cannot adapt to changes in terrain carrying capacity under fire conditions, resulting in decreased accuracy in accessibility assessments and consequently increasing the risk of the firefighting robot getting stuck or overturning. When terrain accessibility is overestimated, the controller may misjudge that the corresponding terrain meets safe passage conditions, failing to adjust obstacle crossing control parameters or plan detour paths in time, ultimately causing the firefighting robot to partially get stuck, overturn, or lose control during movement. Conversely, when terrain accessibility is underestimated, the controller may over-adjust obstacle crossing control parameters and select detour paths, leading to decreased operational efficiency and delays in rescue operations. Both of these situations interrupt the firefighting robot's mission execution, block evacuation and rescue routes, increase the risks during rescue operations, and fail to meet the stringent requirements for obstacle crossing safety and mission continuity for firefighting robots in complex firefighting environments.

[0005] In view of this, the present invention proposes an adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning to solve the above problems. Summary of the Invention

[0006] To overcome the aforementioned shortcomings of the prior art and achieve the above objectives, the present invention provides the following technical solution: an adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning, comprising: Collect ground geometry information and ground bearing pressure information, perform time alignment and spatial correlation on the ground geometry information and ground bearing pressure information to form a spatially correlated terrain dataset; Terrain features are learned based on spatially correlated terrain datasets to construct a joint terrain feature matrix; Based on the joint terrain feature matrix, the bearing margin and collapse probability are dynamically calculated to generate a terrain bearing risk map with marked risk levels. Calculate the adaptive obstacle crossing control parameter set based on the terrain bearing risk map and the joint terrain feature matrix; The obstacle-crossing path is planned and optimized based on the adaptive obstacle-crossing control parameter set, and the fire-fighting robot is controlled to move along the optimized obstacle-crossing path according to the adaptive obstacle-crossing control parameter set.

[0007] Furthermore, methods for forming spatially correlated terrain datasets include: Record the ground position and corresponding collection time of the fire-fighting robot during the data collection process to form the movement trajectory of the fire-fighting robot; By combining the movement trajectory of the firefighting robot, the ground geometry information and ground bearing pressure information are correlated according to the collection time; The ground geometric information and ground bearing pressure information, which are corresponding to the collection time, are mapped to ground locations to form a spatially correlated terrain dataset.

[0008] Furthermore, methods for constructing a joint terrain feature matrix include: Ground units are divided based on their location in the spatially correlated terrain dataset; Based on ground geometry information and ground bearing pressure information, the estimated bearing capacity of each ground unit is calculated. According to the corresponding ground location, the ground geometry information, ground bearing pressure information, and bearing capacity estimate are written into the corresponding ground unit; Based on the movement trajectory of the firefighting robot, determine the arrangement order of each ground unit; Arrange the surface units according to the given order to construct a joint terrain feature matrix.

[0009] Furthermore, methods for calculating the estimated bearing capacity of each surface unit include: Extracting local pressure distribution from ground bearing pressure information; Material matching is performed based on ground geometry information to obtain the ground material type; Based on the ground material type, the preset material bearing capacity benchmark range table is extracted to obtain the bearing capacity benchmark range; Structural characterization information is extracted based on ground geometric information and ground bearing pressure information, and the structural characterization information is matched with preset structural parameter rules to obtain preset structural parameters; The bearing capacity reference range is corrected based on the pre-set structural parameters to obtain the bearing capacity correction range; Extract the lower limit of the bearing capacity correction range as the bearing capacity estimate.

[0010] Furthermore, methods for dynamically calculating load margin include: Ground units are dynamically merged based on a joint terrain feature matrix to obtain ground region blocks; Within the ground area block, the minimum value of the estimated bearing capacity corresponding to the ground unit is selected as the benchmark value of the regional bearing capacity. Extract the local pressure peak and local pressure mean within the ground area block; The proportion of edge units is obtained by calculating the ratio of the number of ground units corresponding to the preset edge regions within the ground area block to the total number of ground units within the ground area block. The edge unit proportion is compared with multiple preset proportion ranges to determine the edge proportion range; The pressure concentration factor is determined based on the preset coefficient value rules corresponding to the edge proportion interval; Using the pressure concentration factor and 1 minus the pressure concentration factor as weights, the local pressure peak and local pressure mean are weighted and summed to obtain the actual load value of the region; The bearing margin is calculated based on the ratio of the difference between the regional bearing capacity benchmark value and the actual regional load value to the regional bearing capacity benchmark value.

[0011] Furthermore, methods for dynamically merging ground units based on a joint terrain feature matrix include: According to the arrangement order of ground units in the joint terrain feature matrix, the ground geometric information, ground bearing pressure information and bearing capacity estimate corresponding to adjacent ground units are extracted sequentially. Calculate the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates between adjacent ground units; The differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates are compared with their respective preset merging thresholds. When the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimation are all no greater than the corresponding merging threshold, adjacent ground units are merged into the same ground area block. When at least one of the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates is greater than the corresponding merging threshold, the corresponding adjacent ground units are divided into different ground area blocks.

[0012] Furthermore, methods for calculating the collapse probability include: The ratio of the actual load on the area to the benchmark value of the area's bearing capacity is calculated to obtain the compressive ratio. Calculate the ratio of the peak local pressure to the mean local pressure within a ground area block to obtain the pressure concentration. Calculate the ratio of the difference between the current regional carrying capacity benchmark value and the previous regional carrying capacity benchmark value to the previous regional carrying capacity benchmark value to obtain the carrying capacity attenuation degree; The edge influence degree is obtained by statistically analyzing the proportion of the number of ground units corresponding to the preset edge regions within the ground area block to the total number of ground units within the ground area block. The pressure ratio, pressure concentration, load attenuation, and edge influence are all adjusted to the same dimension. The collapse probability is obtained by weighting and summing the pressure ratio, pressure concentration, bearing capacity attenuation and edge influence after the adjustment based on the preset weighting coefficients.

[0013] Furthermore, methods for generating terrain bearing capacity risk maps with labeled risk levels include: Read the bearing capacity and collapse probability of each surface area block; The first risk level is determined by comparing the load margin with the preset load margin range. The collapse probability is compared with the preset collapse probability range to determine the second risk level; Compare the first risk level and the second risk level to determine the target risk level; Write the target risk level into the corresponding ground area block; Arrange the ground blocks with the target risk level entered according to their ground location, and generate a topographic bearing capacity risk map labeled with the risk level.

[0014] Furthermore, methods for calculating the adaptive obstacle-crossing control parameter set include: Read the bearing margin, subsidence probability, and risk level of the ground area blocks in the terrain bearing risk map; Read the ground geometric information, ground bearing pressure information and bearing capacity estimate corresponding to the ground unit in the joint terrain feature matrix; Based on the arrangement order of the ground area blocks, determine the main control area of ​​the ground area block to be entered; Based on the risk level, load margin, and collapse probability of the main control area, calculate the driving torque and travel speed; Based on the ground geometry information corresponding to the main control area, calculate the vehicle attitude adjustment amount; Based on the ground bearing pressure information and bearing capacity estimate of the main control area, calculate the ground pressure distribution of the wheel set and track; Based on the risk level, bearing capacity, collapse probability, and ground geometry information corresponding to the main control area, determine the obstacle crossing method; The driving torque, travel speed, vehicle attitude adjustment, wheel and track ground pressure distribution, and obstacle crossing method are used as the adaptive obstacle crossing control parameter set.

[0015] Furthermore, methods for planning and optimizing obstacle-crossing paths based on adaptive obstacle-crossing control parameter sets include: Based on the obstacle-crossing methods corresponding to the ground area blocks, the paths of adjacent ground area blocks are combined to form a set of candidate paths; Based on the risk level of the ground area blocks traversed by the candidate path, the path risk value is calculated cumulatively. Based on the local pressure peak values ​​corresponding to the ground area blocks traversed by the candidate path, the path pressure value is calculated cumulatively. Based on the difference in vehicle attitude adjustment between adjacent ground area blocks in the candidate path, the path attitude change value is calculated cumulatively. The path length is calculated cumulatively based on the spatial distance between adjacent ground area blocks in the candidate path. The path risk value, path stress value, path attitude change value, and path length value are processed to be of the same dimension and then weighted and summed according to the preset path weight coefficient to obtain the path evaluation result. Compare the path evaluation results of each candidate path, and select the candidate path with the smallest path evaluation result as the optimized obstacle-crossing path.

[0016] Compared with existing technologies, the technical effects and advantages of the adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning of the present invention are as follows: (1) This invention collects ground geometric information and ground bearing pressure information, and performs terrain feature learning based on the ground geometric information and ground bearing pressure information to construct a joint terrain feature matrix, so that the geometric morphological features and bearing state features of each ground unit form a corresponding relationship in the same data structure. The expression of ground state is no longer based solely on geometric information such as obstacle height, slope, surface roughness and step shape, but also covers ground load state and bearing capacity change information, thereby reducing the judgment bias caused by relying solely on terrain geometric features to determine drivability.

[0017] (2) This invention dynamically calculates the bearing margin and collapse probability of each area of ​​the ground based on the joint terrain feature matrix, and generates a terrain bearing risk map with risk levels marked. The changes in ground bearing capacity under the influence of fire, high temperature, local burn, water immersion and secondary collapse are included in the risk expression process. The appearance is flat ground but the internal bearing capacity has been reduced, such as steel grid, carbonized wood board, burnt sandwich floor and partially collapsed ceiling. It can be distinguished from ordinary low-risk passable ground in the risk assessment process, so that the terrain assessment results can maintain a better correspondence with the actual operation risk.

[0018] (3) This invention calculates an adaptive obstacle-crossing control parameter set based on the terrain bearing risk map and the joint terrain feature matrix, and plans and optimizes the obstacle-crossing path based on the adaptive obstacle-crossing control parameter set, controlling the fire-fighting robot to move along the optimized obstacle-crossing path. The driving torque, travel speed, vehicle posture adjustment amount, wheel and track ground pressure distribution, and obstacle-crossing mode can be adjusted accordingly with the ground risk state and terrain state. Unsuitable paths can be identified and avoided, thereby reducing local collapse, overturning, and loss of control caused by overestimation of terrain passability, and reducing over-adjustment and unnecessary detours caused by underestimation of terrain passability. Attached Figure Description

[0019] Figure 1 This is a flowchart of the adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning, according to an embodiment of the present invention. Figure 2 This is a flowchart of the dynamic merging process of ground units based on the joint terrain feature matrix in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0021] Example 1: Please see Figure 1 As shown, this embodiment provides an adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning, including: collecting ground geometric information and ground bearing pressure information, performing time alignment and spatial correlation on the ground geometric information and ground bearing pressure information to form a spatially correlated terrain dataset.

[0022] It should be added that the firefighting robot is an operational platform used to perform movement and obstacle crossing tasks in complex firefighting environments, and is used to complete movement operations in environments affected by fire, high temperature, localized burns, water immersion, and secondary collapse; the ground geometry information includes obstacle height, slope, surface roughness, and step shape, used to describe the terrain features of the current working surface of the firefighting robot; the ground bearing pressure information includes local pressure distribution, local pressure peak value, and local pressure average value, used to characterize the bearing status and potential risks of the current working surface.

[0023] It should be noted that the preferred range for collecting ground geometry information and ground bearing pressure information is the ground area that the fire-fighting robot's wheels and tracks have contacted, are currently bearing load on, or are about to enter, so as to ensure that the subsequent spatially associated terrain dataset corresponds to the actual obstacle-crossing range of the fire-fighting robot.

[0024] Specifically, before and during the operation of the firefighting robot, the obstacle height, slope, surface roughness, and step shape of the current working ground are collected to form ground geometric information; the local pressure distribution, local pressure peak, and local pressure average of the current working ground are collected simultaneously to form ground bearing pressure information; the ground position and corresponding collection time of the firefighting robot during the collection process are recorded to form the firefighting robot's movement trajectory; combined with the firefighting robot's movement trajectory, the ground geometric information and ground bearing pressure information are mapped according to the collection time; the ground geometric information and ground bearing pressure information mapped according to the collection time are mapped according to the ground position to form a spatially correlated terrain dataset.

[0025] It should be noted that the sampling frequency is preferably between 10Hz and 30Hz; when the ground has many undulations, the fire-fighting robot moves at a high speed, or the local pressure changes rapidly, a larger sampling frequency should be used; when the ground is relatively flat, the fire-fighting robot moves at a low speed, or the local pressure changes slowly, a smaller sampling frequency should be used. For time alignment, a preset time range is preferably set, preferably between 0.02s and 0.2s; when the ground changes rapidly or the movement speed is high, a smaller preset time range should be used; when the ground changes slowly or the movement speed is low, a larger preset time range should be used. For spatial association, a preset spatial range is preferably set, preferably between 5mm and 20mm; when there are many ground edges and frequent changes in obstacle height, a smaller preset spatial range should be used; when the ground is relatively flat and there are many continuous gentle slopes, a larger preset spatial range should be used. The purpose of this step is to form a spatially associated terrain dataset that can be directly used for subsequent terrain feature learning.

[0026] Terrain features are learned based on spatially associated terrain datasets, and a joint terrain feature matrix is ​​constructed.

[0027] It should be added that the terrain feature learning described herein is a process of extracting, merging, and jointly representing the correspondence between the geometric morphology features and bearing capacity features of various ground units based on spatially correlated terrain datasets. This process forms a joint terrain feature matrix that can be used for bearing capacity risk assessment and obstacle crossing control. The joint terrain feature matrix is ​​a dataset used to uniformly represent the ground geometry and bearing capacity information of various ground units, describing the geometric morphology and bearing capacity features of ground units in front of the firefighting robot and within its current operating area. Each ground unit in the joint terrain feature matrix corresponds to a fixed set of data, including ground geometry information, ground bearing capacity information, and bearing capacity estimates. Terrain feature learning serves to uniformly represent ground geometry and bearing capacity information within the same ground unit, enabling a one-to-one correspondence between terrain morphology features and ground bearing capacity.

[0028] Specifically, ground units are defined based on the ground location in the spatially associated terrain dataset. When there are many ground edges, rapid changes in obstacle height, and rapid changes in local pressure distribution, a smaller ground unit size is used; when the ground is relatively flat, the changes in obstacle height are gradual, and the changes in local pressure distribution are gradual, a larger ground unit size is used. The preferred ground unit size is 20mm to 100mm.

[0029] It should be added that after the ground units are divided, the same ground location corresponds to only one ground unit, thereby avoiding the duplicate assignment of the same ground location.

[0030] Specifically, based on ground geometry information and ground bearing pressure information, the estimated bearing capacity of each ground unit is calculated.

[0031] It should be added that the estimated bearing capacity is data used to characterize the load level that each ground unit can withstand under the current material and structural conditions. The calculation process of the estimated bearing capacity includes: extracting the local pressure distribution from the ground bearing pressure information; performing material matching based on ground geometry information to obtain the ground material type; extracting a preset material bearing reference range table based on the ground material type to obtain the bearing capacity reference range; extracting structural characterization information based on ground geometry information and ground bearing pressure information, and matching the structural characterization information with preset structural parameter rules to obtain preset structural parameters; correcting the bearing capacity reference range according to the preset structural parameters to obtain the corrected bearing capacity range; and extracting the lower limit of the corrected bearing capacity range as the estimated bearing capacity.

[0032] Specifically, during material matching, rule-based matching is performed based on surface opening features, edge continuity features, surface undulation features, and step shape features in the ground geometry information. When the surface opening features have a periodic grid distribution and the edge continuity features correspond to a discrete strip structure, the ground material type is determined to be steel grating; when the surface undulation features correspond to continuous plate-like undulations and the step shape is relatively weak, the ground material type is determined to be sandwich panel; when the ground geometry information corresponds to wood-like plate-like undulation features, the ground material type is determined to be wood-based panel; when the ground surface continuity is high and the surface roughness distribution is relatively uniform, the ground material type is determined to be concrete floor.

[0033] It should be added that the preferred preset ranges for material load-bearing capacity are: 60 kPa to 180 kPa for steel grating, 30 kPa to 120 kPa for sandwich panels, 15 kPa to 80 kPa for wood-based panels, and 120 kPa to 400 kPa for concrete floors. When the floor material is significantly affected by fire, high temperature, or localized burning, the upper limit of the corresponding load-bearing capacity range is not directly increased. Instead, the load-bearing capacity range is adjusted downwards using preset structural parameters to avoid overestimating the load-bearing capacity.

[0034] Specifically, the structural characterization information preferably includes support span characteristics, edge opening density characteristics, slab surface continuity characteristics, local pressure diffusion width characteristics, and local pressure concentration characteristics. Support span characteristics are extracted by the distance between adjacent edge locations on the ground; edge opening density characteristics are extracted by the number of openings or edge interruptions per unit area; slab surface continuity characteristics are extracted by the proportion of continuous plane length; local pressure diffusion width characteristics are extracted by the diffusion range of local pressure distribution; and local pressure concentration characteristics are extracted by the relationship between the local pressure peak value and the local pressure mean value.

[0035] It should be added that, when matching structural characterization information with preset structural parameter rules, it is necessary to determine which preset parameter interval each of the support span characteristics, edge opening density characteristics, plate surface continuity characteristics, local pressure diffusion width characteristics, and local pressure concentration degree characteristics falls into, and then determine the preset structural parameters based on the combination of the parameter intervals they fall into. Preferred preset structural parameters include thickness grade, support spacing grade, material strength correction factor, burn-off correction factor, and water immersion correction factor.

[0036] Furthermore, the support spacing is preferably divided into three levels: 100mm to 300mm, 300mm to 600mm, and 600mm to 1000mm; the thickness is preferably divided into three levels: 5mm to 15mm, 15mm to 30mm, and 30mm to 50mm; the burn-off correction factor is preferably 0.3 to 1.0; and the water immersion correction factor is preferably 0.5 to 1.0. When the ground edge opening density characteristic is high, the slab surface continuity characteristic is low, the local pressure diffusion width is narrow, or the local pressure concentration degree is high, the structural preset parameters are taken as the corresponding lower load-bearing level; when the ground edge opening density characteristic is low, the slab surface continuity characteristic is high, the local pressure diffusion width is wide, or the local pressure concentration degree is low, the structural preset parameters are taken as the corresponding higher load-bearing level.

[0037] Specifically, the bearing capacity baseline range is corrected based on the pre-set structural parameters to obtain the bearing capacity correction range. Once the bearing capacity baseline range corresponding to the ground material type has been determined, the upper and lower limits of the bearing capacity baseline range are simultaneously or progressively lowered according to thickness grade, support spacing grade, material strength correction factor, burn damage correction factor, and water immersion correction factor, forming the bearing capacity correction range. When the ground material is significantly affected by fire, high temperature, localized burn damage, or water immersion, the bearing capacity correction range shifts generally towards a lower value range; when the ground material is relatively intact and continuous support is sufficient, the bearing capacity correction range remains in a higher range. After the bearing capacity correction range is formed, the lower limit of the bearing capacity correction range is extracted as the bearing capacity estimate. Extracting the lower limit as the bearing capacity estimate reflects the minimum bearing level under localized weak conditions, and can avoid overestimating the ground's load-bearing capacity in subsequent bearing risk calculations.

[0038] Specifically, according to the corresponding ground location, the ground geometry information, ground bearing pressure information, and bearing capacity estimate are written into the corresponding ground unit; the arrangement order of each ground unit is determined based on the movement trajectory of the fire-fighting robot; the ground units are arranged according to the arrangement order to construct a joint terrain feature matrix.

[0039] It should be added that the arrangement order is preferably determined by combining the order in the forward direction and the order in the lateral deployment direction of the fire-fighting robot; the ground units in the forward direction are the primary arrangement direction, and the ground units in the lateral deployment direction are the secondary arrangement direction. When a ground unit is located at the edge of a threshold, the edge of a step, the edge of a steel grating opening, a joint of a panel, a local collapse boundary, or a local overhang boundary, the ground unit is recorded independently in the joint terrain feature matrix; when a ground unit is located on a continuous plane or a continuous gentle slope, and the corresponding ground geometry information, ground bearing pressure information, and bearing capacity estimate change within a continuous range, the ground unit is recorded in the joint terrain feature matrix in a continuous positional order.

[0040] Furthermore, the joint terrain feature matrix is ​​preferably updated continuously according to a matrix update cycle, which is preferably between 0.03s and 0.1s. When the ground changes rapidly or the fire-fighting robot moves at a high speed, the matrix update cycle takes a smaller value; when the ground changes slowly or the fire-fighting robot moves at a low speed, the matrix update cycle takes a larger value. The purpose of this step is to form a joint terrain feature matrix that uniformly incorporates ground geometric information, ground bearing pressure information, and bearing capacity estimates.

[0041] Finally, the joint terrain feature matrix is ​​output. Each ground cell in the joint terrain feature matrix contains corresponding ground geometry information, ground bearing pressure information, and bearing capacity estimates. The joint terrain feature matrix serves as a unified data input for subsequent calculations of bearing margin and subsidence probability of ground blocks.

[0042] Based on the joint terrain feature matrix, the bearing capacity margin and collapse probability are dynamically calculated to generate a terrain bearing capacity risk map with marked risk levels.

[0043] It should be added that the calculation objects for the bearing margin and collapse probability are ground area blocks. A ground area block is a local ground region formed by dynamically merging multiple adjacent ground units in the joint terrain feature matrix, used to characterize the load-bearing area of ​​the firefighting robot on its continuous contact path. The bearing margin characterizes the remaining bearing space of a ground area block under the current load state; the collapse probability characterizes the possibility of bearing failure of a ground area block under the current load state and during continuous loading; and the terrain bearing risk map characterizes the risk level distribution corresponding to each ground area block.

[0044] Specifically, ground units are dynamically merged based on a joint terrain feature matrix to obtain ground region blocks. See also... Figure 2 As shown, during dynamic merging, the ground geometric information, ground bearing pressure information, and bearing capacity estimate corresponding to adjacent ground units are extracted sequentially according to the arrangement order of ground units in the joint terrain feature matrix. The differences in ground geometric information, ground bearing pressure information, and bearing capacity estimate corresponding to adjacent ground units are calculated. These differences are then compared with corresponding preset merging thresholds. When the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimate are all no greater than the corresponding merging threshold, adjacent ground units are merged into the same ground region block. When at least one of the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimate is greater than the corresponding merging threshold, the corresponding adjacent ground units are divided into different ground region blocks.

[0045] It should be added that the ground geometric information differences preferably include obstacle height differences, slope differences, surface roughness differences, and step shape differences; the ground bearing pressure information differences preferably include local pressure distribution differences, local pressure peak differences, and local pressure mean differences. When calculating the ground geometric information differences corresponding to adjacent ground units, obstacle height differences, slope differences, surface roughness differences, and step shape differences are calculated separately; when calculating the ground bearing pressure information differences corresponding to adjacent ground units, local pressure distribution differences, local pressure peak differences, and local pressure mean differences are calculated separately; when the aforementioned differences are not greater than the corresponding merging thresholds, it is determined that adjacent ground units meet the merging conditions for the corresponding information items.

[0046] Furthermore, the differences in step shape are preferably compared based on the height of the step's leading edge, the length of the step's upper surface, and the continuity of the step's edge. When the differences in all three items are not greater than their respective preset difference thresholds, the differences in step shape are determined to meet the merging condition. When at least one of the three items is greater than its corresponding preset difference threshold, the differences in step shape are determined not to meet the merging condition. The differences in local pressure distribution are preferably compared based on the differences in local pressure diffusion width and the offset of local pressure concentration location. When both the differences in local pressure diffusion width and the offset of local pressure concentration location are not greater than their respective preset thresholds, the differences in local pressure distribution are determined to meet the merging condition.

[0047] Furthermore, the obstacle height merging threshold is preferably 5mm to 30mm, the slope merging threshold is preferably 2° to 8°, the surface roughness merging threshold is preferably 0.5mm to 3mm, the local pressure peak merging threshold is preferably 5kPa to 20kPa, the local pressure average merging threshold is preferably 3kPa to 15kPa, and the bearing capacity estimate merging threshold is preferably 5kPa to 25kPa. When there are many ground edges and the ground changes rapidly, each merging threshold takes a smaller value; when the ground is relatively continuous and the ground changes gently, each merging threshold takes a larger value. The purpose of this step is to form the ground area block required for subsequent bearing margin and collapse probability calculations.

[0048] Specifically, the method for dynamically calculating the carrying capacity margin includes: dynamically merging ground units based on a joint terrain feature matrix to obtain ground area blocks; within each ground area block, selecting the minimum value of the estimated carrying capacity corresponding to the ground unit as the benchmark value of the regional carrying capacity; extracting the local pressure peak value and local pressure mean value within the ground area block; calculating the proportion of ground units corresponding to preset edge areas within the ground area block to the total number of ground units within the ground area block, obtaining the edge unit proportion; comparing the edge unit proportion with multiple preset proportion intervals to determine the edge proportion interval; determining the pressure concentration coefficient according to the preset coefficient value rules corresponding to the edge proportion interval; using the pressure concentration coefficient and 1 minus the pressure concentration coefficient as weights, performing a weighted summation of the local pressure peak value and local pressure mean value to obtain the actual regional load value; and calculating the carrying capacity margin based on the proportion of the difference between the regional carrying capacity benchmark value and the actual regional load value to the regional carrying capacity benchmark value.

[0049] It should be added that, when extracting the local pressure peak value within a ground area block, the local pressure peak value corresponding to all ground units within the ground area block is extracted, and the maximum value among them is selected as the local pressure peak value corresponding to the ground area block; when extracting the local pressure average value within a ground area block, the local pressure average value corresponding to all ground units within the ground area block is extracted, and all local pressure average values ​​are averaged and then used as the local pressure average value corresponding to the ground area block.

[0050] Furthermore, the pre-defined edge region is preferably the area formed by extending inward from the edge of a threshold, step, steel grating opening, panel joint, local collapse boundary, and local overhang boundary by one to two ground units. The preferred ranges for the proportion of edge units are ≤10%, >10% and ≤25%, >25% and ≤40%, and >40%. The corresponding compressive concentration factors are preferably set to 0.60, 0.65, 0.72, and 0.80. When the ground area block is located in a continuous area with fewer edge locations, a smaller compressive concentration factor is used; when the ground area block is located in a complex area with more edge locations, a larger compressive concentration factor is used. The regional bearing capacity benchmark value is not an average or maximum value, but a minimum value is used so that the minimum bearing boundary of the ground area block can reflect the limiting effect of locally weak ground units. The regional bearing capacity benchmark value characterizes the minimum bearing boundary of the ground area block, the actual load value characterizes the current load level of the ground area block, and the bearing margin characterizes the remaining bearing space of the ground area block.

[0051] Furthermore, the bearing margin is preferably recorded as a percentage range, with the preferred bearing margin ranges being greater than 20%, greater than 10% and less than or equal to 20%, greater than or equal to 0 and less than or equal to 10%, and less than 0. When the ground material is severely affected by fire, high temperature, localized burning, water immersion, and secondary collapse, the bearing margin range boundaries are subject to stricter requirements; when the ground material is relatively intact, has many continuous planes, and the local pressure distribution is relatively uniform, the bearing margin range boundaries are subject to conventional requirements.

[0052] Specifically, the method for calculating the collapse probability includes: calculating the ratio of the actual load value of the area to the benchmark value of the area's bearing capacity to obtain the pressure ratio; calculating the ratio of the peak local pressure to the average local pressure within the ground area block to obtain the pressure concentration; calculating the ratio of the difference between the current benchmark value of the area's bearing capacity and the previous benchmark value to the previous benchmark value of the area's bearing capacity to obtain the bearing capacity attenuation; calculating the ratio of the number of ground units corresponding to the preset edge areas within the ground area block to the total number of ground units within the ground area block to obtain the edge influence; performing dimensionless normalization on the pressure ratio, pressure concentration, bearing capacity attenuation, and edge influence; and weighting and summing the normalized pressure ratio, pressure concentration, bearing capacity attenuation, and edge influence according to preset weighting coefficients to obtain the collapse probability.

[0053] It should be added that the preferred correspondence rule between the current regional carrying capacity benchmark value and the previous regional carrying capacity benchmark value is as follows: Correspondence is based on the ground location and arrangement order of the ground area blocks in the joint terrain feature matrix. When the same ground location has corresponding ground area blocks in two adjacent matrix update cycles, these two are determined as the corresponding regional carrying capacity benchmark values. When the ground location offset of a ground area block in two adjacent matrix update cycles is no more than one ground cell, it is still determined as a correspondence. When the ground area block corresponding to the current value does not exist in the previous value, the carrying capacity attenuation is recorded as 0.

[0054] Further, the pressure ratio rounding interval is preferably set to less than or equal to 0.5, greater than 0.5 and less than or equal to 0.8, greater than 0.8 and less than or equal to 1.0, and greater than 1.0; the pressure concentration rounding interval is preferably set to less than or equal to 1.2, greater than 1.2 and less than or equal to 1.6, greater than 1.6 and less than or equal to 2.0, and greater than 2.0; the load-bearing attenuation rounding interval is preferably set to less than or equal to 5%, greater than 5% and less than or equal to 15%, greater than 15% and less than or equal to 30%, and greater than 30%; the edge influence rounding interval is preferably set to less than or equal to 10%, greater than 10% and less than or equal to 25%, greater than 25% and less than or equal to 40%, and greater than 40%. The pressure ratio weighting coefficient in the preset weighting coefficients is preferably 0.35 to 0.5, the pressure concentration weighting coefficient is preferably 0.2 to 0.3, the load-bearing attenuation weighting coefficient is preferably 0.15 to 0.25, and the edge influence weighting coefficient is preferably 0.1 to 0.2. The probability of collapse is preferably recorded according to probability intervals, which are preferably set as less than or equal to 5%, greater than 5% and less than or equal to 15%, greater than 15% and less than or equal to 35%, and greater than 35%. When the ground area is located in a steel grating area, carbonized wood board area, burned sandwich floor area, or partially collapsed ceiling area after a fire, the collapse probability interval limit shall be subject to stricter requirements; when the ground area is located in a complete and continuous plane area or a complete and continuous gentle slope area, the collapse probability interval limit shall be subject to conventional requirements.

[0055] Specifically, the method for generating a topographic bearing capacity risk map with risk levels marked includes: reading the bearing margin and subsidence probability corresponding to each ground area block; comparing the bearing margin with a preset bearing margin interval to determine a first risk level; comparing the subsidence probability with a preset subsidence probability interval to determine a second risk level; comparing the first risk level and the second risk level to determine a target risk level; writing the target risk level into the corresponding ground area block; and arranging the ground area blocks after writing the target risk level according to the ground location of each ground area block to generate a topographic bearing capacity risk map with risk levels marked.

[0056] It should be added that both the first and second risk levels are preferably set as low risk, medium risk, high risk, and extremely high risk levels. When comparing the first and second risk levels, the one with the higher risk level is preferred as the target risk level. For ease of subsequent path evaluation, the low risk, medium risk, high risk, and extremely high risk levels are preferably mapped to risk level values ​​1, 2, 3, and 4, respectively. The purpose of this step is to unify the bearing capacity margin and collapse probability into risk level results that can be directly used for control decisions.

[0057] Finally, a terrain bearing capacity risk map with labeled risk levels is output. Each ground area block in the terrain bearing capacity risk map has a corresponding bearing margin, collapse probability, and target risk level, and maintains a consistent correspondence with the ground location in the joint terrain feature matrix. The terrain bearing capacity risk map with labeled risk levels serves to provide direct risk input for subsequent calculation of adaptive obstacle crossing control parameter sets.

[0058] Based on the terrain bearing risk map and the joint terrain feature matrix, the adaptive obstacle crossing control parameter set is calculated.

[0059] It should be added that the adaptive obstacle-crossing control parameter set is a set of parameters used to control the firefighting robot's movement and obstacle crossing in different ground areas. It reflects the firefighting robot's driving torque, travel speed, vehicle attitude adjustment, wheel and track ground pressure distribution, and obstacle-crossing method under the current ground load and terrain conditions. The driving torque characterizes the output driving force of the firefighting robot in the corresponding ground area; the travel speed characterizes the passing speed of the firefighting robot in the corresponding ground area; the vehicle attitude adjustment characterizes the range of vehicle attitude changes made by the firefighting robot in the corresponding ground area to adapt to obstacle height, slope, and step shape; the wheel and track ground pressure distribution characterizes the adjustment relationship of the force ratio of the wheel and track in the corresponding ground area; and the obstacle-crossing method characterizes the control method used by the firefighting robot in the corresponding ground area, including direct passage, unloaded passage, edge avoidance passage, low-speed crossing passage, or stopping entry.

[0060] Specifically, the method for calculating the adaptive obstacle crossing control parameter set includes: reading the bearing margin, collapse probability, and risk level of the ground area blocks corresponding to the terrain bearing risk map; reading the ground geometry information, ground bearing pressure information, and bearing capacity estimate of the ground units in the joint terrain feature matrix; determining the main control area of ​​the ground area block to be entered based on the arrangement order of the ground area blocks; calculating the driving torque and travel speed based on the risk level, bearing margin, and collapse probability of the main control area; calculating the vehicle attitude adjustment amount based on the ground geometry information of the main control area; calculating the wheel and track ground pressure distribution based on the ground bearing pressure information and bearing capacity estimate of the main control area; determining the obstacle crossing method based on the risk level, bearing margin, collapse probability, and ground geometry information of the main control area; and using the driving torque, travel speed, vehicle attitude adjustment amount, wheel and track ground pressure distribution, and obstacle crossing method as the adaptive obstacle crossing control parameter set.

[0061] It should be added that the primary control area is preferably the ground area block that is closest to and has not yet been passed in the current direction of travel of the firefighting robot. When the distance between a ground area block and the leading edge of the firefighting robot is less than or equal to one vehicle length, the ground area block is identified as the ground area block to be entered; when the distance between a ground area block and the leading edge of the firefighting robot is greater than one vehicle length, it is not considered as the primary control area for the current time. The preset distance threshold is preferably one to two ground unit lengths. When there are multiple ground area blocks to be entered that are at the same distance from the leading edge of the firefighting robot or whose distance difference is not greater than the preset distance threshold, the ground area block with the higher risk level is prioritized as the primary control area; when the risk levels are the same, the ground area block with the smaller load margin is prioritized as the primary control area.

[0062] Furthermore, when the risk level is low, the driving torque is preferably 60% to 90% of the rated output of the fire-fighting robot, and the traveling speed is preferably 0.4 m / s to 1.0 m / s; when the risk level is medium, the driving torque is preferably 45% to 75% of the rated output of the fire-fighting robot, and the traveling speed is preferably 0.2 m / s to 0.6 m / s; when the risk level is high, the driving torque is preferably 30% to 55% of the rated output of the fire-fighting robot, and the traveling speed is preferably 0.05 m / s to 0.3 m / s; when the risk level is extremely high, the driving torque is preferably a holding torque of 0% to 10% of the rated output of the fire-fighting robot, and the traveling speed is 0 m / s. When the bearing capacity of the ground area is in a low range, or the probability of collapse is in a high range, both the driving torque and the traveling speed are taken from the lower values ​​within the corresponding risk level range; when the bearing capacity of the ground area is in a high range and the probability of collapse is in a low range, both the driving torque and the traveling speed are taken from the higher values ​​within the corresponding risk level range.

[0063] Furthermore, the vehicle attitude adjustment amount is preferably between 0° and 12°. When the obstacle height corresponding to the ground area block is large, the slope is steep, or the step shape is obvious, a larger value is taken for the vehicle attitude adjustment amount; when the ground area block is located in a continuous plane or a continuous gentle slope area, a smaller value is taken for the vehicle attitude adjustment amount. The distribution of ground pressure of the wheel set and track is preferably recorded according to the load ratio of the front wheel set and track, the rear wheel set and track, and the left and right wheel sets and tracks. When the front wheel set and track are about to cross the edge of the threshold or the edge of the step, the load ratio borne by the front wheel set and track is preferably 35% to 55%, and the load ratio borne by the rear wheel set and track is preferably 45% to 65%; when the estimated load capacity of the ground area block is low, the difference in the load ratio of the left and right wheel sets and tracks is preferably controlled within 10%; when the estimated load capacity of the ground area block is high and the surface roughness is small, the difference in the load ratio of the left and right wheel sets and tracks is preferably controlled within 20%.

[0064] Furthermore, when the risk level is low, the load-bearing margin is large, the probability of collapse is low, and the obstacle height is small, the obstacle crossing method is determined to be direct passage; when the load-bearing margin is in the medium range, the probability of collapse is in the medium range, and the slope is gentle, the obstacle crossing method is determined to be unloaded passage; when the ground area block is located near the edge of a threshold, the edge of a steel grating opening, a panel splice, or the boundary of a local collapse, the obstacle crossing method is determined to be edge avoidance passage; when the obstacle height of the ground area block is large and the load-bearing margin is in the low range, the obstacle crossing method is determined to be low-speed crossing passage; when the ground area block is marked as extremely high risk level, the obstacle crossing method is determined to be stop entry. The purpose of this step is to transform the risk results and terrain status into an adaptive obstacle crossing control parameter set that the fire-fighting robot can directly access.

[0065] Finally, an adaptive obstacle-crossing control parameter set is output. Each main control region in the adaptive obstacle-crossing control parameter set has corresponding drive torque, travel speed, vehicle attitude adjustment, wheel and track ground pressure distribution, and obstacle-crossing mode, and maintains a consistent ground position correspondence and matrix update cycle correspondence with the terrain bearing risk map and joint terrain feature matrix. The adaptive obstacle-crossing control parameter set serves to provide control inputs and path constraints for subsequent optimization of obstacle-crossing path planning.

[0066] The obstacle-crossing path is planned and optimized based on the adaptive obstacle-crossing control parameter set, and the fire-fighting robot is controlled to move along the optimized obstacle-crossing path according to the adaptive obstacle-crossing control parameter set.

[0067] It should be added that the optimized obstacle-crossing path is a passage path used to guide the firefighting robot to move and cross obstacles within the current ground area block. The optimized obstacle-crossing path corresponds at least to the passage sequence, entry position, crossing position, avoidance position, and passage direction of the ground area block. The entry position is the initial contact position when the firefighting robot enters the target ground area block. The crossing position is the target passage position when the firefighting robot passes through the edge of a threshold, the edge of a step, a slope turn, or a local collapse boundary. The avoidance position is the detour position when the firefighting robot avoids areas corresponding to extremely high risk levels, areas corresponding to high risk levels, the edge of steel grating openings, panel seams, and local collapse boundaries. The passage direction is the forward direction of the firefighting robot when continuously traversing between adjacent ground area blocks. The optimized obstacle-crossing path has the function of converting the adaptive obstacle-crossing control parameter set into a continuous passage trajectory.

[0068] Specifically, the method for planning and optimizing obstacle-crossing paths based on an adaptive obstacle-crossing control parameter set includes: combining paths between adjacent ground area blocks according to their obstacle-crossing methods to form a candidate path set; accumulating path risk values ​​based on the risk levels of the ground area blocks traversed by the candidate paths; accumulating path pressure values ​​based on the local pressure peak values ​​of the ground area blocks traversed by the candidate paths; accumulating path attitude change values ​​based on the differences in vehicle attitude adjustment between adjacent ground area blocks in the candidate paths; accumulating path length values ​​based on the spatial distance between adjacent ground area blocks in the candidate paths; performing dimensionless normalization on the path risk values, path pressure values, path attitude change values, and path length values, and weighting and summing them according to preset path weight coefficients to obtain path evaluation results; comparing the path evaluation results of each candidate path, and selecting the candidate path with the smallest path evaluation result as the optimized obstacle-crossing path.

[0069] It should be added that when the candidate path set is formed, only ground areas where the obstacle crossing method is direct passage, unloaded passage, edge avoidance passage, or low-speed crossing passage are allowed. Ground areas where the obstacle crossing method is "stop to enter" are not allowed. The number of candidate paths is preferably 2 to 6; when there are fewer passable branches ahead, the number of candidate paths should be smaller; when there are more passable branches ahead, the number of candidate paths should be larger.

[0070] Furthermore, the preferred rule for calculating the cumulative path risk value is as follows: low risk level is mapped to 1, medium risk level to 2, high risk level to 3, and extremely high risk level to 4. The risk level mapping values ​​corresponding to each ground area block traversed by the candidate path are summed cumulatively to obtain the path risk value. The preferred rule for calculating the cumulative path pressure value is as follows: local pressure peak values ​​corresponding to each ground area block traversed by the candidate path are extracted, and the local pressure peak values ​​are summed cumulatively to obtain the path pressure value. The preferred rule for calculating the cumulative path attitude change value is as follows: vehicle attitude adjustment amounts corresponding to adjacent ground area blocks in the candidate path are extracted, and the differences between adjacent vehicle attitude adjustment amounts are summed cumulatively to obtain the path attitude change value; when the candidate path contains only one ground area block, the path attitude change value is recorded as 0. The preferred rule for calculating the cumulative path length value is as follows: spatial distances between adjacent ground area blocks in the candidate path are extracted, and the spatial distances of each segment are summed cumulatively to obtain the path length value.

[0071] Furthermore, the path risk value rounding interval is preferably set to 4 to 40, the path pressure value rounding interval is preferably set to 20 kPa to 300 kPa, the path attitude change value rounding interval is preferably set to 0° to 60°, and the path length value rounding interval is preferably set to 0 m to 5 m. The preset path weighting coefficients are preferably 0.4 to 0.6 for the path risk value, 0.15 to 0.25 for the path pressure value, 0.15 to 0.25 for the path attitude change value, and 0.1 to 0.2 for the path length value. When the ground bearing risk ahead is high, the path risk value weighting coefficient should be larger; when the ground bearing risk ahead is low but the path smoothness requirement is high, the path attitude change value weighting coefficient should be larger; when the local mission requires rapid passage, the path length value weighting coefficient should be larger.

[0072] Furthermore, when at least two candidate paths have the same path evaluation result, the candidate path with the lower path risk value is selected first; when the path risk values ​​are the same, the candidate path with the lower path length value is selected first; when the path length values ​​are still the same, the candidate path with the lower path attitude change value is selected first. The purpose of this step is to provide clear data evaluation basis for optimizing obstacle-crossing paths.

[0073] Specifically, based on the optimized obstacle-crossing path, the passage sequence, entry position, crossing position, avoidance position, and passage direction of the ground area blocks are determined. The entry and crossing positions are preferably selected within a range of 20% to 80% of the width of the corresponding ground area block. When the difference in ground pressure distribution between the wheel set and track corresponding to the ground area block is small, the entry and crossing positions are taken from the middle range; when one side of the ground area block is close to the edge of the steel grating opening, plate joint, local overhang boundary, or local collapse boundary, the entry and crossing positions are far from that edge. The avoidance position is preferably set at a position 10% to 35% away from the edge, or at a position with a high estimated bearing capacity and relatively uniform local pressure distribution within 1 to 3 ground units in front of the crossing position. The aforementioned position selection rules help reduce local pressure concentration and avoid weak edge positions.

[0074] Specifically, the firefighting robot is controlled to move along an optimized obstacle-crossing path based on an adaptive obstacle-crossing control parameter set. During control, the robot's passage sequence, entry position, crossing position, avoidance position, and passage direction for each ground area are determined based on the optimized obstacle-crossing path. Corresponding drive torque, travel speed, vehicle attitude adjustment, wheel and track ground pressure distribution, and obstacle-crossing mode are then activated to control the robot's obstacle-crossing movement along the optimized path. When the obstacle-crossing mode corresponding to the main control area ahead is "stop entering," a "stop entering" control is output; when the obstacle-crossing mode corresponding to the main control area ahead is not "stop entering," corresponding drive torque, travel speed, vehicle attitude adjustment, and wheel and track ground pressure distribution control are output. The obstacle-crossing movement includes straight-through movement, unloaded movement, edge avoidance movement, low-speed crossing movement, and stop entering control. The purpose of this step is to create a continuous closed loop between the aforementioned terrain feature learning, risk assessment, parameter calculation, and path planning.

[0075] Finally, the optimized obstacle-crossing path is output. Each ground area block in the optimized obstacle-crossing path has a corresponding passage order, entry position, crossing position, avoidance position, and passage direction, and maintains a consistent ground position correspondence and matrix update cycle correspondence with the adaptive obstacle-crossing control parameter set. Ground geometry information and ground bearing pressure information are used to form a spatially associated terrain dataset. The spatially associated terrain dataset is used for terrain feature learning and to generate a joint terrain feature matrix. The joint terrain feature matrix is ​​used to dynamically merge ground area blocks and to calculate the bearing margin and collapse probability of the corresponding ground area blocks, generating a terrain bearing risk map with risk levels marked. The terrain bearing risk map and the joint terrain feature matrix are used to generate an adaptive obstacle-crossing control parameter set. The adaptive obstacle-crossing control parameter set is used to generate a candidate path set, path evaluation results, and an optimized obstacle-crossing path, and is used to control the fire-fighting robot to move along the optimized obstacle-crossing path. The data input and output relationship is continuous. The optimized obstacle-crossing path plays a role in completing continuous path planning based on ground bearing status, terrain morphology characteristics, and obstacle-crossing control requirements, which can improve the safety and task execution continuity of the fire-fighting robot during movement and obstacle crossing in complex fire-fighting environments.

[0076] 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 variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should 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.

[0077] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fire-fighting robot adaptive obstacle-crossing control method based on terrain feature learning, characterized in that, include: Collect ground geometry information and ground bearing pressure information, perform time alignment and spatial correlation on the ground geometry information and ground bearing pressure information to form a spatially correlated terrain dataset; Terrain features are learned based on spatially correlated terrain datasets to construct a joint terrain feature matrix; Based on the joint terrain feature matrix, the bearing margin and collapse probability are dynamically calculated to generate a terrain bearing risk map with marked risk levels. Calculate the adaptive obstacle crossing control parameter set based on the terrain bearing risk map and the joint terrain feature matrix; The obstacle-crossing path is planned and optimized based on the adaptive obstacle-crossing control parameter set, and the fire-fighting robot is controlled to move along the optimized obstacle-crossing path according to the adaptive obstacle-crossing control parameter set.

2. The fire-fighting robot adaptive obstacle-crossing control method based on terrain feature learning according to claim 1, characterized in that, Methods for generating spatially correlated terrain datasets include: Record the ground position and corresponding collection time of the fire-fighting robot during the data collection process to form the movement trajectory of the fire-fighting robot; By combining the movement trajectory of the firefighting robot, the ground geometry information and ground bearing pressure information are correlated according to the collection time; The ground geometric information and ground bearing pressure information, which are corresponding to the collection time, are mapped to ground locations to form a spatially correlated terrain dataset.

3. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 2, characterized in that, Methods for constructing a joint terrain feature matrix include: Ground units are divided based on their location in the spatially correlated terrain dataset; Based on ground geometry information and ground bearing pressure information, the estimated bearing capacity of each ground unit is calculated. According to the corresponding ground location, the ground geometry information, ground bearing pressure information, and bearing capacity estimate are written into the corresponding ground unit; Based on the movement trajectory of the firefighting robot, determine the arrangement order of each ground unit; Arrange the surface units according to the given order to construct a joint terrain feature matrix.

4. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 3, characterized in that, Methods for calculating the estimated bearing capacity of each surface element include: Extracting local pressure distribution from ground bearing pressure information; Material matching is performed based on ground geometry information to obtain the ground material type; Based on the ground material type, the preset material bearing capacity benchmark range table is extracted to obtain the bearing capacity benchmark range; Structural characterization information is extracted based on ground geometric information and ground bearing pressure information, and the structural characterization information is matched with preset structural parameter rules to obtain preset structural parameters; The bearing capacity reference range is corrected based on the pre-set structural parameters to obtain the bearing capacity correction range; Extract the lower limit of the bearing capacity correction range as the bearing capacity estimate.

5. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 1, characterized in that, Methods for dynamically calculating load margin include: Ground units are dynamically merged based on a joint terrain feature matrix to obtain ground region blocks; Within the ground area block, the minimum value of the estimated bearing capacity corresponding to the ground unit is selected as the benchmark value of the regional bearing capacity. Extract the local pressure peak and local pressure mean within the ground area block; The proportion of edge units is obtained by calculating the ratio of the number of ground units corresponding to the preset edge regions within the ground area block to the total number of ground units within the ground area block. The edge unit proportion is compared with multiple preset proportion ranges to determine the edge proportion range; The pressure concentration factor is determined based on the preset coefficient value rules corresponding to the edge proportion interval; Using the pressure concentration factor and 1 minus the pressure concentration factor as weights, the local pressure peak and local pressure mean are weighted and summed to obtain the actual load value of the region; The bearing margin is calculated based on the ratio of the difference between the regional bearing capacity benchmark value and the actual regional load value to the regional bearing capacity benchmark value.

6. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 5, characterized in that, Methods for dynamically merging ground units based on a joint terrain feature matrix include: According to the arrangement order of ground units in the joint terrain feature matrix, the ground geometric information, ground bearing pressure information and bearing capacity estimate corresponding to adjacent ground units are extracted sequentially. Calculate the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates between adjacent ground units; The differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates are compared with their respective preset merging thresholds. When the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimation are all no greater than the corresponding merging threshold, adjacent ground units are merged into the same ground area block. When at least one of the differences in ground geometric information, ground bearing pressure information, and bearing capacity estimates is greater than the corresponding merging threshold, the corresponding adjacent ground units are divided into different ground area blocks.

7. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 6, characterized in that, Methods for calculating the probability of collapse include: The ratio of the actual load on the area to the benchmark value of the area's bearing capacity is calculated to obtain the compressive ratio. Calculate the ratio of the peak local pressure to the mean local pressure within a ground area block to obtain the pressure concentration. Calculate the ratio of the difference between the current regional carrying capacity benchmark value and the previous regional carrying capacity benchmark value to the previous regional carrying capacity benchmark value to obtain the carrying capacity attenuation degree; The edge influence degree is obtained by statistically analyzing the proportion of the number of ground units corresponding to the preset edge regions within the ground area block to the total number of ground units within the ground area block. The pressure ratio, pressure concentration, load attenuation, and edge influence are all adjusted to the same dimension. The collapse probability is obtained by weighting and summing the pressure ratio, pressure concentration, bearing capacity attenuation and edge influence after the adjustment based on the preset weighting coefficients.

8. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 7, characterized in that, Methods for generating terrain bearing risk maps with labeled risk levels include: Read the bearing capacity and collapse probability of each surface area block; The first risk level is determined by comparing the load margin with the preset load margin range. The collapse probability is compared with the preset collapse probability range to determine the second risk level; Compare the first risk level and the second risk level to determine the target risk level; Write the target risk level into the corresponding ground area block; Arrange the ground blocks with the target risk level written in them according to their ground location, and generate a terrain bearing risk map with the risk level marked.

9. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 1, characterized in that, Methods for calculating the adaptive obstacle-crossing control parameter set include: Read the bearing margin, subsidence probability, and risk level of the ground area blocks in the terrain bearing risk map; Read the ground geometric information, ground bearing pressure information and bearing capacity estimate corresponding to the ground unit in the joint terrain feature matrix; Based on the arrangement order of the ground area blocks, determine the main control area of ​​the ground area block to be entered; Based on the risk level, load margin, and collapse probability of the main control area, calculate the driving torque and travel speed; Based on the ground geometry information corresponding to the main control area, calculate the vehicle attitude adjustment amount; Based on the ground bearing pressure information and bearing capacity estimate of the main control area, calculate the ground pressure distribution of the wheel set and track; Based on the risk level, bearing capacity, collapse probability, and ground geometry information corresponding to the main control area, determine the obstacle crossing method; The driving torque, travel speed, vehicle attitude adjustment, wheel and track ground pressure distribution, and obstacle crossing method are used as the adaptive obstacle crossing control parameter set.

10. The adaptive obstacle-crossing control method for firefighting robots based on terrain feature learning according to claim 1, characterized in that, Methods for planning and optimizing obstacle-crossing paths based on adaptive obstacle-crossing control parameter sets include: Based on the obstacle-crossing methods corresponding to the ground area blocks, the paths of adjacent ground area blocks are combined to form a set of candidate paths; Based on the risk level of the ground area blocks traversed by the candidate path, the path risk value is calculated cumulatively. Based on the local pressure peak values ​​corresponding to the ground area blocks traversed by the candidate path, the path pressure value is calculated cumulatively. Based on the difference in vehicle attitude adjustment between adjacent ground area blocks in the candidate path, the path attitude change value is calculated cumulatively. The path length is calculated cumulatively based on the spatial distance between adjacent ground area blocks in the candidate path. The path risk value, path stress value, path attitude change value, and path length value are processed to be of the same dimension and then weighted and summed according to the preset path weight coefficient to obtain the path evaluation result. Compare the path evaluation results of each candidate path, and select the candidate path with the smallest path evaluation result as the optimized obstacle-crossing path.

Images