A bionic gait humanoid robot intelligent inspection method, device and equipment

Abstract

The application provides a bionic gait humanoid robot intelligent inspection method, device and equipment, task point distribution data is acquired, a necessary node sequence is determined based on coverage requirements, and an inspection state mapping table is established; a path probability distribution tree is constructed by using probability backtracking, a main path family and an alternative path family are identified, and a probability inspection chain is formed; the inspection chain is decomposed into continuous road sections, a traffic difficulty coefficient and a path complexity index are extracted, gait energy consumption distribution is predicted, an obstacle avoidance beat is determined, and path compensation parameters are generated; risk assessment is performed on the road sections, a risk balance scheme is formed by hedging combination of high-risk paths and safe paths, coordination rules are formulated; an inspection efficiency coordination factor is acquired based on path intersection overlap analysis, beat navigation parameters are generated; dynamic inspection units are generated by matching the beat parameters and the probability inspection chain, task arrival timing is adjusted, an elastic inspection process is constructed, and intelligent autonomous inspection under bionic gait control is realized.

Description

Technical Field

[0001] This invention relates to the field of robot path planning and control technology, and in particular to a biomimetic gait humanoid robot intelligent inspection method, device and equipment. Background Technology

[0002] With the continuous improvement of industrial automation, the demand for inspection in large factories, substations, data centers, and other scenarios is increasing. Humanoid robots, with their human-like morphology and mobility, can adapt to work environments designed for humans, demonstrating unique advantages in inspection tasks across complex terrains. However, existing humanoid robot inspection systems generally suffer from rigid path planning, limited gait control, and poor environmental adaptability, making it difficult to cope with dynamic changes and unexpected situations in real-world inspection scenarios.

[0003] Current technical solutions mostly adopt a pre-set fixed route inspection mode, lacking comprehensive consideration of multi-dimensional factors such as task priority, environmental risks, and energy consumption. This results in low inspection efficiency, incomplete coverage, and slow emergency response. In particular, in terms of gait control, traditional methods fail to fully utilize bionic principles, resulting in stiff robot movement, high energy consumption, and poor stability. Robots are prone to instability in complex terrains such as stairs, slopes, and narrow passages. Therefore, there is an urgent need to develop an autonomous inspection method that can achieve high efficiency, safety, and energy saving. Summary of the Invention

[0004] This invention provides a humanoid robot intelligent inspection method, device, and equipment with biomimetic gait. It aims to achieve intelligent allocation of task points and dynamic priority adjustment by constructing a multi-level inspection task planning system; adopt a probabilistic inspection chain and risk balancing mechanism to comprehensively consider inspection efficiency and safety; introduce a biomimetic gait control strategy to adaptively adjust the robot's motion mode according to terrain features and energy consumption distribution; and optimize multi-machine collaboration through beat navigation parameters to ultimately form a humanoid robot inspection scheme with environmental perception, intelligent decision-making, and flexible execution capabilities.

[0005] The first aspect of this invention proposes an intelligent inspection method for a humanoid robot with biomimetic gait, comprising the following steps:

[0006] Obtain task point distribution data of the inspection area, generate inspection coverage requirements based on the task point distribution data, determine the necessary node sequence according to the inspection coverage requirements, and establish an inspection status mapping table based on the necessary node sequence.

[0007] The inspection status mapping table is subjected to probabilistic backtracking processing to generate a path probability distribution tree. Branch analysis is performed on the path probability distribution tree to obtain the reachability probability value. Based on the reachability probability value, the main path family and the alternative path family are determined. A probabilistic inspection chain is generated based on the main path family and the alternative path family.

[0008] The probabilistic inspection chain is decomposed into continuous road segments. The passage difficulty coefficient and path complexity index are obtained from the continuous road segments. The gait energy consumption distribution is predicted based on the passage difficulty coefficient. The obstacle avoidance rhythm is determined based on the path complexity index. The path compensation parameters are generated by combining the gait energy consumption distribution and the obstacle avoidance rhythm. The path optimization configuration is generated using the path compensation parameters.

[0009] Risk assessment is performed on the continuous road segments to obtain traffic risk attributes. Based on the traffic risk attributes, high-risk paths and safe paths are determined. The high-risk paths and safe paths are combined to form risk-balanced path pairs. Path coordination rules are generated based on the risk-balanced path pairs.

[0010] Based on the path coordination rules and the path optimization configuration, the timing of integrated inspection is determined. The task coverage analysis is performed through the integrated inspection timing to determine the path intersection and overlap area. The inspection efficiency coordination factor is obtained from the path intersection and overlap area. The beat navigation parameters are generated based on the inspection efficiency coordination factor.

[0011] The rhythmic navigation parameters are matched with the probabilistic inspection chain to generate a dynamic inspection unit. The arrival time of the task point is adjusted based on the dynamic inspection unit, and an elastic inspection process is generated according to the arrival time to complete the intelligent inspection of the humanoid robot with biomimetic gait.

[0012] A second aspect of this invention provides a humanoid robot intelligent inspection device with biomimetic gait, comprising:

[0013] The task planning module is used to acquire task point distribution data of the inspection area, generate inspection coverage requirements based on the task point distribution data, determine the necessary node sequence according to the inspection coverage requirements, and establish an inspection status mapping table based on the necessary node sequence.

[0014] The path generation module is used to perform probabilistic backtracking processing on the inspection status mapping table to generate a path probability distribution tree, perform branch analysis on the path probability distribution tree to obtain reachability probability values, determine the main path family and alternative path family based on the reachability probability values, and generate a probabilistic inspection chain based on the main path family and alternative path family.

[0015] The path optimization module is used to decompose the probabilistic inspection chain into continuous road segments, obtain the passage difficulty coefficient and path complexity index from the continuous road segments, predict the gait energy consumption distribution based on the passage difficulty coefficient, determine the obstacle avoidance rhythm based on the path complexity index, generate path compensation parameters by combining the gait energy consumption distribution and the obstacle avoidance rhythm, and generate path optimization configuration using the path compensation parameters.

[0016] The risk balancing module is used to perform risk assessment on the continuous road segments to obtain traffic risk attributes, determine high-risk paths and safe paths based on the traffic risk attributes, combine the high-risk paths and the safe paths to form risk balancing path pairs, and generate path coordination rules based on the risk balancing path pairs.

[0017] The timing coordination module is used to determine the timing of integrated inspection based on the path coordination rules and the path optimization configuration, perform task coverage analysis through the integrated inspection timing to determine the path intersection and overlap area, obtain the inspection efficiency coordination factor from the path intersection and overlap area, and generate the beat navigation parameters based on the inspection efficiency coordination factor.

[0018] The dynamic execution module is used to match the beat navigation parameters with the probability inspection chain to generate dynamic inspection units, adjust the arrival time of task points based on the dynamic inspection units, generate an elastic inspection process according to the arrival time, and complete the intelligent inspection of the humanoid robot with bionic gait.

[0019] A third aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a biomimetic gait humanoid robot intelligent inspection method disclosed in the first aspect.

[0020] The beneficial effects of this invention are reflected in the following points: First, by using probabilistic inspection chains and dynamic path planning technology, it overcomes the limitations of traditional fixed routes, enabling the robot to autonomously select the optimal path based on real-time environmental changes and task priorities. This achieves full coverage and efficient execution of inspection tasks, improving the flexibility and adaptability of inspections. Second, by introducing biomimetic gait control and energy consumption prediction mechanisms, the robot can adaptively switch gait modes according to different terrain features, significantly reducing energy consumption while ensuring motion stability, thus solving the key technical challenges of humanoid robot motion control in complex environments. Third, a comprehensive risk assessment and multi-machine collaboration mechanism is established. Through risk-balanced path pairs and rhythmic navigation control, an optimal balance between safety and efficiency is achieved, effectively avoiding potential risks and resource conflicts during the inspection process, providing reliable assurance for intelligent inspection in industrial scenarios.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0022] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0023] Unless otherwise specified or defined, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0024] Figure 1 This is a flowchart illustrating an intelligent inspection method for a humanoid robot with biomimetic gait according to the present invention.

[0025] Figure 2 This is a structural block diagram of a humanoid robot intelligent inspection device with biomimetic gait according to the present invention.

[0026] Figure 3 This is a schematic diagram of the structure of a computer device according to the present invention. Detailed Implementation

[0027] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0028] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0029] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0030] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0031] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0032] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0033] The technical solutions of the embodiments of this application will be described below.

[0034] like Figure 1 As shown, this embodiment of the invention provides a biomimetic gait humanoid robot intelligent inspection method, including the following steps S110-S160:

[0035] Step S110: Obtain task point distribution data of the inspection area, generate inspection coverage requirements based on task point distribution data, determine the sequence of necessary nodes based on inspection coverage requirements, and establish an inspection status mapping table based on the sequence of necessary nodes.

[0036] Specifically, a three-dimensional rasterized representation is used, dividing the space into a three-dimensional grid, with each grid cell measuring 0.5m × 0.5m × 0.5m. The task point distribution data contains four core types of information: location coordinates P_i = (x_i, y_i, z_i), representing the three-dimensional spatial position of the i-th task point; task attributes including task type, priority (levels 1-5), and estimated execution time; environmental features recording obstacle density, terrain complexity, and lighting conditions; and time constraints specifying the time window for task execution. Data acquisition utilizes LiDAR to obtain point cloud data, visual cameras to capture RGB-D images, and IMU sensors to record terrain features. Task point clustering analysis identifies multiple task clusters, each representing a functional area. The data structure uses a hash table for storage, with grid coordinates as the key. When generating coverage requirements, the hash table data is reorganized into a matrix format for batch calculation, ultimately forming task point distribution data containing four dimensions: location, attributes, environment, and time.

[0037] Inspection coverage requirements are generated based on task point distribution data. The influence range of each task point is calculated using a Gaussian kernel function R_i(d) = exp(-d). 2 / 2σ_i 2), where R_i(d) is the influence intensity of task point i at a distance d, d is the Euclidean distance from the task point, and σ_i is the influence radius, dynamically adjusted according to task priority: 5 meters for level 1, 4 meters for level 2, 3 meters for level 3, 2 meters for level 4, and 1 meter for level 5, with priority weights of 5, 4, 3, 2, and 1 respectively. The influence ranges of all task points are superimposed to form a coverage intensity field F(x,y,z)=Σ(priority_i×R_i), where F(x,y,z) is the coverage intensity value at spatial location (x,y,z). The time dimension expansion converts the static coverage field into spatiotemporal coverage requirements, maintaining the original intensity value within the task time window and attenuating to zero outside the window. Environmental constraint fusion obtains the regional obstacle density by taking the average obstacle density in the 3×3 grid neighborhood around the task point, and corrects the coverage intensity by combining it with terrain complexity. Coverage levels are classified based on field strength thresholds: areas with an intensity greater than 0.8 are core coverage areas, 0.5-0.8 are important coverage areas, 0.2-0.5 are general coverage areas, and areas below 0.2 are optional coverage areas. The generated coverage requirement matrix uses spatial grids as rows and time slices as columns, with matrix elements recording the coverage level of each spatiotemporal point, thus realizing the transformation from discrete task points to a continuous coverage field.

[0038] Based on the coverage level distribution and gradient information of the coverage intensity field in the coverage demand matrix, a graph theory algorithm is used to determine the sequence of necessary nodes. Node selection involves scanning the coverage demand matrix spatially, taking the maximum coverage level for each grid across all time slices, identifying core coverage area grids with a maximum value greater than 0.8, and aggregating consecutive high-level grids into blocks. The geometric center of each block is used as a candidate necessary node. Coverage contribution is calculated using a 3×3 neighborhood window, summing the coverage levels around each candidate point, and simultaneously assessing its importance based on the gradient magnitude of the coverage intensity field at that point. When constructing the node connectivity graph, the gradient direction is extracted from the coverage intensity field, and the angle between the path and the gradient between adjacent nodes is calculated. The edge weight w_ij = d_ij × (1 + 0.2 × terrain_complexity) × (1 - 0.1 × gradient_alignment), where w_ij is the path cost from node i to node j, d_ij is the Euclidean distance between nodes, terrain_complexity is the terrain complexity (value 0-1), gradient_alignment is the gradient alignment degree (value 0-1), and the coefficients 0.2 and 0.1 are the terrain influence factor and gradient correction factor, respectively. The Traveling Salesman Algorithm (TSA) is used to solve the traversal sequence, converting the time slice information in the matrix into time window constraints for each node. The time slice with the highest coverage level for that spatial location is taken as the preferred visit time, and the time window is extended by one time slice before and after. The output sequence of required nodes includes the actual spatial coordinates, estimated arrival time, suggested stay duration (3-5 minutes in the core area, 2-3 minutes in the important area), and turning angle.

[0039] Based on the node coordinates, arrival time, dwell time, and turning angle in the necessary node sequence, an inspection state mapping table reflecting the robot's motion state is constructed. State definitions directly assign node coordinates to the position component and turning angles to the heading angle in the attitude angles. Arrival time is used to calculate the speed constraints between adjacent nodes; the maximum allowable speed is obtained by dividing the distance between nodes by the available time. Actions are categorized based on dwell time: less than 1 minute for rapid passage, 1-3 minutes for routine scanning, and more than 3 minutes for detailed inspection. The state transition probability matrix is ​​constructed considering time feasibility: if the current time plus travel time and dwell time exceeds the target node's specified arrival time, the transition probability is set to zero; otherwise, the probability value equals the base success rate multiplied by the time margin coefficient. The base success rate is 0.95, and the time margin coefficient = remaining time / required time, where the remaining time is the target arrival time minus the current time, and the required time is the travel time plus the dwell time. The path planning module of the mapping table decomposes the turning angle of the node sequence into a series of small-angle turns, each not exceeding 30 degrees. The estimated energy consumption includes dwell time and movement energy consumption, with the scanning power set at 50W. The sensor configuration automatically switches scanning modes based on dwell time. Each element of the mapping table matrix fully records the state transition scheme, including intermediate waypoints, time budget, action sequence, and energy requirements, forming a complete inspection state mapping table.

[0040] Step S120: Perform probabilistic backtracking processing on the inspection status mapping table to generate a path probability distribution tree, perform branch analysis on the path probability distribution tree to obtain reachability probability values, determine the main path family and alternative path family based on the reachability probability values, and generate a probabilistic inspection chain based on the main path family and alternative path family.

[0041] Specifically, the last node in the necessary node sequence is set as the inspection endpoint, serving as the starting point for probabilistic backtracking. Using the state transition probability matrix and mapping elements in the inspection state mapping table, a probability distribution tree for path evolution is constructed using a probabilistic backtracking algorithm. Backtracking begins at the target endpoint and traces back all possible paths leading to that point. The probability calculation for each path comprehensively considers all elements recorded in the mapping table. The intermediate waypoint sequence is used to refine the path trajectory. The waypoint coordinates stored in the mapping table are concatenated to form a complete trajectory from the starting point to the endpoint, with each waypoint marked with its state information upon passage. Time budget directly affects probability calculation. The time budget for each path segment is extracted from the mapping table, accumulated to obtain the total time consumption, and compared with the task time window. A penalty coefficient is applied to the probability of paths exceeding the time limit. The action sequence determines the node type. The probability of quickly passing nodes remains unchanged, while corresponding probability adjustment coefficients are applied to regular scan nodes and detailed inspection nodes, reflecting the time uncertainty of different action types. Energy demand is used for feasibility assessment. A baseline battery capacity is determined based on the projected energy consumption calculated in S110 and the robot's standard operating time. The energy consumption for movement and dwell time across all segments of the path is accumulated. If the total demand exceeds a high percentage threshold of the battery capacity baseline, the probability of that path is significantly reduced, and it is only considered as an emergency backup. During tree structure construction, the root node is the inspection endpoint. Each node stores the cumulative probability, cumulative time, cumulative energy consumption, and executed action sequence for reaching that location. Branch generation is based on the connection relationships in the mapping table; each valid preceding state forms a child node. After backtracking is completed, a path probability distribution tree containing all feasible paths and their probability attributes is formed.

[0042] In some embodiments, the step of performing branch analysis on the path probability distribution tree to obtain the reachability probability value includes: identifying the main branch and secondary branches based on the path probability distribution tree; performing probability decay analysis on the main branch to obtain a decay coefficient; correcting the original probabilities of the main branch and the secondary branches according to the decay coefficient; and determining a threshold based on the corrected probabilities to obtain the reachability probability value.

[0043] Based on the structural features and probability distribution of the path probability distribution tree, branches of different importance levels are identified. Main branches are identified through depth-first traversal, calculating the path probability product from leaf nodes to the root node. Paths with probability products exceeding a preset threshold are marked as main branches. These main branches typically have three characteristics: fewer nodes traversed (shorter paths), fewer detailed inspection actions (faster execution), and reasonable cumulative energy consumption. Secondary branches are paths with probability products in the moderate range, often containing more intermediate waypoints or requiring more scanning actions. Branch importance also considers their uniqueness; if a branch traverses a critical area not covered by other paths, it will be marked as an important branch even with a slightly lower probability. The ratio of main branches to secondary branches is determined based on the statistical characteristics of the path probability distribution.

[0044] Probability decay analysis is performed on the identified main branches to quantify the impact of path complexity on execution reliability. The decay analysis considers three main factors: path length decay (each additional node reduces the probability); action complexity decay (detailed inspection actions and routine scans correspond to different probability decays); and time urgency decay (additional decay is applied when the path usage time approaches the budget limit). The decay coefficient is calculated using an exponential decay model α = base_factor^L × action_factor^D × time decay factor, where L is the number of path nodes and D is the number of detailed inspection actions, with the exponential form reflecting the cumulative impact of complexity on reliability. For paths requiring frequent turns, an additional turning decay coefficient is applied. Energy constraints also affect decay; if the path energy consumption approaches the battery capacity, the decay coefficient is further reduced. Each decay parameter is determined based on the task type and robot performance characteristics, ultimately resulting in a specific decay coefficient for each main branch.

[0045] Based on the calculated attenuation coefficients, the original probabilities of the main branch and secondary branches are corrected. The probability correction for the main branch uses its corresponding attenuation coefficient: P_modified_main = P_original × α, where P_modified_main and P_original are the corrected and original probabilities of the main branch, respectively, reflecting the decrease in reliability during actual execution. The correction for secondary branches considers their correlation with the main branch. If a secondary branch shares more than a set proportion of nodes with a main branch, the attenuation coefficient of that main branch is used; otherwise, a more conservative attenuation is used: P_modified_secondary = P_original × α × additional_factor, where P_modified_secondary is the corrected probability of the secondary branch. For branches containing alternative action sequences, probability compensation is given based on the number of alternative schemes. The correction process maintains the relative magnitude of probabilities; that is, branches with high original probabilities retain high probabilities after correction. After all branches have been corrected, the probability distribution is recalculated to ensure the rationality of the physical meaning.

[0046] Based on the corrected probabilities, a multi-level threshold determination is performed to identify the reachability level and specific probability value of each path. A four-level standard is used for threshold setting, with specific thresholds determined according to task reliability requirements. Paths above the high threshold are considered highly reachable, with the highest expected success rate and can be used as the primary execution plan; paths within the medium range are considered moderately reachable and suitable as alternatives; paths within the lower range are considered low reachable and are only considered in special cases; paths below the minimum threshold are considered unreachable and are excluded from subsequent processing. The final determination of the reachability probability value also considers the path coverage completeness. If a low-probability path is the only option to reach a specific critical area, its reachability probability value will be increased to ensure that no critical area is missed. Each path outputs its reachability probability value (a specific numerical value between 0 and 1) and reachability level label.

[0047] Based on the reachability probability values ​​and reachability level labels obtained from branch analysis, a clustering algorithm is used to divide the path into main path families and candidate path families. The main path families are selected from highly reachable paths. First, the spatial similarity between paths is calculated, and the proportion of shared nodes between two paths is counted. Paths with similarity exceeding a set threshold are aggregated into the same family. Each main path family retains several representative paths that are highly similar macroscopically (passing through the same key areas) but differ in local details (such as specific intermediate waypoint selections and different choices at branch points). The complementarity of paths within a family is evaluated by calculating the union of coverage areas, requiring the union coverage rate to be significantly higher than that of a single path. Candidate path families are constructed from moderately reachable paths, focusing on paths whose spatial distribution is complementary to that of the main path families. The minimum spatial distance from candidate paths to all main path families is calculated, and paths with distances exceeding a set threshold are selected to construct candidate families, ensuring completely different alternatives when the main path is blocked. Special function path families include the shortest time family (prioritizing fast passage actions), the lowest energy consumption family (avoiding climbing and frequent acceleration / deceleration), and the maximum coverage family (passing through all available coverage areas). Ultimately, this results in a structured organization with multiple main path families and alternative path families.

[0048] In some embodiments, generating a probabilistic inspection chain based on the main path family and the alternative path families includes: constructing a primary propagation skeleton based on the main path family; performing extended analysis on the primary propagation skeleton to obtain extended nodes; performing multi-level supplementation on the extended nodes according to the alternative path families to generate an enhanced path network; and performing probability assignment processing on the enhanced path network to generate a probabilistic inspection chain.

[0049] The main framework of the inspection network is constructed based on the core paths of the main path families. The first-level propagation skeleton extracts the path with the highest reachability probability from each main path family as the family representative, and incorporates all nodes on these representative paths into the skeleton node set. The skeleton nodes include all mandatory nodes determined in S110, as well as important relay nodes on the main paths (locations with long dwell times or requiring detailed inspection). The connections between nodes retain the state transition information of the original path, including transition probabilities, estimated times, and recommended actions. The spatial distribution of the skeleton is determined according to the inspection density requirements, with corresponding node densities set in key and transition areas. After the skeleton is constructed, a basic network structure covering all core inspection tasks is formed.

[0050] Structural analysis is performed on the primary propagation backbone to identify weak links requiring reinforcement. The expansion analysis focuses on three types of problems: spatially sparse segments (regions where the distance between adjacent backbone nodes exceeds a set threshold, lacking intermediate monitoring points); connectionally weak points (nodes connected by only a single path, becoming inaccessible if the path is blocked); and coverage blind spots (task points beyond the coverage radius of the nearest backbone node, potentially missed). Spatial analysis algorithms identify a certain proportion of backbone nodes with these problems, marking these locations as expansion nodes. Expansion nodes also include key bifurcation points of the backbone path, suitable for inserting alternative paths to increase network redundancy. Each expansion node records its problem type, distance from the backbone, and suggested supplementary directions.

[0051] Based on the path characteristics of the candidate path families and the need for additional nodes, multi-level path supplementation is implemented. Level 1 supplementation targets sparse segments, extracting path fragments from the candidate path families that pass near these sparse segments and using these fragments as the backbone for parallel selection and integration into the network. Level 2 supplementation addresses connection fragility by finding at least one independent reachable path from the candidate path families for each vulnerable node, ensuring double protection. Level 3 supplementation covers blind spots by inserting neighboring nodes from the candidate path families at the blind spot locations, integrating them into the network through short-distance connections. The supplementation process maintains the integrity of the original paths and does not disrupt existing action sequences and timing arrangements. After all levels of supplementation are completed, the total number of network nodes reaches a preset multiple of the backbone node count.

[0052] The enhanced path network is probabilistically assigned to form a complete inspection chain with probabilistic attributes. Node probability assignment follows an additive principle: if a node belongs only to the skeleton, its original reachability probability is maintained; if a node is traversed by multiple paths, the sum of all path probabilities is taken (with an upper limit set to avoid exceeding the range). Edge probability assignment considers path origin and supplementation level; skeleton edges retain their original transition probabilities, while supplementation edges at each level are subject to corresponding reliability decay coefficients, reflecting decreasing reliability. Probability adjustments for special nodes include forcibly setting the arrival probability of mandatory nodes to the highest value (must be reached), and appropriately increasing the probability of extended nodes based on their supplementation effect. Probability propagation verification ensures that the sum of all path probabilities from the starting point to any reachable node is greater than a set threshold. The final generated probabilistic inspection chain contains network nodes with approximately a preset multiple of the number of mandatory nodes. Each node is labeled with its arrival probability, suggested path, alternative solutions, and expected dwell time; each edge is labeled with its transition probability, estimated time, and recommended action, ultimately forming a complete probabilistic inspection chain.

[0053] Step S130: Decompose the probabilistic inspection chain into continuous road segments, obtain the passage difficulty coefficient and path complexity index from the continuous road segments, predict the gait energy consumption distribution based on the passage difficulty coefficient, determine the obstacle avoidance rhythm based on the path complexity index, generate path compensation parameters by combining the gait energy consumption distribution and obstacle avoidance rhythm, and generate path optimization configuration using the path compensation parameters.

[0054] Specifically, inheriting the action type classification defined in S110, the recommended actions in the probabilistic inspection chain include three categories: quick passage, routine scan, and detailed inspection. Segment decomposition is based on changes in node action type; when the recommended action of an adjacent node changes from "quick passage" to "routine scan" or "detailed inspection," a segment boundary is formed. Each continuous segment contains a starting node, an ending node, a sequence of intermediate nodes, and a corresponding set of edges. Segment attributes are inherited from the probabilistic inspection chain; the segment probability is the product of the transition probabilities of all its included edges; the segment duration is the sum of the expected time consumption of each edge; and the segment action sequence is the concatenation of the recommended actions of each node. The physical characteristics of a segment are calculated using the coordinates of its included nodes and environmental characteristics. The segment length is the sum of the Euclidean distances between nodes; height changes record the maximum ascent and descent values; and the number of turns counts the number of nodes whose direction changes exceed a set angle. Time characteristics are extracted from the expected dwell time of nodes; movement time is the pure travel time; and operation time is the sum of the dwell times of each node. Road segment types are categorized based on the dominant action: moving road segments are mostly for quick passage, scanning road segments contain multiple regular scans, and inspection road segments contain detailed inspections. Typical decomposition results in several consecutive road segments, each with a length determined by task density, fully covering all nodes and edges of the probabilistic inspection chain.

[0055] The passage difficulty coefficient and path complexity index are obtained from continuous road segments. The passage difficulty coefficient D_i comprehensively evaluates the physical challenge of the i-th road segment, and is calculated by the formula D_i = w1 × slope_factor + w2 × obstacle_density, where D_i is the passage difficulty coefficient of the i-th road segment, w1 and w2 are the weight coefficients of slope and obstacles, respectively, slope_factor is the slope factor (calculated based on the changes in road segment height, with ascent and descent values ​​converted into equivalent slopes), and obstacle_density is the obstacle density (extracted from the environmental characteristics of the area traversed by the road segment). The weight coefficients are determined based on the robot's performance characteristics and terrain adaptability, reflecting the relative importance of each factor. The path complexity metric C_i quantifies the navigation difficulty of a road segment, where C_i is the path complexity metric for the i-th road segment, comprising three dimensions: geometric complexity (equal to the number of turns divided by the segment length and then multiplied by a standardization coefficient, reflecting the frequency of directional changes per unit distance); spatial complexity (equal to the proportion of narrow passage length to the total length, calculated based on passage width information in the road segment's environmental characteristics); and action complexity (equal to the number of action switches multiplied by the switch type weight, statistically derived from the action sequence of the road segment). The final path complexity is obtained through weighted summation, with weights allocated according to the performance characteristics of the navigation system. The numerical ranges of the traffic difficulty coefficient and the path complexity metric are determined based on terrain conditions and task complexity; these two metrics comprehensively characterize the traffic characteristics of each road segment.

[0056] Gait energy consumption distribution is predicted based on the difficulty coefficient. The base energy consumption rate E_base = base coefficient × Di, where E_base is the base energy consumption rate (watts / meter). The base coefficient reflects the robot's power characteristics under standard terrain, representing the energy consumption per unit distance at that difficulty level. The gait adjustment factor is determined based on the terrain component in the difficulty coefficient. When Di_i is close to the baseline value (flat terrain dominant), a standard gait is used; when Di_i is in a medium range (slope dominant), a small-step, high-frequency gait is used; and when Di_i exceeds a high threshold (staircase dominant), a large-step, climbing gait is used. Each gait corresponds to a different energy consumption adjustment factor, reflecting the impact of different terrains on energy consumption. Instantaneous power P(t) = E_base × v(t) × gait adjustment factor, where P(t) is the instantaneous power (watts) at time t, and v(t) is the instantaneous speed (m / s) at time t. The speed setting is determined based on D_i, and is determined according to the robot's safety performance and terrain adaptability. Normal speed is used for low difficulty, speed is appropriately reduced for medium difficulty, and speed is significantly reduced for high difficulty. Energy consumption distribution prediction considers acceleration and deceleration phases. The power during the start-up phase is the increase factor of the steady-state power, and the power during the deceleration phase is the decrease factor of the steady-state power. Total energy consumption of a road segment E_segment = ∫P(t)dt, where E_segment is the total energy consumption of the road segment (joules), and the integration interval is the travel time of the road segment. The energy consumption peak occurs at the point of abrupt change in the difficulty coefficient, and the peak power can be several times the average power. Gait switching energy consumption is related to the change in difficulty; additional energy consumption is generated when the change in D_i exceeds the set threshold. The complete gait energy consumption distribution includes the power value P(t) of the time series and the cumulative energy consumption curve E(t).

[0057] The obstacle avoidance cycle time is determined based on path complexity. The obstacle avoidance cycle time T_cycle = base cycle × (1 + C_i), where T_cycle is the obstacle avoidance cycle time (milliseconds), and the base cycle is the system's baseline time interval. Higher complexity requires more frequent environmental perception and decision-making. The sensor scanning cycle time is adjusted according to spatial complexity; the LiDAR scanning frequency is increased in narrow passages and reduced to a normal frequency in open areas. The path planning update cycle time is related to geometric complexity; local paths are updated frequently on high-turning-frequency road sections, while the update interval can be extended on straight road sections. The decision cycle time follows a hierarchical architecture, with the lowest layer having the shortest obstacle avoidance decision cycle, the middle layer having a moderate path tracking cycle, and the highest layer having the longest task planning cycle. A cycle time synchronization mechanism ensures coordinated decision-making across layers; the lowest layer cycle time is the baseline, the middle layer cycle time is an integer multiple of the lowest layer cycle, and the highest layer cycle time is an integer multiple of the middle layer cycle. Special case cycle time adjustments include temporarily shortening the obstacle avoidance cycle time when dynamic obstacles are detected; and further refining the control cycle time when performing precise docking. The beat parameter set includes {T_avoid, T_scan, T_plan, T_control}, which correspond to the obstacle avoidance cycle, perception cycle, planning cycle, and control cycle (all in milliseconds), respectively. These parameters are dynamically configured according to the path complexity.

[0058] In some embodiments, generating path compensation parameters by combining the gait energy consumption distribution and the obstacle avoidance beat includes: performing energy density calculation on the gait energy consumption distribution to obtain an energy concentration distribution; identifying key energy consumption regions from the energy concentration distribution; performing temporal fusion of the key energy consumption regions and the obstacle avoidance beat to obtain compensation timing features; and constructing path compensation parameters based on the compensation timing features.

[0059] Energy density distribution is obtained by calculating energy density from gait energy consumption distribution. The energy density calculation uses a sliding window method: ρ_e(s) = ΔE / Δs, where ρ_e(s) is the energy density at position s (joules / meter), ΔE is the energy consumption within the sliding window (joules), and Δs is the corresponding distance traveled (meters). The density distribution exhibits obvious clustering characteristics, with significantly higher energy density in terrain-changing areas and areas with dense turning than in straight sections. Energy concentration is defined as the ratio of local density to average density: γ(s) = ρ_e(s) / ρ_avg, where γ(s) is the energy concentration at position s (dimensionless), and ρ_avg is the average energy density of the entire road segment (joules / meter). The concentration distribution curve identifies multiple peak regions, with peak positions corresponding to points of high-energy-consuming actions. Spatial smoothing employs a filtering algorithm to eliminate instantaneous fluctuations and preserve the main energy concentration trends. The time dimension extension maps spatial energy density to the time axis, forming a spatiotemporal energy density distribution ρ_e(s,t), which fully describes the spatiotemporal evolution of energy consumption.

[0060] Key energy-consuming regions are identified from the energy concentration distribution. The threshold for identifying key regions is determined based on the statistical characteristics of the energy distribution, i.e., regions where the energy density exceeds a set multiple of the average. The identification algorithm uses connected component analysis to mark all points exceeding the threshold and connects neighboring points through morphological processing to form continuous key regions. Each key region records its spatial extent, peak concentration, and total energy consumption. Region classification is based on cause: terrain-induced types are mainly caused by slope, action-induced types are generated by scanning or inspection actions, and composite types involve the superposition of terrain and action. Typical identification results contain several key regions, accounting for a certain proportion of the total distance but consuming the majority of the total energy. The time span of the regions is converted using velocity curves to form a time-domain representation.

[0061] The energy consumption critical area is fused with the obstacle avoidance cycle to obtain compensation timing characteristics. The fusion algorithm calculates the phase difference between the start time of the critical area and the boundary of the nearest cycle: φ=(t_startmodT_avoid) / T_avoid, where φ is the phase difference (dimensionless, between 0 and 1), t_start is the start time of the critical area (milliseconds), T_avoid is the obstacle avoidance cycle period (milliseconds), mod is the modulo operation, and the phase difference reflects the cycle state when entering the high energy consumption area. The compensation timing feature extraction includes: lead time Δt_lead = T_plan - φ × T_plan, where Δt_lead is the compensation lead time (milliseconds) and T_plan is the path planning cycle (milliseconds), ensuring path adjustment is completed before entering the critical area; lag time Δt_lag = T_control, where Δt_lag is the compensation lag (milliseconds), reserving control response time; and duration T_duration = t_end - t_start + Δt_lead + Δt_lag, where T_duration is the compensation duration (milliseconds) and t_end is the end time of the critical area (milliseconds), covering the complete high-energy-consuming process. The timing features also consider cycle time switching; if the critical area crosses an action switching point, the compensation timing needs to include the transition period for cycle time parameter switching. For multi-area coordination, when the interval between critical areas is less than a set threshold, their compensation timings are merged to avoid frequent adjustments. The compensation timing feature vector extension includes energy consumption intensity information, with peak concentration and total regional energy consumption as additional attributes of the timing feature, forming a complete feature vector {Δt_lead, T_duration, beat adjustment scheme, γ_max, E_region}, where γ_max is the regional peak concentration (dimensionless), E_region is the total regional energy consumption (joules), γ_max is used for subsequent velocity adjustment, and E_region is used for energy reservation calculation.

[0062] Path compensation parameters are constructed based on the timing characteristics of compensation. Deceleration begins well before the critical region, with the deceleration rate calculated based on the speed difference and available time. The target speed is based on peak concentration to ensure that the speed has dropped to an appropriate level before entering the high-energy-consumption zone. A high-efficiency gait is switched to in advance, with the switching timing being the region's start time minus the time required for gait switching. A minimum energy threshold is set for entering the critical region, determined based on the region's predicted energy consumption and safety margin. In the high-energy-consumption region, the original path points are smoothed, with the smoothing degree inversely proportional to the energy concentration. The sensor sampling rate within the critical region is dynamically increased based on the concentration. PID controller parameters are adjusted according to the region's characteristics: position loop bandwidth is adjusted to improve stability, and the speed loop integral term is adjusted to improve tracking performance.

[0063] Path optimization configurations are generated using path compensation parameters. Configuration generation is performed segment by segment, with each segment selecting a subset of compensation parameters based on the number and type of critical areas it contains. Velocity curve reconstruction replaces the original uniform speed assumption with a variable speed curve; critical areas use compensated low speeds, transition zones use gradually changing speeds, and non-critical areas maintain normal speeds. Gait sequence orchestration forms a complete gait switching schedule, including switching times, target gait, and transition durations. Energy management strategies set energy budgets for each segment, allocating a safety multiple of predicted energy consumption to critical areas to ensure sufficiency, while budgets can be appropriately compressed in non-critical areas to improve overall efficiency. Sensor configuration schedules are synchronized with segment configurations, activating high-precision mode before entering complex areas and reverting to normal mode upon exiting. Control parameter scheduling tables provide segmented PID parameters, updating controller configurations at the entrance of each segment. The final path optimization configuration is output in structured data format, including parameter sequences with time indices, which can be directly loaded into the robot control system to achieve prediction- and compensation-based optimization execution.

[0064] Step S140: Conduct a risk assessment on continuous road segments to obtain traffic risk attributes, determine high-risk paths and safe paths based on traffic risk attributes, combine high-risk paths and safe paths to form risk-balanced path pairs, and generate path coordination rules based on risk-balanced path pairs.

[0065] Specifically, risk assessment is based on measurable features of continuous road segments to obtain the traffic risk attributes of each segment. The assessment starts from the inherent characteristics of the road segments to identify potential factors affecting safe passage. Physical risks stem from the terrain features of the road segments; ascending sections pose an instability risk, descending sections pose a braking failure risk, and sections with dense turns pose a collision risk. Environmental risks are based on environmental features obtained in S110; areas with high obstacle density pose a collision risk, areas with poor lighting conditions pose a perception risk, and areas with high terrain complexity pose a navigation deviation risk. Execution risks are related to the type of action on the road segment; scanning sections require frequent stops, increasing the risk of positioning deviation, while inspection sections require long-term operation, leading to energy depletion risks. Risk quantification uses a scoring method based on road segment features, with the comprehensive risk value R_total = f(segment length, height change, number of turns, obstacle density, action complexity), and specific weights determined based on the robot's safety performance. Time-dimensional risks consider task deadline pressure; the closer to the deadline, the higher the risk due to rushed execution. The range of the comprehensive risk value R_total is determined based on the distribution of road segment features. The traffic risk attributes include the overall risk value, the main types of risk sources, the risk level classification, and recommended countermeasures, and a risk assessment file is established for each road segment.

[0066] In some embodiments, determining high-risk paths and safe paths based on the access risk attributes includes: performing risk gradient analysis based on the access risk attributes to obtain a gradient distribution; performing clustering processing on the gradient distribution to identify risk clustering areas; performing connectivity analysis on the risk clustering areas to generate high-risk paths; and performing low-value region search on the gradient distribution to connect the searched low-risk nodes to construct safe paths.

[0067] Risk gradient analysis is performed based on traffic risk attributes to obtain gradient distribution. Gradient analysis calculates the risk change rate between adjacent road segments: ∇R=(R_j-R_i) / d_ij, where ∇R is the risk gradient (risk value / meter), R_i and R_j are the risk values ​​of adjacent road segments, and d_ij is the distance between the center points of the road segments (meters). The gradient vector reflects the spatial trend of risk change; a positive gradient indicates the direction of risk increase, and a negative gradient indicates the direction of risk decrease. The gradient field is constructed by extending the risk of discrete road segments into a continuous distribution through interpolation algorithms, using radial basis functions to ensure a smooth transition. The gradient magnitude |∇R| identifies areas of drastic risk change, where |∇R| is the gradient magnitude (risk value / meter), and the threshold is determined based on the sensitivity to risk changes. Gradient direction analysis identifies risk propagation paths; gradients continuously pointing to high risks form risk channels that need to be avoided or have enhanced protection. The time dimension is extended to consider the dynamic characteristics of risk change. The gradient distribution map is presented in the form of a heatmap, intuitively reflecting the spatial distribution characteristics of risk.

[0068] Gradient distribution is clustered to identify risk clusters. The DBSCAN clustering algorithm is used, with gradient magnitude as a feature, and clustering parameters determined based on the spatial distribution characteristics of road segments. The clustering results form multiple risk clusters, each representing a potential location for concentrated risk outbreaks. Cluster features extracted include center location (centroid coordinates), coverage area (bounding box size), average risk level, and dominant risk type. Area classification is based on risk characteristics: sudden-onset clusters show a sudden increase in risk, gradual-onset clusters show a gradual accumulation of risk, and complex clusters exhibit multiple overlapping risks. Spatial correlation analysis calculates the distance and directional relationships between clusters to identify risk transmission chains.

[0069] Connectivity analysis is performed on risk clusters to generate high-risk paths. Connectivity analysis constructs a risk network graph, with cluster centers as nodes and risk transmission intensity as edge weights. Transmission intensity calculation considers spatial distance and risk correlation; closer distances and more similar risk types result in stronger transmission. Path search algorithms identify paths connecting multiple clusters, where the risk accumulation effect is significant. High-risk paths are defined as those passing through a set number of risk clusters or whose cumulative risk value exceeds a preset threshold. Path characteristics include total length, number of clusters traversed, cumulative risk value, and maximum single-point risk. Key node identification identifies the intersections of multiple high-risk paths; these locations are key areas for risk control. Path classification is determined based on risk level and safety requirements: Level 1 high-risk paths must be avoided, Level 2 high-risk paths require special protection, and Level 3 high-risk paths require enhanced monitoring.

[0070] A low-value region search is performed on the gradient distribution. Combined with the risk value in the traffic risk attribute, the searched low-risk nodes are connected to construct a safe path. A low-value region is defined by simultaneously meeting two conditions: the risk value R_total extracted from the traffic risk attribute is below a set threshold, and the gradient magnitude |∇R| extracted from the gradient distribution is below a gradient threshold, indicating low and stable risk. The search algorithm employs dual threshold segmentation. First, low-risk road segments are selected based on the risk value R_total, and then low-gradient regions are further selected from these segments. The safe node selection criteria comprehensively consider the proportion of risk values ​​below the average, the gradient magnitude indicating no risk abrupt changes, and the probability of successful passage of the road segment. Node connection uses the minimum risk path algorithm, prioritizing connections with both low risk and low gradient values ​​while ensuring connectivity. Safe path construction follows the principles of avoiding all risk clusters as much as possible, maintaining a safe distance from high-risk paths (distance determined based on the robot's obstacle avoidance capabilities), and prioritizing spacious and flat areas. Path optimization improves the continuity and accessibility of the path by adding a small number of medium-risk nodes. The safe path verification ensures that the risk value at any point does not exceed the safety threshold, the gradient value shows no risk mutation points, and emergency handling conditions are available.

[0071] High-risk paths are combined with safe paths to form risk-balanced path pairs. Based on the principle of risk complementarity, each high-risk path is configured with several safe paths as alternatives, forming a primary and backup path pair. The pairing principle considers spatial coverage; safe paths should be able to reach the critical task points of high-risk paths, ensuring that the path reachability based on the probabilistic inspection chain meets the requirements. The comprehensive risk after risk hedging is calculated as: R_combined = w_high × R_high + w_safe × R_safe, where R_combined is the comprehensive risk value of the path pair, R_high is the risk value of the high-risk path, R_safe is the risk value of the safe path, and w_high and w_safe are execution weights, representing the probability of using each path, and their sum is 1. The weights are allocated according to the execution probability and safety policy. Path switching points are set at locations with lower risk gradients to facilitate a smooth transition. Switching conditions include real-time risk monitoring values ​​exceeding preset thresholds, energy reserves falling below the requirements of safe paths, and the detection of sudden environmental changes such as the addition of obstacles.

[0072] Path coordination rules are generated based on risk balancing paths. The formulation of these rules follows the dual objectives of risk minimization and task assurance. Basic rules include executing the main path under normal circumstances and switching to a safe path when risk monitoring is triggered. Dynamic switching rules issue warnings when the current risk exceeds the warning threshold of the path's design risk, and force a switch when it exceeds the mandatory switching threshold, based on real-time risk assessment. Priority rules determine the selection order when multiple paths conflict, with the highest task criticality taking precedence, followed by the smallest risk increment, and the shortest path serving as a supplementary criterion. Collaboration rules handle multi-robot scenarios, restricting single-robot passage on high-risk paths while allowing multiple robots to travel in parallel on safe paths. Time coordination rules are based on the time-varying characteristics of risk, prioritizing safe paths during high-risk periods and selecting efficient paths during low-risk periods. Emergency rules define extreme situation handling, including the minimum-risk evacuation plan when all path risks exceed the threshold and autonomous decision-making logic when communication is interrupted. The rule update mechanism dynamically adjusts based on execution feedback; successful execution reduces the risk assessment of the path, while failed execution increases the risk weight and optimizes the switching threshold. The complete path coordination rule set comprises three levels: basic rules, dynamic rules, and emergency rules, ensuring reasonable decision-making in various scenarios.

[0073] Step S150: Determine the timing of integrated inspection based on path coordination rules and path optimization configuration; determine the path intersection and overlap area through task coverage analysis based on the integrated inspection timing; obtain the inspection efficiency coordination factor from the path intersection and overlap area; and generate the cycle navigation parameters based on the inspection efficiency coordination factor.

[0074] Specifically, the timing of integrated inspections is determined based on the dynamic switching and priority rules in the path coordination rules, combined with the time index of the path optimization configuration. Time-related constraints are extracted from the path coordination rules, including switching conditions triggered by risk monitoring, the processing order of multi-path conflicts, and multi-machine access restrictions in the coordination rules. Key time nodes are identified from the time index of the path optimization configuration, including the start time of each road segment, the sensor mode switching time, and the control parameter update time. The timing optimization algorithm finds time periods during which multiple paths can be executed in parallel or sequentially. Parallel timing requires that different paths be sufficiently separated spatially, while sequential timing requires sufficient buffer time after the preceding path is completed. Specific thresholds are determined based on robot performance and safety requirements. The integration mode is determined based on the path risk level: low-risk paths allow multiple parallel executions, medium-risk paths only allow sequential execution with increased intervals, and high-risk paths occupy a dedicated time window. The timing schedule table generates the following information: path ID, execution start time, estimated duration, other paths allowed to be parallelized, and conflict avoidance measures.

[0075] The path intersection and overlap areas are determined through task coverage analysis using integrated inspection timing. Based on the timing table, the start time and duration of each path are known, allowing the determination of its active time period [t_start, t_start + duration], where t_start is the path start time and duration is the path execution duration. Space occupancy analysis is based on the basic attributes of the paths; each path covers the nodes it passes through and a defined surrounding area, with the coverage radius determined by sensor detection capabilities. Parallel execution detection directly uses the "Other Paths Allowed for Parallel Execution" field in the timing table to identify simultaneously executing path combinations. Time overlap judgment compares the active time periods of different paths; if path i's [t_start_i, t_end_i] intersects with path j's [t_start_j, t_end_j] and the intersection duration exceeds a set threshold, it is marked as time overlap. Spatial proximity inference is based on timing scheduling logic; parallel execution requires spatial separation to meet safe distance requirements, and serial execution intervals to meet buffer time requirements. The characteristics of the intersection and overlap areas include the set of path IDs involved, the overlapping time interval, the inferred spatial range, and the overlap type (parallel overlap / serial handover / proximity interference). The criteria for identifying key overlaps are determined based on the degree of temporal overlap, the number of paths involved, and the intersection of execution windows.

[0076] In some embodiments, obtaining the inspection efficiency coordination factor from the path intersection and overlap area includes: performing intersection interference analysis on the path intersection and overlap area to determine the coordination strength, wherein the intersection interference analysis includes the congestion level, coverage density, and interference degree of adjacent paths in the overlap area; reallocating the traffic weights of paths in the path intersection and overlap area according to the coordination strength to form a traffic weight allocation; formulating an efficiency optimization strategy using the traffic weight allocation; and obtaining the inspection efficiency coordination factor through the efficiency optimization strategy.

[0077] Intersection interference analysis is performed on overlapping paths to determine coordination intensity. Congestion level is determined based on the number of paths simultaneously existing within the overlapping area; the more paths, the higher the congestion level. Coverage density is calculated as the number of coverages per unit area: ρ_coverage = Σ(coverage area of ​​each path) / total area of ​​the region, where ρ_coverage is the coverage density; a density exceeding a set threshold indicates over-coverage. The degree of interference to adjacent paths is assessed through spatial distance and execution time overlap; the closer the distance and the greater the time overlap, the higher the degree of interference. Coordination intensity S_coord integrates three dimensions: S_coord = w1 × congestion level + w2 × coverage density + w3 × interference level, where S_coord is the coordination intensity, and w1, w2, and w3 are weighting coefficients determined according to the influence of each factor. Intensity grading maps S_coord to standard intervals; different value ranges correspond to weak, medium, and strong coordination needs. Time-varying characteristic analysis is used to analyze the variation of coordination intensity over time, identifying peak periods and stable periods.

[0078] The travel weights of paths within overlapping areas are redistributed based on coordination intensity, forming a travel weight allocation. Different allocation strategies are used according to the coordination intensity level: strict priority allocation is used in areas with strong coordination needs, weighted fair allocation is used in areas with medium coordination needs, and the original weights are maintained in areas with weak coordination needs. The specific allocation algorithm is determined based on priority and coordination needs, with high-priority paths receiving larger weights, and the weight ratio is adjusted according to the coordination intensity. A dynamic adjustment mechanism corrects the weights based on real-time execution. If the actual travel time of a path exceeds the set proportion of the allocation time window, its weight is temporarily adjusted and compensated for by waiting paths. Weight normalization ensures that the sum of the weights of all paths within each area is 1. Time slice allocation discretizes continuous time into fixed-length time slices, and each time slice is allocated to a specific path. Spatial partitioning divides large overlapping areas into grids, and different grids can be allocated to different paths for parallel use.

[0079] Efficiency optimization strategies are implemented using traffic weight allocation. The execution order of paths is adjusted by weight allocation, prioritizing high-weight paths and delaying or selecting alternative time windows for low-weight paths. Spatial optimization strategies reduce overlap through path fine-tuning, adjusting path node positions to stagger the coverage areas of different paths while ensuring task coverage. Speed ​​coordination strategies allocate differentiated speeds based on weight, maintaining normal speeds for high-weight paths and appropriately reducing speeds for low-weight paths in overlapping areas to minimize interference. Resource allocation strategies prioritize sensor and computing resources for high-weight paths, ensuring their data acquisition and processing needs are met. Coordination strategies identify complementary path combinations, scheduling them for simultaneous execution to share sensor data and improve overall efficiency. Buffer zones are established at the boundaries of overlapping areas to create transition zones, allowing paths to complete speed and configuration adjustments before entering the core overlapping area. Strategy combinations are selected based on coordination strength: strong coordination employs multiple combination strategies, medium coordination uses partial strategies, and weak coordination requires only fine-tuning.

[0080] The inspection efficiency coordination factor is obtained through efficiency optimization strategies. The coordination factor is based on path execution effect evaluation, comprehensively calculated by considering the degree of optimization of path execution time, the proportion of improvement in coverage efficiency, and the reduction of inter-path conflicts. The coordination factor consists of four components: the time coordination factor η_time reflects the efficiency improvement brought about by timing optimization, determined based on the proportion of waiting time reduction due to timing adjustments; the space coordination factor η_space reflects the effect of path adjustments, calculated based on the proportion of overlap area reduction due to path adjustments; the resource coordination factor η_resource represents the contribution of resource optimization configuration, based on the effect evaluation of sensor configuration optimization; and the interference loss factor η_interference quantifies the negative impact of residual interference, determined based on the degree of impact of unavoidable path conflicts. The comprehensive coordination factor calculation formula is η_coord = η_time + η_space + η_resource - η_interference, where η_coord is the final inspection efficiency coordination factor, and each component factor is determined based on the actual optimization effect. Factor normalization ensures that η_coord falls within the [0,1] interval.

[0081] The cycle navigation parameters are generated based on the inspection efficiency coordination factor. The cycle base period is adjusted based on the obstacle avoidance cycle parameters determined in S130, T_base=T_0 / (1+η_coord), where T_base is the adjusted base period, T_0 is the standard cycle period, and η_coord is the inspection efficiency coordination factor. The higher the coordination factor, the faster the cycle and the more sensitive the response. A multi-layer cycle system is established, with the multiplier relationship determined according to the response time requirements of the decision-making level. The navigation cycle T_nav=T_base is used for path tracking and obstacle avoidance; the coordination cycle T_coord=n1×T_base is used for multi-path coordination decision-making; and the task cycle T_task=n2×T_base is used for task switching and resource allocation, where n1 and n2 are multiplier coefficients. The cycle synchronization mechanism ensures consistent decision coordination at different levels, and the lower-level cycle must be an integer fraction of the higher-level cycle. Dynamic cycle adjustment is based on real-time coordination needs. When entering a high coordination factor area, the cycle frequency is appropriately increased; after leaving, it gradually returns to normal. Tick ​​phase control coordinates the movements of multiple robots, avoiding synchronization conflicts through phase offset. The phase difference Δφ = 2π / N, where Δφ is the phase difference and N is the number of robots. Special tick modes handle critical scenarios: emergency avoidance mode shortens the tick to improve response speed, and precise docking mode extends the tick to ensure stability. The tick navigation parameter set includes T_nav, T_coord, T_task, tick adjustment rules, and phase configuration. Through a unified timing framework, it coordinates the robots' perception, decision-making, and execution actions, ensuring the synchronous operation of each module and the orderly cooperation of multiple robots during inspection.

[0082] Step S160: Match the beat navigation parameters with the probability inspection chain to generate a dynamic inspection unit, adjust the arrival time of the task point based on the dynamic inspection unit, generate an elastic inspection process according to the arrival time, and complete the intelligent inspection of the humanoid robot with bionic gait.

[0083] In some embodiments, the step of matching the beat navigation parameters with the probabilistic inspection chain to generate a dynamic inspection unit includes: constructing a temporal mapping space using the beat navigation parameters; projecting the probabilistic inspection chain onto the temporal mapping space to obtain spatial projection points; performing clustering processing on the spatial projection points to determine a core node cluster; and reconstructing the dynamic inspection unit based on the core node cluster.

[0084] A temporal mapping space is constructed using beat navigation parameters. Based on the multi-layered beat system in the beat navigation parameters, a unified temporal benchmark framework is established, with the navigation beat T_nav as the basic time unit, and the coordination beat T_coord and task beat T_task as the middle and high-level time scales, respectively. The temporal space adopts a hierarchical structure: the bottom layer marks navigation decision moments with T_nav intervals, the middle layer marks coordination decision moments with T_coord intervals, and the top layer marks task switching moments with T_task intervals. The mapping rule converts actual time into beat units, and the time coordinate is equal to the actual time divided by T_nav, ensuring that all temporal information can be uniformly represented. The phase dimension utilizes the phase configuration in the beat navigation parameters to reflect the temporal coordination relationship between multiple robots. The spatial boundary setting ensures the periodicity and continuity of the mapping; the time axis wraps around when the task is completed, and the phase axis cycles at 2π. Distance calculation in the temporal space comprehensively considers the time interval and beat hierarchy differences, and the weights are allocated according to the importance of the decisions.

[0085] The probabilistic inspection chain is projected onto the temporal mapping space to obtain spatial projection points. Based on the expected arrival time, suggested dwell time, and arrival probability of each node in the probabilistic inspection chain, the nodes are mapped to their corresponding positions in the temporal space. The time coordinates are calculated using the node's expected arrival time and the reference unit of the temporal space, giving each node a definite position in the time dimension. The beat level is determined according to the node's action type: fast-pass nodes correspond to the navigation beat level, regular scan nodes correspond to the coordination beat level, and detailed inspection nodes correspond to the task beat level. The phase coordinates are determined based on the node's relative position and phase configuration in the inspection chain. Probability information is retained as a weight attribute of the projection points; high-probability nodes have a greater influence weight, while low-probability nodes have a smaller weight. Path connectivity is converted into directed associations in the space during projection, with the start and end points corresponding to the projected positions of the connected nodes, and the association strength is determined based on the transition probability. Dwell time affects the expansion range of the projection points in the time dimension. After projection, a set of nodes in the temporal space is obtained, with each point retaining the complete attribute information of the original node.

[0086] For example, the step of clustering the spatial projection points to determine the core node cluster includes: decomposing the spatial projection points into primary projection points and secondary projection points; performing density analysis on the primary projection points to obtain clustering seeds; performing adsorption processing on the clustering seeds and the secondary projection points to form clusters; and determining the core node cluster based on the clusters.

[0087] The spatial projection points are decomposed into primary projection points and secondary projection points. The decomposition criteria are based on the probability weight and connectivity of the projection points in the probabilistic inspection chain. Points with a probability weight greater than a set threshold and a number of connected edges exceeding a set limit are defined as primary projection points; the rest are secondary projection points. Primary projection points typically represent key nodes in the probabilistic inspection chain and undertake the main inspection functions. Secondary projection points include low-probability nodes, auxiliary nodes, and connecting nodes. The decomposition process preserves the topological relationships between projection points, recording the association between each secondary point and the nearest primary point, forming a hierarchical structure. Primary points are relatively concentrated in the temporal space, forming functional regions. Secondary points are distributed around the primary points, playing a supplementary and connecting role.

[0088] Density analysis is performed on the main projection points to obtain clustering seeds. Density calculation is based on neighborhood analysis in the temporal space, counting the number of other main points within a certain range around each main point. The neighborhood range is determined according to the time scale in the clockwise navigation parameters, usually set with the coordinated clockwise T_coord as the benchmark. High-density points are defined as locations within their neighborhoods that contain a sufficient number of other main points; these points become clustering seed candidates. Seed selection avoids spatially too close redundant seeds, ensuring that the seed spacing meets the minimum interval requirement determined based on the clockwise parameters. Seed priority is determined based on two factors: local density value and the average value of node probability weights. The temporal characteristics of the seeds reflect their position in the inspection process; early seeds are suitable for forming initial clusters, while late seeds are suitable for forming final clusters.

[0089] Clustering seeds and secondary projection points are adsorbed to form clusters. Adsorption rules are based on distance and node attribute similarity in the temporal space; secondary points are adsorbed to the nearest and functionally compatible seed. Distance calculation is based on time intervals and beat hierarchy differences in the temporal space, comprehensively considering differences in time and function dimensions. Attribute compatibility requires beat type matching to ensure that the execution characteristics of the adsorbed points and seeds are coordinated. Adsorption strength decays with temporal distance, generating adsorption within the effective influence range of the seed, which is determined based on beat navigation parameters. Competitive adsorption handles situations where multiple seeds simultaneously attract a secondary point, comparing the adsorption strength of various seeds and selecting the optimal match. Iterative adsorption allows already adsorbed points to become new adsorption centers, but the adsorption force decreases progressively. If the number of points adsorbed by a certain seed exceeds the processing capacity, it automatically splits into multiple sub-seeds.

[0090] The core node clusters are determined based on clustering. Cluster construction extracts key features from each cluster, including the temporal position of the cluster center, the number of nodes within the cluster, the average probability weight, and the temporal span. Core nodes are selected based on the highest probability weights within each cluster, ensuring coverage of the cluster's main inspection functions. Cluster boundaries define the temporal influence range of each cluster in the temporal space. Inter-cluster relationship analysis considers adjacency and dependency; temporally adjacent clusters are assigned sequential execution relationships, while functionally complementary clusters are assigned coordination relationships. Cluster attributes integrate the characteristics of all nodes within the cluster, including dominant beat type, time window range, and probability weight distribution. Each cluster contains sufficient task execution nodes and possesses relatively independent inspection capabilities.

[0091] The dynamic inspection unit is reconstructed based on the core node cluster. The reconstruction process transforms each core node cluster into an independently executable dynamic unit. The unit's execution time window is determined based on the temporal distribution of nodes within the cluster; the start time is the earliest node's time, and the end time is the latest node's time plus necessary buffering. The tick configuration inherits the cluster's dominant tick type and is fine-tuned according to task characteristics within the cluster. Units dominated by fast tasks use shorter ticks to improve response speed, while units dominated by complex tasks use longer ticks to ensure stable execution. Tasks within the unit are arranged in the logical order of the original probabilistic inspection chain to maintain execution continuity. The overall execution probability of the unit is based on a comprehensive calculation of the probabilities of each node within the cluster, providing a reliability assessment for unit execution. Appropriate sensor bandwidth, computing resources, and energy budgets are allocated according to the cluster size and task type. Adjustable time and execution range are configured for each unit, allowing for adaptive adjustments based on real-time conditions.

[0092] The arrival timing of task points is adjusted based on dynamic inspection units. Timing adjustment uses dynamic inspection units as the basic unit, ensuring that the time arrangement of each task point within a unit is consistent with the unit's cycle time configuration. The expected arrival time and suggested dwell time for each task point are obtained from the probabilistic inspection chain. The execution time of task points is allocated to the unit's time window, ensuring sufficient switching time between tasks; the switching time is determined based on cycle time navigation parameters. The timing arrangement of task points between adjacent units considers the transition time between units to avoid time conflicts. High-priority task points receive ideal timing first, while low-priority task points are flexibly scheduled according to remaining time. When the actual execution time deviates from the plan, the arrival timing of subsequent task points is automatically adjusted; the adjustment range is determined based on the cycle time, maintaining the controllability of the overall progress. The adjusted arrival timing ensures that it meets the task's time constraints and probabilistic requirements.

[0093] A flexible inspection process is generated based on arrival timing to complete intelligent inspection of the humanoid robot with a biomimetic gait. All high-priority task points are connected according to the adjusted time sequence to form a basic inspection route, ensuring the completion of core tasks. Conditional execution branches are designed for optional task points, and execution is determined based on time margin and system status; branch conditions are determined based on the time leeway of arrival timing. A suitable gait mode is selected based on the time interval and spatial distance between adjacent task points: a fast gait is used between task points with short time intervals, a stable gait is used for those with ample time intervals, and an adaptive gait is used when traversing complex terrain. A gait transition mechanism is established between task points, and the gait switching time is determined based on the time allocation of arrival timing to ensure the continuity and stability of movement. The gait frequency is adjusted according to the beat frequency in the beat navigation parameters to achieve synchronization between gait and system beat. The process is allowed to be adjusted during execution based on real-time feedback, including task jumps, path fine-tuning, and gait switching; the adjustment range is determined based on the flexible parameters of the dynamic inspection unit. Task execution progress and gait performance are tracked to evaluate the conformity between actual arrival time and planned timing. Define handling strategies for critical anomalies, such as detour plans when obstacles block the way and degraded execution plans when equipment malfunctions. These strategies are integrated with the branching mechanism of the flexible inspection process. Coordinate perception, decision-making, control, and communication modules to ensure seamless integration of biomimetic gait with inspection tasks, with each module synchronized according to unified rhythmic navigation parameters. Ultimately, end-to-end intelligent control, from high-level task planning to low-level gait control, is achieved, enabling humanoid robots to complete complex inspection tasks with natural and efficient biomimetic gait.

[0094] To implement the above-described method embodiments, a biomimetic gait humanoid robot intelligent inspection method is proposed to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a humanoid robot intelligent inspection device 200 with a biomimetic gait, according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The humanoid robot intelligent inspection device 200 with a biomimetic gait provided in this embodiment includes:

[0095] Task planning module 201 is used to acquire task point distribution data of the inspection area, generate inspection coverage requirements based on the task point distribution data, determine the necessary node sequence according to the inspection coverage requirements, and establish an inspection status mapping table based on the necessary node sequence.

[0096] The path generation module 202 is used to perform probabilistic backtracking processing on the inspection status mapping table to generate a path probability distribution tree, perform branch analysis on the path probability distribution tree to obtain reachability probability values, determine the main path family and alternative path family based on the reachability probability values, and generate a probabilistic inspection chain based on the main path family and alternative path family.

[0097] The path optimization module 203 is used to decompose the probabilistic inspection chain into continuous road segments, obtain the passage difficulty coefficient and path complexity index from the continuous road segments, predict the gait energy consumption distribution based on the passage difficulty coefficient, determine the obstacle avoidance rhythm based on the path complexity index, generate path compensation parameters by combining the gait energy consumption distribution and the obstacle avoidance rhythm, and generate path optimization configuration using the path compensation parameters.

[0098] The risk balancing module 204 is used to perform risk assessment on the continuous road segments to obtain traffic risk attributes, determine high-risk paths and safe paths based on the traffic risk attributes, combine the high-risk paths and the safe paths to form risk balancing path pairs, and generate path coordination rules based on the risk balancing path pairs.

[0099] The timing coordination module 205 is used to determine the integrated inspection timing based on the path coordination rules and the path optimization configuration, perform task coverage analysis through the integrated inspection timing to determine the path intersection and overlap area, obtain the inspection efficiency coordination factor from the path intersection and overlap area, and generate the beat navigation parameters based on the inspection efficiency coordination factor.

[0100] The dynamic execution module 206 is used to match the beat navigation parameters with the probability inspection chain to generate a dynamic inspection unit, adjust the arrival time of the task point based on the dynamic inspection unit, generate an elastic inspection process according to the arrival time, and complete the intelligent inspection of the humanoid robot with bionic gait.

[0101] The aforementioned biomimetic gait humanoid robot intelligent inspection device 200 can implement a biomimetic gait humanoid robot intelligent inspection method according to the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0102] like Figure 3 As shown, the third embodiment of the present invention also provides a computer device, including a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, characterized in that the processor 302 executes the program to implement the steps of the biomimetic gait humanoid robot intelligent inspection method described in the first embodiment of the present invention.

[0103] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0104] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A biomimetic gait-inspired humanoid robot intelligent inspection method, characterized in that, include: Obtain task point distribution data of the inspection area, generate inspection coverage requirements based on the task point distribution data, determine the necessary node sequence according to the inspection coverage requirements, and establish an inspection status mapping table based on the necessary node sequence. The inspection status mapping table is subjected to probabilistic backtracking processing to generate a path probability distribution tree. Branch analysis is performed on the path probability distribution tree to obtain the reachability probability value. Based on the reachability probability value, the main path family and the alternative path family are determined. A probabilistic inspection chain is generated based on the main path family and the alternative path family. The probabilistic inspection chain is decomposed into continuous road segments. The passage difficulty coefficient and path complexity index are obtained from the continuous road segments. The gait energy consumption distribution is predicted based on the passage difficulty coefficient. The obstacle avoidance rhythm is determined based on the path complexity index. The path compensation parameters are generated by combining the gait energy consumption distribution and the obstacle avoidance rhythm. The path optimization configuration is generated using the path compensation parameters. Risk assessment is performed on the continuous road segments to obtain traffic risk attributes. Based on the traffic risk attributes, high-risk paths and safe paths are determined. The high-risk paths and safe paths are combined to form risk-balanced path pairs. Path coordination rules are generated based on the risk-balanced path pairs. Based on the path coordination rules and the path optimization configuration, the timing of integrated inspection is determined. The task coverage analysis is performed through the integrated inspection timing to determine the path intersection and overlap area. The inspection efficiency coordination factor is obtained from the path intersection and overlap area. The beat navigation parameters are generated based on the inspection efficiency coordination factor. The rhythmic navigation parameters are matched with the probabilistic inspection chain to generate a dynamic inspection unit. The arrival time of the task point is adjusted based on the dynamic inspection unit, and an elastic inspection process is generated according to the arrival time to complete the intelligent inspection of the humanoid robot with biomimetic gait.

2. The method according to claim 1, characterized in that, The step of performing branch analysis on the path probability distribution tree to obtain the reachability probability value includes: Identify the main branches and secondary branches based on the path probability distribution tree; Perform probability decay analysis on the main branches to obtain the decay coefficient; The original probabilities of the main branch and the secondary branch are corrected according to the attenuation coefficient; The reachability probability value is obtained by threshold determination based on the corrected probability.

3. The method according to claim 1, characterized in that, The step of generating a probabilistic inspection chain based on the main path family and the alternative path family includes: Construct a primary propagation skeleton based on the aforementioned main path family; Extended analysis is performed on the primary propagation skeleton to obtain extended nodes; Based on the candidate path family, the extended nodes are supplemented at multiple levels to generate an enhanced path network; The enhanced path network is subjected to probability assignment processing to generate a probabilistic inspection chain.

4. The method according to claim 1, characterized in that, The method of combining the gait energy consumption distribution and the obstacle avoidance beat to generate path compensation parameters includes: The energy density distribution is obtained by performing energy density calculation on the gait energy consumption distribution; Identify key energy consumption areas from the energy concentration distribution; The energy consumption key area is time-series fused with the obstacle avoidance rhythm to obtain compensation timing characteristics; Path compensation parameters are constructed based on the compensation timing characteristics.

5. The method according to claim 1, characterized in that, The process of determining high-risk paths and safe paths based on the traffic risk attributes includes: Based on the aforementioned traffic risk attributes, risk gradient analysis is performed to obtain the gradient distribution; Clustering is performed on the gradient distribution to identify risk clustering areas; Connectivity analysis is performed on the risk cluster area to generate high-risk paths; A low-value region search is performed on the gradient distribution, and the low-risk nodes found are connected to construct a safe path.

6. The method according to claim 1, characterized in that, The step of obtaining the inspection efficiency coordination factor from the path intersection and overlap area includes: The coordination intensity is determined by performing intersection interference analysis on the overlapping areas of the paths. The intersection interference analysis includes the congestion level, coverage density and interference degree of adjacent paths in the overlapping areas. Based on the coordination strength, the passage weights of the paths within the path intersection and overlap area are redistributed to form a passage weight allocation; Efficiency optimization strategies are formulated using the aforementioned passage weight allocation; The efficiency optimization strategy described above is used to obtain the inspection efficiency coordination factor.

7. The method according to claim 1, characterized in that, The step of matching the beat navigation parameters with the probabilistic inspection chain to generate a dynamic inspection unit includes: A temporal mapping space is constructed using the beat navigation parameters; The probabilistic inspection chain is projected onto the temporal mapping space to obtain spatial projection points; Clustering is performed on the spatial projection points to determine the core node cluster; The dynamic inspection unit is reconstructed based on the core node cluster.

8. The method according to claim 7, characterized in that, The step of clustering the spatial projection points to determine the core node cluster includes: The spatial projection points are decomposed into primary projection points and secondary projection points; Density analysis is performed on the main projection points to obtain clustering seeds; The clustering seed and the secondary projection point are adsorbed to form a cluster; The core node cluster is determined based on the clustering.

9. A humanoid robot intelligent inspection device with biomimetic gait, characterized in that, include: The task planning module is used to acquire task point distribution data of the inspection area, generate inspection coverage requirements based on the task point distribution data, determine the necessary node sequence according to the inspection coverage requirements, and establish an inspection status mapping table based on the necessary node sequence. The path generation module is used to perform probabilistic backtracking processing on the inspection status mapping table to generate a path probability distribution tree, perform branch analysis on the path probability distribution tree to obtain reachability probability values, determine the main path family and alternative path family based on the reachability probability values, and generate a probabilistic inspection chain based on the main path family and alternative path family. The path optimization module is used to decompose the probabilistic inspection chain into continuous road segments, obtain the passage difficulty coefficient and path complexity index from the continuous road segments, predict the gait energy consumption distribution based on the passage difficulty coefficient, determine the obstacle avoidance rhythm based on the path complexity index, generate path compensation parameters by combining the gait energy consumption distribution and the obstacle avoidance rhythm, and generate path optimization configuration using the path compensation parameters. The risk balancing module is used to perform risk assessment on the continuous road segments to obtain traffic risk attributes, determine high-risk paths and safe paths based on the traffic risk attributes, combine the high-risk paths and the safe paths to form risk balancing path pairs, and generate path coordination rules based on the risk balancing path pairs. The timing coordination module is used to determine the timing of integrated inspection based on the path coordination rules and the path optimization configuration, perform task coverage analysis through the integrated inspection timing to determine the path intersection and overlap area, obtain the inspection efficiency coordination factor from the path intersection and overlap area, and generate the beat navigation parameters based on the inspection efficiency coordination factor. The dynamic execution module is used to match the beat navigation parameters with the probability inspection chain to generate dynamic inspection units, adjust the arrival time of task points based on the dynamic inspection units, generate an elastic inspection process according to the arrival time, and complete the intelligent inspection of the humanoid robot with bionic gait.

10. A computer device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the method as described in any one of claims 1 to 8.

Images