Multi-path cooperation method, device and equipment for cluster quadruped inspection robot

Abstract

The invention provides a multi-path cooperation method, device and equipment for a cluster quadruped inspection robot, and the method comprises the steps: obtaining gait dynamics data and topographic data, and recognizing a topographic adaptive rhythm based on the spectrum decomposition of a foot end contact force sequence; converting the topographic data into a potential energy gradient field to construct a dynamic potential energy diagram, and generating a natural path network; a cooperative trigger point is set based on the stable support moment, a path ripple field is formed, and a multi-machine cooperative potential well is constructed; decomposing the path network into a dominant task chain and an auxiliary task chain, and weaving to form an initial collaborative path; rigid nodes and flexible sections are identified through topological transformation, and random disturbance is injected to reconstruct a self-adaptive collaborative path; gait phase information is extracted to construct a synchronous network, a potential energy flow vector is used for phase alignment to generate a group resonance frequency, and a resonance cooperation path is formed; and finally, an execution instruction is generated based on the resonance cooperation path and the terrain adaptability rhythm, and multi-path cooperation of the cluster quadruped inspection robot in the complex terrain environment is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot-free path planning, and in particular to a multi-path collaboration method, device and equipment for clustered quadruped inspection robots. Background Art

[0002] Multi-path coordination of swarm quadruped inspection robots in complex terrain environments is a key research topic in robotic path planning, with broad application prospects in scenarios such as industrial inspection, environmental monitoring, and security patrols. However, traditional swarm path planning methods, primarily based on geometric path optimization and simple task allocation strategies, lack a deep understanding of terrain characteristics and the dynamics of robot gaits, making it difficult to achieve truly terrain-adaptive multi-path coordination.

[0003] In existing technologies, most quadruped robot swarm path planning systems employ a hierarchical control architecture, separating global path planning from local motion control. While this approach reduces system complexity, it ignores the profound impact of the terrain environment on the robot's gait rhythm and fails to fully utilize the rich dynamic information contained in the robot's foot-ground contact process. Furthermore, traditional methods lack effective modeling of the swarm's internal phase synchronization mechanisms, leading to poor coordination between robots, frequent path conflicts, and low inspection efficiency. In particular, in complex environments such as rugged terrain, soft soil, and varying slopes, multi-path coordination is significantly reduced, and path planning can even fail, severely hindering the practical application of swarm quadruped inspection robot technology. Summary of the Invention

[0004] The present invention provides a multi-path collaboration method, device and equipment for clustered quadruped inspection robots, aiming to construct a terrain adaptability rhythm by deeply exploring the intrinsic relationship between the robot's gait dynamics characteristics and the terrain environment, and based on the potential energy field theory and phase synchronization mechanism, realize the intelligent multi-path collaboration of clustered quadruped inspection robots in complex dynamic terrain.

[0005] The first aspect of the present invention provides a multi-path collaboration method for clustered quadruped inspection robots, comprising the following steps: Acquiring gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extracting a foot-end contact force sequence based on the gait dynamics data, performing spectral decomposition on the foot-end contact force sequence to generate a gait vibration pattern, and identifying a terrain adaptability rhythm through the gait vibration pattern; Converting the environmental terrain data into a potential energy gradient field, constructing a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, performing flow field analysis on the dynamic potential energy map to generate a potential energy flow vector, and generating a natural path network through the potential energy flow vector; Extracting a stable support moment from the foot-end contact force sequence, setting a collaborative trigger point based on the stable support moment and the natural path network, releasing a path exploration signal at the collaborative trigger point, forming a path ripple field through the propagation of the path exploration signal, and constructing a multi-machine collaborative potential well based on the path ripple field; Decomposing the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well, forming a core path skeleton by pulling the dominant task chain, forming a supporting path network by using the auxiliary task chain, and weaving the core path skeleton and the supporting path network to form an initial collaborative path; Performing a topological transformation on the initial collaborative path to generate a path elasticity coefficient, identifying rigid nodes and flexible segments based on the path elasticity coefficient, injecting random perturbations into the flexible segments to generate a path adaptive factor, and reconstructing the adaptive collaborative path using the path adaptive factor; extracting gait phase information from the adaptive collaborative path to form a gait phase matrix, constructing a phase synchronization network based on the gait phase matrix and the rigid nodes, performing group phase alignment using the phase synchronization network and the potential energy flow vector to generate a group resonance frequency, and driving the formation of a resonant collaborative path using the group resonance frequency; The final execution instruction is generated based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robot.

[0006] The second aspect of the present invention provides a multi-path collaboration device for a cluster of quadruped inspection robots, comprising: a data acquisition module for acquiring gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extracting a foot-end contact force sequence based on the gait dynamics data, performing spectral decomposition on the foot-end contact force sequence to generate a gait vibration pattern, and identifying a terrain adaptability rhythm through the gait vibration pattern; a path generation module, configured to convert the environmental terrain data into a potential energy gradient field, construct a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, perform flow field analysis on the dynamic potential energy map to generate a potential energy flow vector, and generate a natural path network using the potential energy flow vector; a collaborative control module, configured to extract a stable support moment from the foot-end contact force sequence, set a collaborative trigger point based on the stable support moment and the natural path network, release a path exploration signal at the collaborative trigger point, form a path ripple field through the propagation of the path exploration signal, and construct a multi-machine collaborative potential well based on the path ripple field; a path optimization module, configured to decompose the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well, form a core path skeleton by pulling the dominant task chain, form a supporting path network by using the auxiliary task chains, and weave the core path skeleton and the supporting path network to form an initial collaborative path; an adaptive adjustment module, configured to perform a topological transformation on the initial collaborative path to generate a path elasticity coefficient, identify rigid nodes and flexible segments based on the path elasticity coefficient, inject random perturbations into the flexible segments to generate a path adaptive factor, and reconstruct the adaptive collaborative path using the path adaptive factor; a phase synchronization module, configured to extract gait phase information from the adaptive collaborative path to form a gait phase matrix, construct a phase synchronization network based on the gait phase matrix and the rigid nodes, utilize the phase synchronization network and the potential energy flow vector to perform group phase alignment to generate a group resonance frequency, and utilize the group resonance frequency to drive the formation of a resonant collaborative path; The execution control module is used to generate a final execution instruction based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robot.

[0007] The third aspect of the present invention proposes a computer device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the steps of a multi-path collaboration method of a clustered quadruped inspection robot disclosed in the first aspect are implemented.

[0008] The beneficial effects of the present invention are reflected in the following points: 1. Through terrain-adaptive rhythm recognition and potential field construction, the generation of the initial collaborative path is realized from terrain perception, and the complex terrain features are converted into an executable multi-path structure, which solves the problem of insufficient utilization of terrain information in traditional methods and improves the matching degree between path planning and inspection environment. 2. Using topological transformation and random perturbation injection technology, the initial collaborative path is optimized into an adaptive collaborative path. By identifying the rigid nodes and flexible segment characteristics of the path, the multi-path has the ability to make real-time adjustments according to environmental changes and execution deviations, thereby improving the robustness and environmental adaptability of the system. 3. A phase synchronization network is established and combined with the potential energy flow vector for group phase alignment, and the adaptive collaborative path is further optimized into a resonant collaborative path. The group resonance frequency is used to achieve high coordination of cluster movement and optimization of energy consumption, ensuring the multi-path collaborative effect and inspection efficiency of the cluster quadruped inspection robot in a complex terrain environment.

[0009] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings herein illustrate specific examples of the technical solutions described in the present invention, and together with the specific implementation methods constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.

[0011] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.

[0012] Figure 1 It is a flow chart of a multi-path collaboration method of a clustered quadruped inspection robot according to the present invention.

[0013] Figure 2 This is a structural block diagram of a multi-path collaborative device of a clustered quadruped inspection robot according to the present invention.

[0014] Figure 3 It is a structural schematic diagram of a computer device of the present invention. DETAILED DESCRIPTION

[0015] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0016] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0017] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0018] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0019] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0020] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in 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 "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0021] The technical solutions of the embodiments of this application are introduced below.

[0022] like Figure 1 As shown, an embodiment of the present invention provides a multi-path collaboration method for clustered quadruped inspection robots, including the following steps S110 to S170: Step S110: Obtain gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extract the foot end contact force sequence based on the gait dynamics data, perform spectral decomposition on the foot end contact force sequence to generate a gait vibration pattern, and identify the terrain adaptability rhythm through the gait vibration pattern.

[0023] Specifically, gait dynamics data acquisition utilizes a high-frequency sampling strategy, acquiring real-time motion state information from IMU sensors, force sensors, and encoders installed on the robot's joints, feet, and torso. The IMU sensors collect triaxial acceleration, angular velocity, and attitude angle data at a frequency of 1000Hz, force sensors measure contact force and torque at each foot end at a frequency of 500Hz, and joint encoders record joint angles and angular velocities at a frequency of 200Hz. Data synchronization utilizes a GPS clock synchronization protocol to ensure that data timestamp errors for all robots in the cluster are within 1ms. Environmental terrain data acquisition utilizes multi-layered perception using lidar, depth cameras, and tactile sensors. The lidar scans at a frequency of 10Hz to generate a 3D point cloud, the depth camera acquires ground texture and slope information at 30fps, and the tactile sensor measures ground hardness, friction coefficient, and roughness in real time. Terrain data is gridded, discretizing the continuous terrain into 0.1m×0.1m grid cells. Each grid cell records elevation, slope angle, surface material, and friction parameters. The cluster communication protocol establishes a high-speed data link based on the WiFi6 standard. Each robot uploads a data packet containing a timestamp, robot ID, sensor type, and value array to the central control node every 100ms.

[0024] Contact force sequence extraction extracts the vertical normal force component Fz and the horizontal tangential force components Fx and Fy at the foot end from force sensor data in the gait dynamics dataset. This is then combined with IMU attitude angle data for gravity compensation to form a three-dimensional contact force vector F(t) = [Fx(t), Fy(t), Fz(t)]. Contact phase identification analyzes the variation pattern of the normal force Fz to determine the stance and swing phases of gait: stance phase is defined when Fz > 5N, swing phase when Fz < 1N, and transition phase when 1N ≤ Fz ≤ 5N. Gait cycle segmentation defines a complete gait cycle as the interval between two consecutive transitions from the swing phase to the stance phase, with a typical cycle length of 0.4-0.8 seconds. Contact force characteristic parameters extracted include peak contact force Fmax (maximum normal force during the stance phase), force loading rate dF / dt (rate of change from contact to peak force), stance time Tsupport (stance phase duration), and force integral ∫F(t)dt (impulse throughout the stance phase). Multi-leg coordination analysis identifies the robot's gait pattern by comparing the phase relationship of the quadruped contact force series: the phase difference between the diagonal foot end forces in a diagonal trot is π, while the phase difference between the ipsilateral forefoot and hindfoot in a pacing gait is π / 2. Time series interpolation uses cubic spline interpolation to unify data with different sampling frequencies to 1000 Hz, while the terrain grid data is spatially and temporally aligned with the robot's position.

[0025] The established foot-end contact force sequence was transformed in the frequency domain and subjected to spectral analysis. The vibration modes that characterize the robot's gait were decomposed for different gait modes and the support phase data. Spectral decomposition used a fast Fourier transform (FFT) to convert the time-domain contact force sequence F(t) into a frequency-domain representation F(ω). The transformation formula is F(ω)=∫F(t)e^(-jωt)dt, where F(t) is the time-domain contact force signal, ω is the angular frequency, and j is the imaginary unit. The power spectral density (PSD) was estimated using the Welch method. The contact force sequence was divided into 50% overlapping time windows, each 2 seconds long. A Hanning window function was applied to minimize spectral leakage. The frequency bands were defined based on the characteristics of the quadruped robot's gait: the low-frequency band of 0.5-2 Hz corresponds to the gait fundamental frequency, the mid-frequency band of 2-10 Hz reflects the foot-end impact response, and the high-frequency band of 10-50 Hz reflects the ground material excitation. The fundamental gait frequency is extracted by identifying the main peak in the PSD curve. This peak frequency corresponds to the robot's gait frequency fs, typically 1-3 Hz. Vibration mode identification uses empirical mode decomposition (EMD) to decompose the contact force signal into multiple intrinsic mode functions (IMFs), each representing a vibration component within a specific frequency range. Gait vibration patterns are constructed by combining the main IMF components. The top five modes with energy exceeding 85% of the total energy are retained as the primary vibration modes and the corresponding terrain parameters are annotated.

[0026] Terrain adaptive rhythms were identified by combining gait vibration patterns with real-time terrain grid data. Slope adaptation identification used slope angle θ acquired by a depth camera. By comparing the fundamental frequency of vibration patterns at different slopes, it was found that cadence decreased to 0.8-1.2 Hz and amplitude increased by 30% during uphill gait (θ > 10°), while cadence increased to 2-2.5 Hz and high-frequency components were enhanced during downhill gait (θ < -10°). Surface material response analysis, using roughness parameters detected by tactile sensors, showed that the power spectral density in the mid-frequency range (2-10 Hz) increased by 2-3 times when the surface material changed from smooth to rough, reflecting frequent foot-to-ground impact. Friction adaptation was characterized by enhanced high-frequency oscillations of the tangential force component on low-friction surfaces (μ < 0.3), with frequencies concentrated in the 15-25 Hz range. Rhythm pattern recognition employed a dynamic time warping (DTW) algorithm to match similar vibration pattern sequences and, combined with terrain parameter clustering, identified stable adaptive rhythms. Terrain-adaptive rhythms are defined as the consistent vibration patterns exhibited by a robot swarm under specific terrain conditions. These rhythms are composed of three core parameters: rhythm frequency (fr), rhythm intensity (Ar), and phase synchronization (Φ). The rhythm frequency is determined by statistically analyzing the dominant frequency of the vibration pattern and mapped to the terrain parameters using the relationship fr = f(θ, μ). Typical values ​​for different terrains are: 1.5-2.0 Hz on flat ground, 1.0-1.5 Hz on uphill slopes, 2.0-2.8 Hz on downhill slopes, and 1.2-1.8 Hz on rough surfaces. Rhythm intensity is expressed as the normalized value of the vibration pattern energy. Based on quantitative correlation analysis between vibration patterns and terrain parameters, terrain-adaptive rhythms reflecting the swarm's adaptability were identified.

[0027] Step S120 , converting the environmental terrain data into a potential energy gradient field, constructing a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, performing flow field analysis on the dynamic potential energy map to generate potential energy flow vectors, and generating a natural path network through the potential energy flow vectors.

[0028] Specifically, the environmental terrain data is converted into a potential energy gradient field. The 0.1m×0.1m grid unit in the terrain dataset is used as the basic discrete unit of the potential energy field. The elevation value, slope angle, surface material and friction parameter of each grid are converted into corresponding potential energy components. The potential energy function is constructed using a multi-component superposition mode, and the calculation formula is U(x,y)=αgh(x,y)+β∇ 2 h(x,y)+γρmat(x,y)+δμ(x,y), where U(x,y) is the total potential energy at position (x,y), α, β, γ, δ are weight coefficients, g is the acceleration of gravity, h(x,y) is the elevation interpolation function, ∇ 2h(x,y) is the Laplace operator of elevation, ρmat(x,y) is the material impedance coefficient, and μ(x,y) is the friction parameter. The potential energy gradient is calculated using the finite difference method, with the gradient vector representing the rate of change of potential energy in each direction. The potential energy field is discretized into a potential energy matrix, and the potential energy difference between adjacent grid cells drives the robot's motion. The gradient field amplitude reflects the steepness of the terrain, and the gradient direction indicates the direction of maximum potential energy drop.

[0029] The dynamic potential energy map construction process integrates the static potential energy gradient field with the three core parameters of the terrain adaptability rhythm. The rhythm frequency fr is used as a temporal modulation factor to periodically modulate the potential energy field, simulating the periodic characteristics of the robot's gait. The rhythm intensity Ar is used as an amplitude amplification factor, which is multiplied by the gradient field amplitude to form an enhanced gradient field, reflecting the robot's motion excitation intensity under different terrains. The phase synchronization degree Φ is converted into a phase delay function φ(x,y)=Φ×√[(x-x0)] through the spatial distance function. 2 +(y-y0) 2 ], where (x0, y0) are the coordinates of the cluster's centroid, establishing a phase coupling relationship related to spatial position. The dynamic potential energy function is expressed as Udyn(x,y,t)=U(x,y)[1+Ar×sin(2πfrt+φ(x,y))], which incorporates both spatial variation and temporal evolution. The temporal discretization of the potential energy map uses a fixed time step to ensure that each rhythm cycle contains sufficient temporal sampling points. Spatial interpolation expands the discrete grid into a continuous potential energy surface, improving spatial resolution.

[0030] The constructed dynamic potential energy map is subjected to fluid dynamics flow field analysis to extract the flow characteristics of the potential energy in the spatiotemporal domain and generate a potential energy flow vector field. The potential energy flow velocity is calculated using Darcy's law: the velocity vector v = -k∇Udyn / μf × [1+κcos(φ(x,y))], where k is the permeability coefficient, μf is the fluid viscosity coefficient, ∇Udyn is the spatial gradient of the dynamic potential energy, and κ is the phase coupling coefficient. Streamline tracing uses numerical integration to solve the streamline differential equation and calculate streamline trajectories from each grid point. Flow field divergence calculation is used to identify potential energy sources and sinks. Positive divergence indicates potential energy divergence, while negative divergence indicates potential energy convergence. Vorticity calculation is used to identify vortex structures and circulation regions in the flow field. Regions with vorticity greater than the threshold |ω|>ωth are marked as path avoidance zones. The potential energy flow vector field is stored as a vector matrix, containing the velocity vector information for each grid point. Critical points are identified by finding points where the velocity is zero. The type of critical point is determined by the eigenvalues ​​of the velocity gradient matrix. Saddle points are considered potential path bifurcation points. Flow calculations integrate the velocity at a given cross-section to determine the potential flow through that cross-section.

[0031] In some embodiments, generating a natural path network through the potential energy flow vector field includes: identifying potential energy convergence points and divergence points based on the potential energy flow vector field; constructing a potential energy transmission channel between the convergence points and the divergence points; branching the potential energy transmission channel to form a tree-like path structure; and meshing the tree-like path structure to form a natural path network.

[0032] Converging points are identified by searching for regions with negative divergences and large amplitudes, with the condition div(v) < -δ, where δ is the divergence threshold, set to twice the standard deviation of the flow field divergence. Diverging points are identified by searching for regions with positive divergences, with the condition div(v) > δ. Point strength is quantified using the divergence amplitude |div(v)|. Points with greater strength have a wider impact on the surrounding flow field. The influence radius is calculated by analyzing the divergence decay pattern around each point. The influence radius R is defined as the distance where the divergence decays to 10% of the peak value. The spatial distribution of converging and diverging points is merged using a clustering algorithm. Points of the same type within a distance of 0.3m are merged into a single feature point. The centroid coordinates of the merged point are used, and the intensity is the sum of the intensities of each point. Saddle point critical points are added as supplementary nodes to the feature point set. Feature point screening retains the top 50% of the intensity ranking as primary converging and diverging points to ensure the importance of network nodes. Point-pair matching uses the shortest distance principle to pair each diverging point with the nearest converging point, forming a source-sink pair for potential energy transfer. The pairing constraint requires that the distance between point pairs does not exceed 5 meters. Isolated points exceeding this distance are marked as boundary points.

[0033] Between identified convergence and divergence points, potential energy transfer channels are constructed along the streamlines of the potential energy flow vector field. Streamline tracing begins at each divergence point and numerically integrates along the velocity vector. The integration step is set to 1 / 10 of the grid spacing to ensure path tracking accuracy and avoid vorticity avoidance zones. Tracking terminates when the convergence point is reached or the velocity drops below the threshold |v| < 0.01 m / s. To improve channel coverage, multi-start point tracking uses five start points within a 0.2 m radius around each divergence point to generate multiple candidate channels. Channel selection utilizes a comprehensive score based on the shortest path and maximum flow rate. The scoring function S = w1 / L + w2∫|v|dl, where L is the path length, ∫|v|dl is the total flow rate along the path, and w1 and w2 are weighting coefficients. Channel width is adaptively set based on flow rate, ranging from 0.5 m for high flow rates to 0.2 m for low flow rates. Channel merging combines parallel channels within a spatial distance of less than 0.3m into a single channel. The combined channel width is the sum of the individual channel widths. Channel verification checks the bidirectional accessibility of channels by tracing reverse streamlines. Unidirectional channels are marked as special paths. Channel parameter records include the start and end coordinates, path length, average width, flow rate, and curvature radius.

[0034] The potential energy transmission channel is branched and grown to form a tree-like path structure. With the main transmission channel as the trunk, branch paths are generated at key channel nodes. Branch point selection criteria include: turning points with a channel curvature radius less than 2m, flow field abrupt changes where the velocity gradient changes by more than 20%, cross-sectional change points where the channel width changes significantly, and saddle point critical points. The branch direction is determined by analyzing the flow field distribution around this point, with three alternative directions selected: angles of 45°, 90°, and 135° to the mainstream direction. A constrained growth model is used for branch length, with a maximum length not exceeding 50% of the main channel length and a minimum length of 0.5m. The growth rule simulates the branching characteristics of natural rivers, and the number of branching levels is controlled by Horton's law, with the number of branches at level n being 1 / 3 of the number at level (n-1). Branch termination conditions include encountering other channels or branches, entering a low potential energy region, the velocity dropping below a threshold, reaching a preset maximum length, and entering a vorticity avoidance zone. The tree structure encoding uses a hierarchical identification system, with the trunk at level 0, first-level branches at 1.1, 1.2, and 1.3, and second-level branches at 1.1.1, 1.1.2, etc. Branch weights inherit the weight of the main channel and decrease according to the number of branch levels. The weight of an n-level branch is 1 / 2^n of the main channel.

[0035] The resulting tree-like path structures are meshed together through lateral connections and loop construction to form a natural path network. First, proximity areas between different tree structures are identified. Potential connections are established when the distance between two branch endpoints is less than 0.6m. Connection feasibility is assessed by analyzing the potential energy distribution and terrain obstacles along the connection path. Connections with smooth potential energy changes, no obvious obstacles, and no eddy avoidance zones are marked as feasible. Lateral connection construction uses a shortest path algorithm to establish straight or simple curve connections between feasible connection points. The connection path width is set to the smaller of the connection branch widths. Loop detection uses a depth-first search algorithm to identify closed loops in the network, recording the loop length and number of nodes involved. Loop processing removes redundant small loops (with a circumference less than 2m) and retains large loops that are useful for path planning. Network density control limits the number of connections per node to no more than 6 by setting a maximum connectivity degree to prevent network overcomplexity. Connected component analysis ensures the connectivity of the entire network, and isolated small components are integrated into the main network by adding connecting edges.

[0036] Step S130: extract the stable support moment from the foot-end contact force sequence, set the collaborative trigger point based on the stable support moment and the natural path network, release the path exploration signal at the collaborative trigger point, form a path ripple field through the propagation of the path exploration signal, and construct a multi-machine collaborative potential well based on the path ripple field.

[0037] Specifically, the extraction of stable support moments is based on the stability analysis of the normal force component Fz in the foot-end contact force sequence F(t). The force change rate indicator is used to determine support stability. The variance of the normal force in a continuous time window is calculated. When the force variance is less than a set threshold, the time period is determined to be a stable support stage. The threshold is set to 10% of the overall force sequence variance. The stable support moment is defined as the time midpoint of the stable support stage. Multi-leg coordination analysis identifies the synchronous support mode of robots in the cluster by comparing the stable support moments of the quadruped contact force sequence. Moment clustering uses the K-means algorithm to group the stable support moments of each robot according to their temporal proximity. The cluster distance threshold is set to 1 / 8 of the gait cycle. The cluster stable support moment is defined as the weighted average of the stable support moments of all robots in the same cluster, and the weight is allocated based on the support force of each robot.

[0038] Based on the extracted cluster stable support moments and the natural path network structure, trigger points for robot collaborative actions are set at key nodes along the path. Using the cluster stable support moment as the temporal benchmark and the key nodes of the natural path network as the spatial benchmark, a spatiotemporal coupling trigger mechanism is established. Path network node screening extracts highly connected nodes from the natural path network graph structure, selecting nodes with a degree greater than 1.5 times the average degree as candidate trigger points. Node importance is assessed using the betweenness centrality metric, calculating the frequency of each node in the network's shortest path. Nodes with high betweenness centrality are prioritized as trigger points. Trigger point time allocation allocates the cluster stable support moment to candidate nodes based on their importance weights, with the weights being normalized based on the node betweenness centrality. Trigger delay takes into account inter-robot communication delay and response time. A base delay of 50ms is calculated based on the communication propagation time and the inter-node distance. Collaborative trigger point parameters include spatial coordinates, trigger moment, influence radius, and trigger strength. The trigger strength is calculated based on a weighted combination of path convergence and support quality. The multi-level trigger mechanism establishes a hierarchical structure of main trigger points and auxiliary trigger points. The main trigger points are responsible for large-scale coordination, and the auxiliary trigger points are responsible for local fine-tuning.

[0039] The path exploration signal release uses a diffusion wave model. Starting at the trigger time t0 of each trigger point, the signal propagates exploration information from the trigger point to the surrounding space. The signal strength function is defined as S(r,t)=I0×exp(-r / R)×sin(2πf(t-t0-r / v)+φ), where I0 is the trigger strength parameter of the trigger point, r is the distance from the trigger point, R is the influence radius parameter of the trigger point, f is the signal frequency based on the importance of the trigger point, φ is the initial phase determined by the spatial orientation of the trigger point in the path network, v=2m / s is the signal propagation speed, and the r / v term represents the propagation delay. Frequency division multiplexing is used for signal frequency allocation. The primary trigger point uses a base frequency of 1Hz, while secondary trigger points use integer multiples of the base frequency based on their importance weights to avoid mutual interference between signals. Signal modulation uses amplitude modulation, encoding path weight information and node attribute information into the signal amplitude. Signal directionality is achieved through the initial phase φ, with the signal phase leading along the primary path and lagging along the secondary path.

[0040] Using the linear superposition principle of multiple signals, the total signal strength at any point in space is the sum of the contributions of the signals at each trigger point. The superposition formula, taking into account propagation delay, is Stotal(x,y,t)=∑Si(ri,t-ri / v). Wave interference analysis identifies constructive and destructive interference regions of the signal. Constructive interference regions indicate areas of good path connectivity, while destructive interference regions indicate areas of poor connectivity. The ripple field amplitude distribution is obtained by calculating the time average of the signal intensity at each point. Regions with high amplitudes indicate the intersection of multiple paths. Phase distribution analysis reveals the temporal characteristics of signal propagation. Regions with consistent phases indicate synchronously arriving signals, while regions with large phase differences indicate the influence of different path lengths. Ripple field gradient calculation identifies the direction of signal intensity change. Regions with large gradients indicate boundary regions where path density changes dramatically. Nodes and antinodes are identified by finding extreme values ​​of the signal amplitude. Antinodes correspond to points of path convergence, while nodes correspond to points of path divergence. Field intensity grading divides the ripple field into strong, medium, and weak regions based on signal strength. Different intensity regions correspond to different path priorities. Phase coherence is calculated as Φc=|∑exp(jθi)| / N, which is used to quantify the degree of phase synchronization of multiple signals. Based on signal propagation and wave interference, a path ripple field is finally formed.

[0041] In some embodiments, constructing a multi-machine collaborative potential well based on the path ripple field includes: reconstructing the field strength of the path ripple field to generate a reconstructed potential field; extracting a collaborative potential energy texture based on the reconstructed potential field; establishing a potential well feature matrix through the collaborative potential energy texture, the potential well feature matrix including depth features, width features, texture features and gradient features; and constructing a multi-machine collaborative potential well based on the potential well feature matrix.

[0042] The path ripple field is reconstructed to generate a reconstructed potential field. Using an inverse transformation method, the signal intensity of the ripple field is converted into a potential energy function. The transformation formula is U(x,y)=-k×ln[|Stotal(x,y)|+ε]-λ×Φc(x,y), where k is the proportionality constant, ε is a small quantity to avoid logarithmic singularities, λ is the phase modulation coefficient, and Φc is the phase consistency. The physical meaning of the reconstructed potential field is that regions of high signal intensity correspond to regions of low potential energy, forming a potential well that attracts the robot; regions of low signal intensity correspond to regions of high potential energy, forming a potential barrier that repels the robot; and regions of high phase consistency further reduce the potential energy, enhancing the cooperative attraction. A Gaussian filter is used for potential field smoothing to eliminate noise and discontinuities during the reconstruction process. The filter kernel size is set to three times the grid spacing. Boundary condition processing sets reflection boundaries at the potential field boundaries to prevent the robot from straying from the intended area. Potential field normalization maps the reconstructed potential energy values ​​to the [0,1] interval for ease of subsequent processing and comparison. Gradient field calculation is used to determine the robot's movement direction, and the gradient direction points to the direction where the potential energy decreases fastest.

[0043] Based on the generated reconstructed potential field distribution features, potential energy texture features reflecting the collaborative motion patterns of the robots are extracted. First, the continuous potential field is discretized into a 256-level grayscale image, and the potential energy values ​​are linearly mapped to grayscale levels. By statistically analyzing the spatial distribution relationships between pixel pairs of different grayscale values ​​in the potential field, a co-occurrence matrix is ​​generated for texture feature extraction. Texture feature parameters include energy (indicating texture uniformity), contrast (indicating texture clarity), correlation (indicating the degree of linear correlation within the texture), and entropy (indicating texture complexity). Directional texture analysis identifies the primary directionality of the potential field by calculating texture features at four directions: 0°, 45°, 90°, and 135°. A local binary pattern operator is used to extract local texture features within the potential field, encoding patterns of potential energy variation in local regions. Texture periodicity detection uses the autocorrelation function to analyze the periodicity of the potential field, with the period length corresponding to the collaborative pace of the robots. Texture roughness is measured by calculating the standard deviation of the potential field gradient. Regions with high roughness indicate areas of drastic path changes. The texture feature vector contains features such as energy, contrast, correlation, entropy, local pattern, directionality, periodicity, and roughness.

[0044] A feature matrix describing the geometric and dynamic properties of the potential well is constructed using the extracted synergistic potential energy texture feature vectors. This potential well feature matrix construction combines texture features with the geometric features of the potential field to form a matrix representation that comprehensively describes the potential well properties. Depth feature extraction determines the potential well depth by analyzing local extreme points in the reconstructed potential field. Depth is defined as the potential energy difference between the bottom of the well and the surrounding potential barriers. The depth feature vector includes average depth, depth variability, distribution skewness, and kurtosis. Width features are determined by analyzing the spatial scale of the potential well. The potential well boundary at different potential energy levels is calculated using the equipotential line tracing method. Width is defined as the maximum span of the boundary. The width feature vector includes the major axis width, minor axis width, major-to-minor axis ratio, and the area enclosed by the well. Gradient features describe the steepness of the potential well boundary and are obtained by calculating the potential energy gradient amplitude at the well boundary. The gradient feature vector includes the maximum gradient, average gradient, gradient variance, and main gradient direction. The feature matrix is ​​organized in a block structure and contains four types of features: depth, width, gradient, and texture. Feature normalization normalizes feature values ​​of different physical dimensions to the same scale.

[0045] By superimposing and coupling potential wells, individual potential wells are combined into a composite potential field suitable for cluster collaboration. Potential well selection involves clustering the feature matrix to identify groups of potential wells with similar characteristics. The most representative potential well within each group is selected as the representative potential well. Potential well weights are assigned based on the importance of each well in the path network and the number of robots allocated, with potential wells corresponding to important path nodes receiving higher weights. The robot allocation strategy uses the Hungarian algorithm to optimally assign robots to different potential wells, with the goal of minimizing the total allocation cost. Inter-potential well coupling is achieved by introducing an interaction term between potential wells. The coupling strength is related to the inter-well distance and the desired degree of coordination between robots. The collaborative potential field function is expressed as Ucoop(x,y)=∑wiUi(x,y)+∑∑CijV(rij), where wi is the weight of the i-th potential well, Ui is the individual potential well function, Cij is the inter-well coupling coefficient, V(rij) is the distance-dependent interaction potential, and rij is the inter-well distance. The multi-machine collaborative potential well finally constructed guides the cluster robots to move collaboratively along the optimized path through the potential energy gradient, realizing cluster navigation control based on terrain adaptability.

[0046] Step S140: Decompose the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well. The dominant task chain is used to pull together a core path skeleton, and the auxiliary task chain is used to form a supporting path network. The core path skeleton and the supporting path network are woven together to form an initial collaborative path.

[0047] Specifically, based on the weight distribution and coupling relationships of potential wells in a multi-machine collaborative potential well, the graph structure of the natural path network is hierarchically divided according to the attraction strength of the potential wells. Dominant task chain identification analyzes the potential well groups with the largest weight wi in the collaborative potential well function Ucoop(x,y)=∑wiUi(x,y)+∑∑CijV(rij), selecting paths passing through these high-weight potential wells as the dominant task chain. Dominant task chain extraction utilizes a shortest path algorithm, using high-weight potential wells as mandatory nodes and calculating the optimal path sequence connecting each potential well. Path importance is scored by combining potential well weights and path lengths using the formula Score=∑(wi / Li), where wi is the weight of the potential wells traversed by the path and Li is the path segment length. Auxiliary task chain identification selects potential wells and connecting paths with lower weights, which primarily serve as diversion and backup routes. The task chain classification standard stipulates that the dominant task chain carries 70% of the robot traffic, while the auxiliary task chain carries 30%. The number of dominant task chains is limited to 20% of the total number of paths to ensure path concentration and efficiency. Auxiliary task chains are distributed around the dominant task chains, forming a supporting and supplementary network. The robot allocation strategy assigns high-priority robots to dominant task chains and low-priority robots to auxiliary task chains.

[0048] Based on the decomposed dominant task chain sequence, a core path skeleton with structural stability is formed through a traction mechanics mechanism. The traction force calculation is based on Hooke's law. The traction force between nodes is F = k(L - L0), where k is the spring stiffness coefficient, L is the current distance, and L0 is the equilibrium distance. The traction force is applied by iteratively updating the node positions: x(t+1) = x(t) + αF / m, where α is the step size coefficient and m is the node virtual mass. The skeleton nodes are selected from the dominant task chains with the highest connectivity and path importance scores as skeleton support points. The number of skeleton support points is set to 30% of the total number of nodes in the dominant task chain. Node stiffness is assigned based on the importance scores of the task chain nodes, with higher-scoring nodes receiving greater stiffness coefficients. The stiffness range is set between 1 and 10 N / m. The equilibrium distance L0 is determined based on the average spacing between nodes in the dominant task chain to ensure the compactness of the skeleton structure. Skeleton stability is determined by calculating the structural eigenfrequency to avoid resonance. The path skeleton geometry uses spline curves to fit the discrete skeleton nodes, generating a smooth path curve. Skeleton branching processing sets bifurcation nodes at the convergence points of multiple paths, supporting multi-directional path extension. The skeleton path width is set to 2.5m based on 70% traffic demand to ensure that the main robot flow can be accommodated.

[0049] Using the decomposed auxiliary task chains, a supporting path network system is constructed around the core path skeleton. The path network is hierarchically divided into three levels of support: first-level support paths directly connect to the dominant task chain, providing primary diversion functionality; second-level support paths connect to the first-level support paths, providing localized path selection; and third-level support paths form the outermost path buffer. Support path density control ensures a balanced distribution of paths at each level. The spacing between first-level support paths is set at 1.5 times the spacing between dominant task chains, second-level support paths at 2 times, and third-level support paths at 3 times. Path network connectivity is achieved by establishing bridges between different levels, with bridge nodes located at key locations at the hierarchy boundaries. Support strength is determined based on the spatial distance and connection density between auxiliary task chains and the dominant task chain. Auxiliary task chains with closer distances and denser connections have higher support strength. Network redundancy ensures that if any supporting path becomes disconnected, other paths can assume its function. Path meshing discretizes continuous supporting paths into grid nodes, facilitating path planning and navigation. The support path width is set in layers according to the 30% flow demand: 1.5m for level one, 1.0m for level two, and 0.5m for level three, in conjunction with the robot allocation of the auxiliary task chain.

[0050] Path weaving uses an interlaced connection mode to establish multi-point connections between the core path skeleton and each layer of the supporting path network. Weaving node selection is based on the principles of spatial distance and functional matching. When the distance between the skeleton node and the supporting node is less than the threshold of 0.8m and the functions are compatible, a connection is established. Multi-layer connections are preferentially established at the skeleton bifurcation nodes. The connection weight distribution adopts the inverse distance principle, and the connection weight w=1 / d 2 , where d is the distance between nodes. Weaving density control ensures a moderate number of connections. The number of connections for each skeleton node is limited to 2-5 to avoid overly complex structures. The weaving density of the dominant task chain area is set to 1.5 times that of the auxiliary task chain area. Weaving strength adjustment is determined according to the importance and frequency of use of the path. Frequently used connections have stronger weaving strength. Path fusion processing uses a smooth transition algorithm at the connection point to ensure the continuity and navigability of the path. The fusion radius is set to half of the node influence range, and path smoothing is performed within this range. The initial collaborative path parameters include path coordinate sequence, node type identification, connection weight matrix, hierarchical structure information and traffic distribution parameters. Based on the fusion processing of the weaving algorithm, the initial collaborative path is finally obtained.

[0051] Step S150 , topological transformation is performed on the initial collaborative path to generate a path elasticity coefficient, rigid nodes and flexible segments are identified based on the path elasticity coefficient, random perturbations are injected into the flexible segments to generate a path adaptive factor, and the path adaptive factor is used to reconstruct the adaptive collaborative path.

[0052] Specifically, the initial collaborative path is topologically transformed to generate path elastic coefficients. Using the flexibility matrix approach from structural mechanics, the path network is considered a structural framework consisting of nodes and connected edges. Node flexibility analysis is determined by calculating the displacement response of each node. The flexibility coefficient of node i is Ci = Δi / Fi, where Δi is the node displacement and Fi is the applied force. Edge flexibility calculation is based on the geometric characteristics of the path segments, with straight path segments exhibiting less flexibility and curved path segments exhibiting greater flexibility. The path segment elastic modulus E is determined based on the connection weight w and is proportional to the weight: E = k × w, where k is the proportionality constant. The section moment of inertia I is calculated based on the path width and the robot's maneuverability, with wider paths exhibiting greater section moments of inertia. The elastic coefficient matrix K is constructed to organize the flexibility information of all nodes and path segments into a system matrix, with the matrix elements Kij representing the displacement response of node i under the applied force of node j. Topological transformations, including node reordering, path segment merging, and redundant connection removal, simplify the path structure and highlight key elastic characteristics. The transformed path elastic coefficient reflects the response ability of the path network to external disturbances and the deformation recovery characteristics.

[0053] Based on the generated path elasticity distribution characteristics, rigid nodes and flexible segments in the path network are identified through threshold segmentation and cluster analysis. Using the elasticity threshold method, nodes are considered rigid when their elasticity is less than a set threshold, set at 50% of the overall mean elasticity. Rigid node characteristics include high connectivity, large sectional moment of inertia, and strong connection weights; these nodes maintain a relatively fixed position during path deformation. Flexible segment identification selects path segments with elasticity greater than the threshold, indicating greater deformability and adaptability. Flexible segments are classified based on elasticity into high-flexibility segments (elasticity > 2× mean), medium-flexibility segments (mean < coefficient < 2× mean), and low-flexibility segments (threshold < coefficient < mean). Node type annotation labels each node in the initial collaborative path as either rigid or flexible, providing basic information for subsequent processing. Rigid node distribution analysis reveals that these nodes are primarily located at path intersections, critical turning points, and high-load areas. Flexible segment connection pattern analysis identifies the connectivity and spatial distribution between flexible segments, forming a flexible segment connectivity graph. The rigid-flexible coupling relationship establishes a constraint relationship between the rigid node and the flexible segment, and the rigid node serves as the boundary constraint point of the flexible segment.

[0054] Inject random disturbances into the flexible segment to generate the path adaptation factor. Based on the Gaussian white noise model, the disturbance signal is expressed as ξ(t)=A×N(0,σ 2 )×sin(2πft+φ), where A is the disturbance amplitude, N(0,σ 2) is zero-mean Gaussian noise, f is the perturbation frequency, and φ is the random phase. The perturbation amplitude A is determined based on the flexible segment type and the path elasticity coefficient at the corresponding location: A=0.5m for highly flexible segments, A=0.3m for moderately flexible segments, and A=0.1m for lowly flexible segments, ensuring that the perturbation intensity matches the deformation capacity. A spectrum separation strategy is used to allocate perturbation frequencies: low-frequency perturbations (0.5-1Hz) are used for highly flexible segments, medium-frequency perturbations (1-2Hz) for moderately flexible segments, and high-frequency perturbations (2-3Hz) for lowly flexible segments. The perturbation injection point is selected at the midpoint of the flexible segment to maximize the perturbation effect. A multi-point perturbation coordination mechanism ensures that the perturbation signals of adjacent flexible segments maintain a certain degree of correlation, avoiding discontinuous path deformation. The perturbation duration is set to 10% of the path planning period, ensuring adaptive effects without compromising overall stability. The path adaptation factor is calculated based on the flexible segment's response to the perturbation: α = Δpath / ξ, where Δpath is the path deformation and ξ is the injected perturbation. The factor normalization process maps the adaptive factors to the [0,1] interval.

[0055] In some embodiments, the use of the path adaptive factor to reconstruct an adaptive collaborative path includes: generating a path deformation control amount based on the path adaptive factor; applying the path deformation control amount to the flexible segment to generate elastic deformation; generating a morphological adjustment parameter based on the elastic deformation and the relative displacement between the rigid node; and performing morphological correction on the initial collaborative path using the morphological adjustment parameter to form an adaptive collaborative path.

[0056] The path deformation control variable is generated based on the generated path adaptive factor distribution. The deformation control variable is calculated using a proportional-derivative controller model: the control variable u = Kp × α + Kd × (dα / dt), where Kp is the proportional coefficient, Kd is the differential coefficient, α is the path adaptive factor, and dα / dt is the factor change rate. The proportional control component determines the deformation intensity based on the current adaptive factor value, while the differential control component predicts the deformation direction based on the factor change trend. The control variable direction is calculated by analyzing the gradient of the adaptive factor, with the gradient direction indicating the primary direction of path deformation. The control variable phase inherits the perturbation phase φ to maintain temporal continuity of the deformation process. The deformation control variable amplitude is limited to a preset range to prevent excessive deformation from rendering the path infeasible. Multidimensional control variable generation includes three components: x-direction, y-direction, and rotation angle, enabling full-scale path adjustment. The control variable timing schedule adopts a segmented control strategy, dividing the deformation process into three phases: startup, adjustment, and stabilization. The control variable allocation mechanism allocates the total control variable based on the importance and length of flexible segments, with more control variable allocated to important flexible segments. The smoothing process uses a low-pass filter to eliminate high-frequency oscillations to ensure the smoothness of the deformation process.

[0057] The generated path deformation control is applied to the identified flexible segment, driving the path to produce elastic deformation that meets the environmental adaptation requirements. The elastic deformation calculation is based on Hooke's law, strain ε = σ / E_seg, where σ is the stress generated by the control and E_seg is the local elastic modulus of the path segment. The deformation propagation model describes the propagation process of the control within the flexible segment, using the wave equation ∂ 2 u / ∂t 2 =c 2 ∇ 2 u, where u is the displacement field and c is the wave velocity. Boundary conditions set the constraints at the connection between the flexible segment and the rigid node, and the position of the rigid node remains fixed. Deformation constraint checking ensures that the elastic deformation does not violate geometric constraints and physical limitations, such as the minimum radius of curvature and the maximum elongation ratio. Deformation mode recognition determines the main deformation modes of the flexible segment through modal analysis and selects the deformation mode with the minimum energy. Deformation stability analysis ensures that the deformation process does not cause path instability or oscillation. The deformation result record includes the displacement vector of each node, the strain distribution of the path segment, and the total deformation energy.

[0058] Exemplarily, generating the morphological adjustment parameters based on the relative displacement between the elastic deformation and the rigid node includes: performing stress analysis on the relative displacement to obtain a stretching factor and a compression factor; performing expansion processing based on the stretching factor to form an expansion parameter; performing contraction processing based on the compression factor to generate a contraction parameter; and fusing the expansion parameter and the contraction parameter to output the morphological adjustment parameter.

[0059] Stress analysis is performed along the relative displacement direction to determine the tensile and compressive characteristics of the path deformation process. The tensile factor is calculated based on the positive stress component. When the stress is tensile, a tensile effect is generated. The tensile factor λ = σ_tensile / σ_yield, where σ_tensile is the tensile stress and σ_yield is the yield stress. The compressive factor is calculated based on the negative stress component. When the stress is compressive, a compressive effect is generated. The compressive factor μ = |σ_compress| / σ_yield, where σ_compress is the compressive stress. Factor normalization normalizes the tensile and compressive factors to the [0, 1] range to eliminate differences in numerical ranges. Stress distribution analysis identifies the primary regions of tension and compression along the path and plots stress distribution contours. Interpolation is used to calculate the tensile and compressive factors at each point along the path, forming a continuous factor field. Factor thresholds are set to distinguish between significant tension, significant compression, and neutral regions, with thresholds set at 0.3 and 0.7, respectively. Bidirectional factor processing considers complex stress states with simultaneous tension and compression, employing a vector synthesis method.

[0060] Based on the obtained stretch factor distribution, the corresponding path area is expanded to adapt to the stretch deformation requirements. The expansion scale calculation formula is Scale=1+k×λ, where k is the expansion coefficient and λ is the stretch factor. The expansion direction is determined based on the main direction of the tensile stress, and expansion is preferentially performed along the maximum stretch direction. The expansion center selects the center of mass of the stretch area as the expansion reference point to ensure the symmetry of the expansion. The expansion constraint sets the maximum expansion scale limit to prevent the path from being over-expanded and becoming infeasible. The progressive expansion strategy adopts a step-by-step expansion method, dividing the total expansion amount into multiple small steps. The expansion parameter record includes the expansion center coordinates, expansion scale, expansion direction and expansion range. The expansion quality assessment calculates the geometric characteristics of the path after expansion, such as length increment, area change and curvature adjustment. The expansion boundary processing ensures a smooth connection between the expanded area and the non-expanded area.

[0061] The shrinkage processing adopts a geometric shrinkage algorithm to determine the shrinkage ratio according to the size of the compression factor. The shrinkage ratio calculation formula is Scale=1-k×μ, where k is the shrinkage coefficient and μ is the compression factor. The shrinkage direction is determined based on the main direction of the compressive stress, and preferential shrinkage is performed along the direction of maximum compression. The shrinkage center selects the centroid of the compressed area as the shrinkage reference point to maintain the uniformity of the shrinkage. The shrinkage constraint sets a minimum shrinkage ratio limit to prevent the path from being disconnected due to excessive shrinkage. The progressive shrinkage strategy adopts a step-by-step shrinkage method to avoid path discontinuity caused by sudden shrinkage. The shrinkage parameter record includes the coordinates of the shrinkage center, shrinkage ratio, shrinkage direction and shrinkage range. The shrinkage quality assessment calculates the geometric characteristics of the path after shrinkage, such as length reduction, density increase and connection strengthening. The shrinkage boundary processing ensures a smooth transition between the shrinkage area and the non-shrinkage area.

[0062] The resulting expansion and contraction parameters are integrated through a fusion algorithm to output unified morphological adjustment parameters. The fusion formula is P_adjust = w_expand × P_expand + w_contract × P_contract, where w_expand and w_contract are normalized weights, and P_expand and P_contract are expansion and contraction parameters, respectively. The weight allocation principle prioritizes adjustment types with a greater impact on path performance; generally, expansion weights slightly outweigh contraction weights. Parameter conflict resolution addresses situations where both expansion and contraction requirements exist in the same area, using the net effect principle to determine the final adjustment direction. Parameter integrity checks ensure that no parameters are lost or duplicated during the fusion process. Fusion parameter classification labels parameters based on adjustment intensity and type.

[0063] The generated morphology adjustment parameters are used to precisely modify the initial collaborative path, completing the construction of an adaptive collaborative path. The coordinate correction formula is P_new = P_old + ΔP_adjust, where P_old represents the original coordinates and ΔP_adjust represents the coordinate increment calculated based on the adjustment parameters. The correction order is based on the priority of the adjustment parameters, with important adjustments prioritized. Correction constraints ensure that the correction process does not disrupt the path's basic topology and connectivity. Correction accuracy controls the precision of the correction operation, preventing cumulative errors from affecting the final result. Correction verification checks the rationality of each correction step, promptly identifying and correcting errors. Path smoothing smoothes the corrected path, eliminating sharp corners and discontinuities introduced during the correction process. Correction results are recorded, including a comparison of the paths before and after correction, correction statistics, and performance changes. Path attribute updates update path attributes such as length, curvature, and connection weights based on the morphology correction results. This adaptive collaborative path can adjust path morphology in real time based on environmental disturbances and changes in cluster status, ensuring the stability of collaborative navigation for quadruped robot swarms in dynamic terrain.

[0064] Step S160: extract gait phase information from the adaptive collaborative path to form a gait phase matrix, construct a phase synchronization network based on the gait phase matrix and rigid nodes, use the phase synchronization network and potential energy flow vector to align group phases to generate a group resonance frequency, and use the group resonance frequency to drive the formation of a resonant collaborative path.

[0065] In some embodiments, extracting gait phase information from the adaptive collaborative path to form a gait phase matrix includes: performing phase gradient analysis on the adaptive collaborative path to identify phase change peak and valley areas; performing evolution trend analysis on the phase change peak and valley areas to obtain phase drift directions; determining a phase synchronization range based on the phase drift directions; and constructing a gait phase matrix based on the synchronization range.

[0066] Phase gradient analysis is performed on the adaptive collaborative path to identify phase variation peaks and valleys. Phase gradient analysis is based on the path's instantaneous phase function φ(s) and calculates the rate of phase change dφ / ds along the path's arc length. The gradient calculation uses the central difference method to avoid the accumulation of errors associated with numerical differentiation. Phase variation peaks are identified by finding local gradient maxima, corresponding to locations with the most dramatic phase variation. Phase variation valleys are identified by finding local gradient minima, corresponding to locations with the most gradual phase variation. Peak-valley regions are defined with the peak or valley point as the center and extending to the point where the gradient variation rate is zero, serving as the region boundaries. Region importance is assessed based on the gradient amplitude of the peak or valley; the larger the amplitude, the more important the peak-valley region. Peak-valley region classification marks peak regions as fast-changing phase regions and valley regions as slow-changing phase regions. Different types of regions have different synchronization characteristics. Regional connectivity analysis identifies the connections between adjacent peak-valley regions and constructs a topological map of the peak-valley regions. Regional geometric feature extraction includes parameters such as region length, average gradient, and gradient variance, which are used to describe the phase variation characteristics of the region. Spatial distribution statistics analyze the spatial distribution pattern of peak and valley areas in the entire path and identify periodic and non-periodic features.

[0067] Exemplarily, the evolution trend analysis of the phase change peak and valley areas to obtain the phase drift direction includes: constructing an initial phase influence range based on the phase change peak and valley areas; performing an extended analysis on the initial phase influence range to obtain phase diffusion parameters; determining a multi-machine phase propagation chain based on the phase diffusion parameters; and extracting the phase drift direction based on the multi-machine phase propagation chain.

[0068] Based on the identified phase change peak and valley regions, the initial phase influence range of each region is established to analyze the phase propagation characteristics. A circular influence range is established with the center point of the peak and valley region as the center of the circle and the characteristic length of the region as the initial radius. The influence radius calculation is based on the geometric characteristics and type of the peak and valley regions. The radius of the peak region (fast change region) is r0=0.8×√(L×W / π), and the radius of the valley region (slow change region) is r0=1.2×√(L×W / π), where L is the region length and W is the average width of the region. The influence intensity distribution is determined according to the phase gradient amplitude of the peak and valley regions. The peak region with a large gradient has a stronger influence. The range boundary weight is described by a Gaussian attenuation function, and the weight w(r)=exp(-r 2 / 2σ 2 ), where r is the distance from the center and σ is the attenuation parameter. Multi-region coupling analysis calculates the degree of overlap between adjacent influence ranges. The phase effect of overlapping regions is affected by the linear superposition principle. Influence range classification divides influence ranges into excitation (peak region) and suppression (valley region) based on peak and valley types. These different types have different propagation characteristics. In-range node identification records the path nodes within the influence range and establishes a mapping relationship between regions and nodes.

[0069] The initial phase influence range is expanded and analyzed to obtain the phase diffusion parameters. The diffusion equation model is used to describe the propagation law of phase influence over time. The diffusion equation is ∂φ / ∂t=D∇ 2 φ + S, where D is the diffusion coefficient and S is the source term. The diffusion coefficient calculation is based on the geometric characteristics of the path and the robot's motion parameters. Straight paths have larger diffusion coefficients, while curved paths have smaller diffusion coefficients. Diffusion boundary conditions are set with zero gradient at the path boundaries, indicating that phase influence does not cross the path boundaries. The diffusion time step is set to 1 / 10 of the robot's gait period to ensure the time accuracy of the diffusion analysis. Phase diffusion parameters are extracted by analyzing the characteristic parameters of the diffusion process, including the diffusion velocity v = √(2D), the diffusion distance d = v × t, and the diffusion intensity decay rate λ. Multi-step extended simulation performs diffusion calculations over multiple time steps to track the dynamic changes in the phase influence range. Diffusion front identification locates the location where the phase influence intensity reaches a threshold, and the effective influence area is within the front. Diffusion directionality analysis calculates the primary and secondary diffusion directions, with the major axis of the elliptical diffusion pattern being the primary diffusion direction.

[0070] Based on the acquired phase diffusion parameters, a multi-machine phase propagation chain for phase information propagation in a robot cluster is established. Based on the coverage of the phase influence range, key nodes that can connect multiple influence ranges are selected as propagation nodes. The propagation distance threshold is set based on the diffusion distance parameter, so only nodes with a distance less than the diffusion distance can establish direct propagation connections. The propagation strength is calculated based on the diffusion strength and distance decay: the propagation strength I = I0 × exp(-d / λ), where I0 is the initial strength, d is the propagation distance, and λ is the decay constant. The propagation chain topology uses the minimum spanning tree algorithm to ensure that all propagation nodes are connected while minimizing the total propagation cost. Chain branching establishes branch nodes at the bifurcation points of the propagation chain to support multi-directional propagation of phase information. Propagation delay is calculated based on the propagation distance and propagation speed, with the delay τ = d / v, for synchronization control. Propagation reliability is assessed by calculating the connectivity and redundancy of the propagation chain to evaluate its stability. Multi-chain parallelism considers multiple independent propagation chains operating in parallel to improve the coverage and reliability of phase propagation.

[0071] Phase drift directions are extracted based on multi-machine phase propagation chains. Direction vector calculation combines the geometric direction of the propagation chain with the propagation strength to form a direction vector v = I × û, where I is the propagation strength and û is the unit direction vector. Main propagation direction identification determines the primary phase drift direction of the cluster by calculating the weighted average of the direction vectors of all propagation chains. Direction consistency analysis measures the degree of consistency between directions across different propagation chains; high consistency indicates a clear overall drift direction. Direction stability assessment evaluates the stability of the drift direction by analyzing how the propagation direction changes over time. Multimodal direction detection identifies whether there are multiple primary drift directions. Multimodal situations require regional analysis. Direction weight allocation assigns weights to different directions based on the importance and coverage of the propagation chain, with more important directions receiving higher weights. Direction vector field construction expands local direction vectors to the entire path space, forming a continuous direction field.

[0072] Phase synchronization range determination considers the consistency of phase drift directions and the matching of drift velocities. Regions with consistent drift directions and similar velocities have good synchronization potential. Synchronization feasibility assessment calculates the phase difference and drift velocity difference between regions. Synchronization is considered possible when the phase difference is less than π / 2 and the velocity difference is less than a threshold. Synchronization radius calculation is based on the physical model of phase propagation. The synchronization influence radius R = v / ω, where v is the phase propagation velocity and ω is the oscillation frequency. Multi-center synchronization region identification: When a long path may contain multiple independent synchronization centers, a clustering algorithm is used to identify the number and boundaries of synchronization regions. Region overlap processing addresses the issue of overlapping adjacent synchronization regions, using a weighted average method to determine the synchronization parameters of the overlapping regions. Synchronization strength grading divides the synchronization range into strong, moderate, and weak synchronization regions based on the degree of phase consistency. Boundary smoothing smoothes the boundaries of the synchronization range to avoid synchronization abrupt changes caused by hard boundaries.

[0073] The gait phase matrix is ​​constructed using the path nodes within each synchronization range as matrix row and column indices. The matrix dimensions are N×N, where N is the total number of synchronized nodes. Matrix element assignment is based on the phase relationship and synchronization strength between nodes. Nodes within the same synchronization range have stronger matrix element values. Phase correlation calculation utilizes a cyclic correlation function to account for the periodic nature of the phase. The correlation coefficient is Rij = cos(φi - φj), where φi and φj represent the node phases. A distance weighting approach accounts for the effect of spatial distance on phase correlation. The modified correlation coefficient is R'ij = Rij × exp(-dij / λ), where dij is the node distance and λ is the distance decay constant. The matrix block structure partitions the matrix into blocks based on the synchronization range. Each block corresponds to a synchronization region, with larger values ​​within the block and smaller values ​​between blocks. Matrix positive definiteness is ensured by adding diagonal elements or eigenvalue correction. Matrix sparsification removes weakly correlated matrix elements to improve computational efficiency.

[0074] Based on the generated gait phase matrix and rigid nodes, a phase synchronization network was constructed to describe the phase propagation and synchronization mechanisms. The network nodes were selected with the rigid nodes as core nodes. Nodes corresponding to the spatial positions of the rigid nodes were selected in the gait phase matrix. Synchronous nodes were identified based on strong coupling relationships with phase matrix element values ​​greater than a threshold of 0.7. Network edge weights were set based on phase correlation and spatial distance. The weight calculation formula was wij = ρij × exp(-dij / σ), where ρij is the phase correlation coefficient, dij is the inter-node distance, and σ is the distance decay parameter. The synchronization radius was determined by setting the synchronization influence radius R, centered on the rigid node. Nodes within this radius participated in the phase synchronization process. Network topology analysis calculated the degree distribution, clustering coefficient, and average path length to assess the network's synchronization propagation capability. The phase propagation model used the Kuramoto oscillator model to describe the phase synchronization dynamics. The phase evolution equation was dφi / dt = ωi + K∑sin(φj-φi), where ωi is the natural frequency and K is the coupling strength. A synchronization threshold calculation determined the minimum coupling strength Kc required for the network to achieve complete synchronization. The coupling strength is set to K = 1.2Kc to ensure stable synchronization by exceeding the synchronization threshold. Network module identification uses a community detection algorithm to identify the module structure in the phase-synchronized network, and each module has strong synchronization characteristics.

[0075] Using a constructed phase-synchronized network and potential energy flow vectors, group phase alignment is performed to generate a unified group resonant frequency. Potential energy flow vector mapping samples the potential energy flow vector field at the nodes of the phase-synchronized network to obtain the potential energy flow direction and strength at each node. The phase adjustment force is calculated based on the divergence and curl of the potential energy flow vector: Fi = ∇·vi + ∇×vi, where vi is the potential energy flow vector at node i. The phase alignment algorithm uses a consensus protocol, and the node phase update rule is φi(t+1) = φi(t) + α∑wijsin(φj(t) - φi(t)) + βFi, where α is the synchronization gain and β is the potential energy flow gain. The iterative convergence condition is set to stop when the phase change is less than a threshold δ = 0.01, indicating that phase alignment has been achieved. Group resonant frequency extraction is performed by analyzing the phase distribution after alignment and calculating the group average frequency ωgroup = (1 / N)∑dφi / dt. Frequency stability is determined by calculating the frequency variance to evaluate the stability of the resonance frequency. When the variance is less than a threshold, it is considered that stable resonance has been achieved.

[0076] In some embodiments, the use of the group resonance frequency to drive the formation of a resonant collaborative path includes: determining the phase locking relationship of each robot based on the group resonance frequency; generating a synchronization trigger sequence according to the phase locking relationship; driving each robot to produce coordinated movement through the synchronization trigger sequence; and integrating the trajectories generated by the coordinated movement to form a resonant collaborative path.

[0077] Phase-locking relationships are determined using a phase difference allocation strategy, assigning target phases based on the robot's position and importance within the path. The phase of the robot in the main path is set to the reference phase φref = ωgroup × t, while the target phases of other robots are determined based on their spatial relationship to the main path. Phase difference calculation is based on the inter-robot distance and the desired motion coordination pattern, taking into account the direction of the potential energy flow vector. The phase advances π / 6 in the downstream direction and lags π / 6 in the upstream direction. The base phase difference Δφij = 2π × dij / λ, where dij is the inter-robot distance and λ is the spatial wavelength. Locking strength is assigned based on the robot's importance in the collaborative task, with more important robots receiving stronger phase-locking constraints. The locking range is set so that each robot's phase lock only affects its neighboring robots, with the locking radius set to twice the average inter-robot distance. Dynamic locking adjustment allows for real-time adjustments to the phase-locking relationship based on task requirements and environmental changes. Locking stability analysis verifies the stability of the phase-locking system using Lyapunov stability theory.

[0078] Based on the determined phase-locked relationship, a precise synchronization trigger sequence is generated to coordinate the motion timing of each robot. This synchronization trigger sequence generation is based on the group resonant frequency and each robot's target phase, calculating the trigger moment for each robot's motion. The trigger moment calculation formula is ti = (φi - φref) / (2πωgroup) + nT, where φi is the target phase of robot i, φref is the reference phase, T is the resonance period, and n is the number of periods. Sequence periodicity ensures that the trigger sequence has a periodicity consistent with the group resonant frequency, achieving stable coordinated motion. Trigger accuracy control accurately determines the trigger moment to the millisecond level, ensuring high-precision synchronization and coordination. The sequence buffering mechanism precalculates trigger sequences for multiple periods to avoid real-time computation delays that affect synchronization accuracy. Trigger priority setting prioritizes trigger times based on importance when multiple robots have conflicting trigger moments. Sequence adaptive adjustment adjusts subsequent trigger moments based on the actual robot execution to compensate for execution errors.

[0079] A generated synchronized trigger sequence drives each robot to produce coordinated motion according to a predetermined time sequence. Motion commands containing position, velocity, and acceleration are generated based on the trigger moment and the robot's current state. Rigid nodes serve as motion anchors to maintain relative fixed positions, around which the other robots perform coordinated motion. Trajectory interpolation uses spline interpolation to generate smooth motion trajectories between trigger moments, ensuring the continuity of robot motion. Motion synchronization control uses PID controllers to correct deviations between the robots' actual motion and the desired trajectory to maintain motion synchronization. Coordinated motion modes include in-phase motion (all robots move synchronously) and out-of-phase motion (with phase differences between robots), selected based on task requirements. Motion constraint checking ensures that coordinated motion meets the robots' kinematic and dynamic constraints, such as maximum velocity and acceleration limits. Collision avoidance incorporates collision detection and avoidance strategies into coordinated motion to prevent collisions between robots. Motion performance monitoring uses sensor feedback to monitor the robots' actual motion and compare it with the desired motion.

[0080] The coordinated motion trajectories generated by each robot are integrated to form a unified resonant collaborative path. Trajectory integration uses a weighted averaging method, assigning weights to trajectories based on the robot's importance and motion quality. Weighting is based on the robot's mission importance, motion accuracy, and communication quality, with trajectories of important robots receiving greater weight. Trajectory synchronization and alignment aligns the trajectories of different robots based on their timestamps, eliminating the effects of communication delays and clock skew. Trajectory filtering uses a Kalman filter to eliminate noise and outliers in the trajectories, improving trajectory quality. A trajectory fusion algorithm fuses multiple robot trajectories into a single representative trajectory, which reflects the overall motion pattern of the swarm. The fused trajectory is smoothed using a B-spline curve to fit the fused trajectory points, generating a smooth resonant collaborative path. A path continuity check ensures the resonant collaborative path's temporal and spatial continuity, avoiding path breaks or jumps. Path geometric features, including path length, average curvature, and velocity distribution, are calculated to assess path quality. Based on multi-robot trajectory fusion and path generation, a resonant collaborative path is ultimately formed. This resonant collaborative path enables energy-optimal collaborative navigation of a swarm of quadruped robots in complex dynamic terrain through a swarm phase synchronization mechanism.

[0081] Step S170 , generating a final execution instruction based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robots.

[0082] Specifically, the trajectory coordinate sequence and timestamp of the resonant cooperative path are used as the space-time reference. In combination with the three core parameters of the terrain adaptive rhythm, namely, the rhythm frequency fr, the rhythm intensity Ar, and the phase synchronization Φ, an inverse kinematics solution is used to convert the path points in Cartesian space into angular commands in the robot joint space. A quintic polynomial interpolation is then used to generate smooth and continuous joint trajectories between discrete path points. A terrain adaptive modulation mechanism modulates the basic commands in real time based on the terrain adaptive rhythm parameters: fr modulates the robot's gait frequency to adapt to the terrain characteristics, reducing it on rough terrain to improve stability and increasing it on flat terrain to enhance efficiency; Ar modulates the joint drive torque output to match the load requirements of different terrains; and Φ modulates the phase relationship between the four legs to achieve terrain-adaptive gait changes. Multi-machine coordinated command generation establishes a unified execution sequence for the cluster based on the phase-locked relationship of the resonant cooperative path. The specific execution time of each robot is precisely determined by the group's resonant frequency ωgroup and phase difference distribution. Cluster synchronization control utilizes a distributed consensus algorithm to exchange state information in real time to maintain overall coordination. Command distribution utilizes a three-tiered control architecture: a global coordination controller generates macro-coordination strategies based on resonant collaborative paths, a mid-tier path controller generates detailed trajectory commands based on terrain rhythm parameters, and a bottom-tier joint controller converts these commands into motor drive signals for precise control. Execution monitoring uses multi-sensor fusion to acquire real-time robot status, calculate execution errors, and generate compensation commands using an adaptive control algorithm. When the error exceeds a threshold, subsequent command parameters are dynamically adjusted based on terrain changes, ensuring high-precision collaborative inspection operations for quadruped robot swarms in complex and dynamic terrain environments.

[0083] In order to implement a multi-path collaboration method of clustered quadruped inspection robots corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The following is a block diagram of a multi-path coordination device 200 for a clustered quadruped inspection robot provided in an embodiment of the present application. For ease of illustration, only the parts relevant to this embodiment are shown. The multi-path coordination device 200 for a clustered quadruped inspection robot provided in an embodiment of the present application includes: A data acquisition module 201 is configured to acquire gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extract a foot contact force sequence based on the gait dynamics data, perform spectral decomposition on the foot contact force sequence to generate a gait vibration pattern, and identify a terrain adaptability rhythm through the gait vibration pattern; a path generation module 202 for converting the environmental terrain data into a potential energy gradient field, constructing a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, performing flow field analysis on the dynamic potential energy map to generate potential energy flow vectors, and generating a natural path network using the potential energy flow vectors; a collaborative control module 203 configured to extract a stable support moment from the foot-end contact force sequence, set a collaborative trigger point based on the stable support moment and the natural path network, release a path exploration signal at the collaborative trigger point, form a path ripple field through propagation of the path exploration signal, and construct a multi-machine collaborative potential well based on the path ripple field; A path optimization module 204 is configured to decompose the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well, form a core path skeleton by pulling the dominant task chain, form a supporting path network by using the auxiliary task chains, and weave the core path skeleton and the supporting path network to form an initial collaborative path; An adaptive adjustment module 205 is configured to perform a topological transformation on the initial collaborative path to generate a path elasticity coefficient, identify rigid nodes and flexible segments based on the path elasticity coefficient, inject random perturbations into the flexible segments to generate a path adaptive factor, and reconstruct the adaptive collaborative path using the path adaptive factor; a phase synchronization module 206 configured to extract gait phase information from the adaptive collaborative path to form a gait phase matrix, construct a phase synchronization network based on the gait phase matrix and the rigid nodes, utilize the phase synchronization network and the potential energy flow vector to perform group phase alignment to generate a group resonance frequency, and utilize the group resonance frequency to drive the formation of a resonant collaborative path; The execution control module 207 is used to generate a final execution instruction based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robots.

[0084] The aforementioned multi-path coordination device 200 for a clustered quadruped inspection robot can implement a multi-path coordination method for a clustered quadruped inspection robot according to the aforementioned method embodiment. The optional options in the aforementioned method embodiment also apply to this embodiment and are not described in detail here. The remaining contents of the embodiments of this application can be referred to the contents of the aforementioned method embodiment and are not further described in this embodiment.

[0085] 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 when the processor 302 executes the program, the steps of a multi-path collaboration method of a clustered quadruped inspection robot described in the first embodiment of the present invention are implemented.

[0086] The purpose of the above embodiments is to exemplify and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosed content of the present invention, and it does not limit the scope of protection of the present invention.

[0087] The above embodiments are not exhaustive and may include many other embodiments not listed above. Any replacements and improvements made without violating the concept of the present invention are within the scope of protection of the present invention.

Claims

1. A multi-path collaboration method for clustered quadruped inspection robots, characterized in that: include: Acquiring gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extracting a foot-end contact force sequence based on the gait dynamics data, performing spectral decomposition on the foot-end contact force sequence to generate a gait vibration pattern, and identifying a terrain adaptability rhythm through the gait vibration pattern; Converting the environmental terrain data into a potential energy gradient field, constructing a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, performing flow field analysis on the dynamic potential energy map to generate a potential energy flow vector, and generating a natural path network through the potential energy flow vector; Extracting a stable support moment from the foot-end contact force sequence, setting a collaborative trigger point based on the stable support moment and the natural path network, releasing a path exploration signal at the collaborative trigger point, forming a path ripple field through the propagation of the path exploration signal, and constructing a multi-machine collaborative potential well based on the path ripple field; Decomposing the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well, forming a core path skeleton by pulling the dominant task chain, forming a supporting path network by using the auxiliary task chain, and weaving the core path skeleton and the supporting path network to form an initial collaborative path; Performing a topological transformation on the initial collaborative path to generate a path elasticity coefficient, identifying rigid nodes and flexible segments based on the path elasticity coefficient, injecting random perturbations into the flexible segments to generate a path adaptive factor, and reconstructing the adaptive collaborative path using the path adaptive factor; extracting gait phase information from the adaptive collaborative path to form a gait phase matrix, constructing a phase synchronization network based on the gait phase matrix and the rigid nodes, performing group phase alignment using the phase synchronization network and the potential energy flow vector to generate a group resonance frequency, and driving the formation of a resonant collaborative path using the group resonance frequency; The final execution instruction is generated based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robot.

2. The method according to claim 1, characterized in that Generating a natural path network through the potential energy flow vector includes: identifying potential energy convergence points and divergence points based on the potential energy flow vector; Constructing a potential energy transmission channel between the convergence point and the divergence point; Performing branch growth on the potential energy transmission channel to form a tree-like path structure; The tree-like path structures are meshed to form a natural path network.

3. The method according to claim 1, characterized in that The method of constructing a multi-machine cooperative potential well based on the path ripple field includes: Reconstructing the field intensity of the path ripple field to generate a reconstructed potential field; extracting a synergistic potential energy texture based on the reconstructed potential field; Establishing a potential well feature matrix through the synergistic potential energy texture, wherein the potential well feature matrix includes depth features, width features, texture features and gradient features; A multi-machine collaborative potential well is constructed based on the potential well characteristic matrix.

4. The method according to claim 1, wherein The reconstructing the adaptive collaborative path by using the path adaptive factor includes: generating a path deformation control amount based on the path adaptive factor; Applying the path deformation control amount to the flexible segment to generate elastic deformation; generating a morphology adjustment parameter according to the elastic deformation and the relative displacement between the rigid nodes; The initial collaborative path is morphologically modified using the morphological adjustment parameters to form an adaptive collaborative path.

5. The method according to claim 1, wherein The step of extracting gait phase information from the adaptive collaborative path to form a gait phase matrix includes: Performing phase gradient analysis on the adaptive collaborative path to identify phase change peak and valley regions; Performing an evolution trend analysis on the phase change peak and valley regions to obtain a phase drift direction; determining a phase synchronization range based on the phase drift direction; A gait phase matrix is ​​constructed according to the synchronization range.

6. The method according to claim 1, characterized in that The forming of a resonance collaborative path by driving the group resonance frequency includes: Determining a phase locking relationship between the robots based on the group resonance frequency; generating a synchronous trigger sequence according to the phase-locked relationship; Drive each robot to produce coordinated motion through the synchronous trigger sequence; The trajectories generated by the coordinated motion are integrated to form a resonant collaborative path.

7. The method according to claim 4, characterized in that Generating the morphology adjustment parameter according to the elastic deformation and the relative displacement between the rigid nodes includes: Performing stress analysis on the relative displacement to obtain a tensile factor and a compressive factor; Performing an expansion process based on the stretch factor to form an expansion parameter; Performing a shrinkage process based on the compression factor to generate a shrinkage parameter; The expansion parameter and the contraction parameter are fused to output a morphology adjustment parameter.

8. The method according to claim 5, characterized in that The performing evolution trend analysis on the phase change peak and valley regions to obtain the phase drift direction includes: Constructing an initial phase influence range based on the phase change peak and valley regions; Performing an extended analysis on the initial phase influence range to obtain phase diffusion parameters; determining a multi-machine phase propagation chain according to the phase diffusion parameters; The phase drift direction is extracted based on the multi-machine phase propagation chain.

9. A multi-path collaborative device for clustered quadruped inspection robots, characterized in that: include: a data acquisition module for acquiring gait dynamics data and environmental terrain data of each quadruped robot in the cluster, extracting a foot-end contact force sequence based on the gait dynamics data, performing spectral decomposition on the foot-end contact force sequence to generate a gait vibration pattern, and identifying a terrain adaptability rhythm through the gait vibration pattern; a path generation module, configured to convert the environmental terrain data into a potential energy gradient field, construct a dynamic potential energy map based on the potential energy gradient field and the terrain adaptability rhythm, perform flow field analysis on the dynamic potential energy map to generate a potential energy flow vector, and generate a natural path network using the potential energy flow vector; a collaborative control module, configured to extract a stable support moment from the foot-end contact force sequence, set a collaborative trigger point based on the stable support moment and the natural path network, release a path exploration signal at the collaborative trigger point, form a path ripple field through the propagation of the path exploration signal, and construct a multi-machine collaborative potential well based on the path ripple field; a path optimization module, configured to decompose the natural path network into a dominant task chain and an auxiliary task chain according to the multi-machine collaborative potential well, form a core path skeleton by pulling the dominant task chain, form a supporting path network by using the auxiliary task chains, and weave the core path skeleton and the supporting path network to form an initial collaborative path; an adaptive adjustment module, configured to perform a topological transformation on the initial collaborative path to generate a path elasticity coefficient, identify rigid nodes and flexible segments based on the path elasticity coefficient, inject random perturbations into the flexible segments to generate a path adaptive factor, and reconstruct the adaptive collaborative path using the path adaptive factor; a phase synchronization module, configured to extract gait phase information from the adaptive collaborative path to form a gait phase matrix, construct a phase synchronization network based on the gait phase matrix and the rigid nodes, utilize the phase synchronization network and the potential energy flow vector to perform group phase alignment to generate a group resonance frequency, and utilize the group resonance frequency to drive the formation of a resonant collaborative path; The execution control module is used to generate a final execution instruction based on the resonant collaborative path and the terrain adaptability rhythm to complete the multi-path collaboration of the clustered quadruped inspection robot.

10. A computer device, characterized in that: The method comprises a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method according to any one of claims 1 to 8 when executing the computer program.

Images