A quadruped robot dog motion risk assessment method and system

Abstract

The application discloses a kind of quadruped robot dog movement risk assessment method and system, it is related to machine movement assessment technical field.A kind of quadruped robot dog movement risk assessment system, including have: data acquisition module, topological network module, load coupling module, delay instability module, static instability module, risk guide module and comprehensive evaluation module.The application is by load coupling risk assessment result, the delay instability risk assessment result of external force interference and static instability risk assessment result are unified as the risk flow input of different types, risk flow coupling dynamics model is established in combination with risk transmission impedance parameter and capacity parameter, the coupling propagation relationship of different types risk flow in risk flow topological network and its superposition effect to overall movement risk are described, realize the system level description of the coupling mechanism of multiple-source risk under complex working condition.

Description

Technical Field

[0001] This invention relates to the field of machine motion assessment technology, and in particular to a method and system for assessing the motion risk of a quadruped robot dog. Background Technology

[0002] Quadruped robot dogs, as a typical biomimetic mobile robot, possess strong obstacle-crossing capabilities and good adaptability to complex terrains, and are increasingly being applied in scenarios such as inspection, search and rescue, exploration, and special operations. In these applications, the stable operation of a quadruped robot dog relies on the coordinated changes of multiple state variables, including the forces acting on its feet, the joint driving torques, and the robot's posture and motion state. Typically, it is necessary to assess its motion stability based on these state variables and adjust its motion strategy as needed to avoid risks such as posture instability, falls, or structural damage.

[0003] In existing technologies, methods for stability assessment of quadruped robots are mostly based on analysis of single or a few stability indicators such as zero-torque points, center-of-mass projections, capture points, energy indices, or empirical thresholds. They focus more on the dynamic constraints of local joints or the overall center of mass, lacking means to characterize the propagation law of risk between rigid components from the perspective of the overall system structure. Different types of risk factors, such as load changes, unstructured external force disturbances, and changes in support state under low speed or micro-gait, are usually modeled and controlled separately, making it difficult to reflect their coupling effects in time and space. Most methods can only provide the stability margin at a certain moment, making it difficult to predict the delayed instability caused by external forces or load disturbances after a period of evolution. They also lack mechanisms at the system level to actively guide the transfer of risk from weak components to redundant or more dissipative components. Summary of the Invention

[0004] This invention characterizes the rigid components of a quadruped robot dog and their dynamic relationships in a unified model, and represents multiple risks such as load changes, unstructured external force interference, and static support state changes as a risk flow propagating in the network. Based on this, it realizes comprehensive assessment and guidance control of load coupling risk, external force interference delayed instability risk, and static instability risk.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for assessing the movement risks of a quadruped robot dog, comprising:

[0007] Acquire multi-source state data of the quadruped robot dog during its movement, including foot force data, joint driving torque data, and body posture and motion state data;

[0008] Construct a risk flow topology network and calculate the risk transmission impedance and capacity of each edge in the risk flow topology network based on real-time status data;

[0009] The force data at the foot end and the driving torque data of the joint are mapped to the risk sources in the risk flow topology network. The load coupling risk assessment results are obtained by solving the minimum risk cut set in the risk flow topology network.

[0010] Based on the body posture and motion state data, the unstructured external force interference experienced by the quadruped robot dog is identified and injected into the risk flow topology network as a risk source. The time-series propagation characteristics are analyzed based on the network feature values ​​to obtain the delay instability risk assessment results.

[0011] When the robot dog is in a low-speed or micro-gait state, the current support state is obtained and mapped as a virtual stable subgraph. The evolution characteristics of its connectivity and minimum cut capacity are analyzed to obtain the time-related static instability risk assessment results.

[0012] The above three types of risk assessment results are used as different risk flows propagating in the risk flow topology network. A coupled dynamic model of risk flow is established. The risk flow is monitored, and the network topology is reconstructed by actively adjusting the motion strategy or adjusting the edge weights to guide the risk flow to the preset risk pool or to make it propagate in staggered time.

[0013] Based on the state prediction of the risk flow topology network after risk guidance, a comprehensive motion risk assessment result including the evaluation of risk decoupling effect is generated, and closed-loop adjustment is carried out accordingly.

[0014] As a preferred embodiment of the present invention, the risk flow topology network includes a set of nodes formed by the rigid components of the quadruped robot dog, and a set of edges formed by the dynamic relationships between the rigid components. Each edge is associated with a risk transmission impedance parameter and a capacity parameter. The risk transmission impedance parameter is obtained by sensitivity calculation of the state changes of the two nodes connected to the edge within a preset time window. The capacity parameter is obtained by statistically analyzing the peak values ​​of the foot force data and joint driving torque data corresponding to the edge within the preset time window, and combining this with the body posture and motion state of the corresponding node.

[0015] As a preferred technical solution of the present invention, the acquisition of the load coupling risk assessment result includes: mapping the foot-end force data to the node associated with the corresponding support foot in the risk flow topology network according to the corresponding support foot; mapping the joint driving torque data to the edge connecting the corresponding node in the risk flow topology network according to the corresponding joint; determining the risk flow injection intensity of each node and each edge using the foot-end force data and the joint driving torque data; performing flow distribution analysis on the risk flow topology network under the condition of considering the risk transmission impedance parameter and capacity parameter constraints of each edge; constructing a cut set optimization model with the risk flow intensity and capacity occupancy degree of the edges in the cut set as evaluation indicators; identifying the bottleneck transmission path of load propagation between different rigid components by solving the minimum risk cut set in the risk flow topology network; and generating the load coupling risk assessment result based on the capacity margin and risk flow intensity corresponding to the minimum risk cut set.

[0016] As a preferred technical solution of the present invention, the acquisition of the delayed instability risk assessment result includes: performing time series analysis on the body posture and motion state data within a preset time window, calculating the deviation between the body posture and the preset reference posture trajectory and detecting the moment when the posture change rate exceeds a threshold, identifying the action time and location of unstructured external force interference based on the deviation direction and amplitude, injecting the unstructured external force interference as a risk source into the node corresponding to the disturbed rigid component and its adjacent edges in the risk flow topology network; after injecting the risk source, establishing a linear approximate dynamic model describing the propagation of the risk flow in the network based on the current risk flow topology network, and constructing a matrix characterizing the network connection relationship and edge weights from the linear approximate dynamic model, solving for the eigenvalues ​​and corresponding characteristic modes of the matrix; based on the magnitude of the real part of the eigenvalues, the change trend of characteristic mode energy over time, and the degree of modal coupling at different body motion state stages, assessing the cumulative impact of unstructured external force interference on the body posture stability within several subsequent time windows, and obtaining the delayed instability risk assessment result of the external force interference.

[0017] As a preferred technical solution of the present invention, the acquisition of the static instability risk assessment result includes: determining the set of supporting feet currently in the support phase based on foot force data and body posture and motion state data, and constructing the current support polygon based on the projection of the contact position between the supporting feet and the ground on a preset reference plane; extracting the nodes corresponding to the supporting feet and the nodes corresponding to the body center of gravity calculated from the body posture and motion state data from the risk flow topology network, constructing a virtual stable subgraph, and determining the edge weights of each edge in the virtual stable subgraph based on the foot force data to characterize the static support capability of each supporting foot for the body; performing connectivity analysis on the virtual stable subgraph, calculating the minimum cutting capacity from the node corresponding to the body center of gravity to the node corresponding to the boundary of the support polygon under the condition of including the node, and obtaining the static stability margin at the current moment; tracking the changing trend of the minimum cutting capacity in multiple consecutive time windows, and when the minimum cutting capacity continues to decrease and is lower than a preset threshold, the static instability risk is judged to be increased, and a time-related static instability risk assessment result is generated based on the changing trend.

[0018] As a preferred technical solution of the present invention, the coupled dynamics model includes: using the risk state of each node and the risk flow intensity of each edge in the risk flow topology network as state variables, using the load-type risk flow, external force disturbance-type risk flow, and static instability risk flow corresponding to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result as input terms, and combining the risk transmission impedance parameter and capacity parameter to establish the state equation and constraint conditions describing the evolution of the state variables over time, thus forming a coupled dynamics model.

[0019] As a preferred technical solution of the present invention, the monitoring of the risk flow includes: calculating the risk status of each node and the risk flow intensity of each edge in the risk flow topology network in real time based on the coupled dynamics model; comparing the risk flow intensity with the capacity parameter of the corresponding edge; comparing the node risk status with the risk threshold set according to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result; identifying nodes and edges whose risk flow intensity exceeds the corresponding capacity parameter or whose node risk status exceeds the risk threshold, and recording them as target nodes and target edges respectively; and generating control parameters for actively adjusting the motion strategy based on the position of the target nodes and target edges in the risk flow topology network and their corresponding body posture and motion state.

[0020] As a preferred technical solution of the present invention, the guidance of the risk flow includes: adjusting the motion strategy of the quadruped robot dog according to the control parameters, modifying the edge connection relationship between the nodes corresponding to the support feet in the risk flow topology network by changing the landing sequence of the support feet and the support combination, reconstructing the topology of the risk flow topology network, or changing the risk transmission impedance parameter by adjusting the edge weight corresponding to the joint driving torque, so that different types of risk flows preferentially flow to the preset risk sink or are staggered in time in the risk flow topology network.

[0021] As a preferred technical solution of the present invention, the generation of the comprehensive motion risk assessment result includes: using the risk status of each node and the risk flow intensity of each edge in the current risk flow topology network as initial conditions, predicting the state of the risk flow topology network within a preset future time window to obtain the node risk status and edge risk flow intensity at each time step; comparing the predicted load coupling risk assessment result, delay instability risk assessment result, and static instability risk assessment result with the assessment results at the corresponding historical time before risk guidance, calculating the coupling degree change of different types of risk flows in spatial distribution and temporal evolution as an evaluation index of risk decoupling effect; and generating a comprehensive motion risk assessment result containing the current comprehensive risk level and risk evolution trend based on the evaluation index of risk decoupling effect and the comprehensive risk level predicted within the preset future time window.

[0022] A quadruped robot dog motion risk assessment system includes:

[0023] Data acquisition module: Acquires multi-source state data of the quadruped robot dog during its movement;

[0024] Topology Network Module: Constructs risk flow topology networks;

[0025] Load Coupling Module: Obtains load coupling risk assessment results;

[0026] Delay instability module: Obtains delay instability risk assessment results;

[0027] Static instability module: Obtains static instability risk assessment results;

[0028] Risk guidance module: Monitors and guides risk flows;

[0029] Comprehensive assessment module: Generates comprehensive sports risk assessment results and makes closed-loop adjustments accordingly.

[0030] The present invention has the following advantages:

[0031] This invention introduces risk transmission impedance and capacity parameters into the risk flow topology network, and solves for the minimum risk cut set based on foot force data and joint driving torque data. It identifies the bottleneck transmission path of load propagation between different rigid components and quantitatively characterizes the spatial distribution and weak points of load coupling risk. By using body posture and motion state data to identify unstructured external force disturbances, it models external force disturbances as risk source injections in the risk flow topology network, and analyzes the temporal propagation characteristics of risk flow based on network eigenvalues ​​and characteristic modes to predict the cumulative impact of external force disturbances on body posture stability within several subsequent time windows.

[0032] This invention maps the relationship between the current supporting leg and the robot's center of gravity into a virtual stable subgraph in a risk flow topology network when the quadruped robot dog is in a low-speed or micro-gait state. It then analyzes the evolution of the connectivity and minimum cut capacity of this subgraph over time, providing a time-dependent static instability risk assessment result. This result not only reflects whether the current support state is safe, but also reflects the trend of the support structure evolving towards an unstable state.

[0033] This invention unifies the results of load coupling risk assessment, delayed instability risk assessment of external force interference, and static instability risk assessment as inputs for different types of risk flows. It establishes a risk flow coupling dynamic model by combining risk transmission impedance parameters and capacity parameters, which characterizes the coupling and propagation relationship of different types of risk flows in the risk flow topology network and their superposition effect on the overall motion risk, thus realizing a system-level description of the multi-source risk coupling mechanism under complex working conditions.

[0034] This invention monitors the risk status of each node and the risk flow intensity of each edge in the network based on a risk flow coupling dynamics model. It identifies target nodes and target edges whose risk flow intensity is close to or exceeds the capacity parameter and whose node risk status is close to or exceeds the risk threshold. Based on this, it actively adjusts the landing sequence, support phase combination, and edge weights corresponding to the joint driving torque of the quadruped robot dog. This allows the risk flow to preferentially converge to risk sedimentation tank components with larger capacity parameters or higher dissipation capacity, or to be transmitted in a staggered manner in time, avoiding the resonant superposition of risks on weak components.

[0035] This invention predicts the network state within a preset future time window based on a coupled dynamics model of the risk flow topology network after risk flow guidance is completed. The results of the load coupling risk assessment, the delayed instability risk assessment, and the static instability risk assessment after guidance are compared with the historical results before guidance. The change in risk coupling degree is calculated and a comprehensive motion risk assessment result including the comprehensive risk level and risk evolution trend is formed. This result is used to drive the closed-loop adjustment of the quadruped robot dog's motion strategy, thereby constructing a closed-loop control system of assessment-guidance-prediction-reassessment. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the structure of a quadruped robot dog motion risk assessment system used in an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0039] Example 1: A method for assessing the movement risks of a quadruped robot dog, comprising the following steps:

[0040] Step S1: Acquire multi-source state data of the quadruped robot dog during its movement, including foot force data, joint driving torque data, and body posture and motion state data;

[0041] In one embodiment of the present invention, multi-source state data refers to a set of state quantities synchronously collected by different sensors mounted on a quadruped robot dog within the same motion process and the same sampling period. The quadruped robot dog, for example, includes four legs and a main body. Each leg contains multiple rotational joints, the main body is equipped with an inertial measurement unit, force / torque sensors are installed at the ends of each leg, and joint position and driving torque measuring devices are installed at each joint. The multi-source state data is discretized and recorded on the time axis according to a fixed sampling period. Each sampling moment forms a complete data set, which is used for subsequent construction of a risk flow topology network and for risk assessment and risk guidance.

[0042] The foot-end force data is collected by torque sensors installed at the end of each leg. For each leg of the quadruped robot dog, the foot-end force data includes contact force components when the foot contacts the ground, including at least a vertical support force component and a horizontal tangential force component, as well as a torque component reflecting the torsional state of contact with the ground. The foot-end force data also includes a contact state marker for the foot to distinguish between the support phase and the swing phase, for example, by comparing the vertical support force with a preset threshold to determine whether the foot is currently in a support state.

[0043] The joint driving torque data is acquired by the drive device or associated sensor located at each joint. For example, for the hip joint, knee joint, and other rotational joints of each leg, the joint driving torque data includes the actual output torque of each joint at the current sampling moment. The torque data is obtained by converting the motor output current and the motor torque constant, or by measuring it with a built-in torque sensor.

[0044] The aircraft attitude and motion data are acquired by an inertial measurement unit mounted on the main body of the aircraft and an external attitude estimation device. The aircraft attitude data includes the attitude representation of the aircraft relative to the world coordinate system or reference coordinate system, such as roll angle, pitch angle and yaw angle expressed by Euler angles or quaternions; the motion data includes the position, linear velocity and angular velocity of the aircraft's center of mass or base in space, as well as the position of the aircraft's center of mass and its projection position on the support plane calculated from the current attitude and the configuration of each joint.

[0045] Step S2: Construct the risk flow topology network and calculate the risk transmission impedance and capacity of each edge in the risk flow topology network based on real-time status data;

[0046] The risk flow topology network includes a set of nodes corresponding to the rigid components of the quadruped robot dog, and a set of edges corresponding to the dynamic relationships between the rigid components. Each edge is associated with a risk transmission impedance parameter and a capacity parameter. The risk transmission impedance parameter is obtained by sensitivity calculation of the state changes of the two nodes connected to the edge within a preset time window. The capacity parameter is obtained by statistically analyzing the peak values ​​of the foot force data and joint driving torque data corresponding to the edge within the preset time window, and combining them with the body posture and motion state of the corresponding node.

[0047] In one embodiment of the present invention, the risk flow topology network is a directed weighted network model that abstracts the overall mechanical structure and dynamic relationships of a quadruped robot dog. The quadruped robot dog includes a main body, rigid components such as hip links, thigh links, lower leg links, and foot contact parts for each leg. Each type of rigid component corresponds to a node in the risk flow topology network, and all nodes constitute a node set. If there is a direct dynamic transmission path between any two rigid components, such as through joint connections, through stiffness coupling of the body structure, or through the reaction force transmission path generated by the foot contacting the ground, then a directed edge is set between the corresponding two nodes in the risk flow topology network to represent the transmission direction and path of load or external force disturbance between the two nodes. All directed edges constitute an edge set.

[0048] In a risk flow topology network, each edge is associated with a set of parameters characterizing the propagation characteristics of the risk flow: risk transmission impedance and capacity. The risk transmission impedance describes the amplification or attenuation capability of the risk flow between the two nodes connected by the edge under a given load or external disturbance. The capacity parameter describes the maximum risk flow intensity that the edge can withstand while maintaining the body's attitude and motion without instability. In implementation, based on multi-source state data, a set of state variables is selected for each node in the network. These variables may include, for example, the attitude, linear velocity, and angular velocity of the rigid component, as well as comprehensive indices of foot force and joint driving torque near the component. This set of state variables is then used to construct a node state time series over consecutive sampling periods.

[0049] Within a preset time window, such as spanning several control cycles and covering the critical support phase of a gait cycle, the state time series of two nodes connected by the same edge are compared and analyzed to obtain the degree of response of the downstream node's state change relative to the upstream node's state change under load changes or external disturbances. This degree of response is used as a sensitivity index between state change quantities. The risk transmission impedance parameter is set based on the signal sensitivity index. When the response amplitude of the downstream node to the disturbance of the upstream node is high, the risk transmission impedance parameter of the corresponding edge is set to a lower value to reflect that the risk propagation attenuation on that edge is weak; when the response amplitude of the downstream node to the disturbance of the upstream node is low, the risk transmission impedance parameter of the corresponding edge is set to a higher value to reflect that the risk propagation attenuation on that edge is strong.

[0050] The capacity parameter is calculated based on the mechanical transmission relationship corresponding to the edge. For two rigid components connected by a joint, the capacity parameter is set according to the peak value of the joint driving torque data corresponding to that joint within a preset time window, combined with the safe operating range of the joint under the current body posture and motion state. For example, when the joint driving torque is close to the joint's rated torque and the body posture is close to the predetermined stability margin boundary, the capacity parameter of that edge is set to a lower value to indicate that the connection can bear a smaller risk flow margin under the current posture. For an edge formed by the foot contact path, the capacity parameter is set according to the peak values ​​of the foot's vertical support force and tangential force within a preset time window, as well as the geometric relationship of the support area under the corresponding body posture and motion state. When a certain supporting foot bears a high proportion of the load and the center of gravity projection is close to the boundary of the support polygon, the capacity parameter of the corresponding edge is reduced accordingly.

[0051] In constructing the risk flow topology network, the structure of the node set and edge set is determined by the mechanical design of the quadruped robot, while the risk transmission impedance parameters and capacity parameters are dynamically updated within a sliding time window based on continuously collected foot force data, joint driving torque data, and body posture and motion state data. Taking the continuous task of the quadruped robot walking along uneven terrain while carrying a fixed load as an example, in each control cycle, the risk transmission impedance parameters and capacity parameters of each edge are updated according to the multi-source state data within the current time window. This ensures that the risk flow topology network reflects both the robot's inherent structure and the current load distribution and motion state in real time.

[0052] Step S3: Map the foot force data and joint driving torque data to risk sources in the risk flow topology network, and obtain the load coupling risk assessment results by solving the minimum risk cut set in the risk flow topology network;

[0053] The acquisition of the load coupling risk assessment results includes: mapping the foot-end force data to the nodes associated with the corresponding supporting foot in the risk flow topology network; mapping the joint driving torque data to the edges connecting the corresponding nodes in the risk flow topology network; determining the risk flow injection intensity of each node and each edge using the foot-end force data and joint driving torque data; performing flow distribution analysis on the risk flow topology network under the constraints of risk transmission impedance parameters and capacity parameters of each edge; constructing a cut set optimization model with the risk flow intensity and capacity occupancy of the edges in the cut set as evaluation indicators; identifying the bottleneck transmission path of load propagation between different rigid components by solving the minimum risk cut set in the risk flow topology network; and generating the load coupling risk assessment results based on the capacity margin and risk flow intensity corresponding to the minimum risk cut set.

[0054] In one embodiment of the present invention, a risk source refers to the location and intensity of risk flow injection based on actual mechanical load. Risk sources act on the risk flow topology network in the form of nodal risk sources and edge risk sources. Nodal risk sources describe the load injection at the contact point between the supporting foot and the ground, while edge risk sources describe the load transfer injection between rigid components at joints. For a quadruped robot dog performing a continuous task of carrying a fixed load along uneven terrain, at any sampling moment, the set of supporting feet in the supporting phase is first determined based on the force data at the foot tip. A mapping relationship is established between each foot contact point in the supporting foot set and the corresponding rigid component node in the risk flow topology network. The combined magnitude of the vertical support force and tangential force at the foot tip is mapped to the nodal risk source injection intensity of that node, representing the load risk introduced to the overall structure at that location.

[0055] For joint driving torque data, for each joint, based on its position in the robot's mechanical structure and its connected upstream and downstream rigid components, the corresponding directed edge in the risk flow topology network is found. The magnitude of the driving torque of that joint at the current sampling moment is mapped to the edge risk source injection intensity. The closer the driving torque is to the joint's rated working upper limit, the greater the corresponding edge risk source injection intensity, indicating that the connection path where that joint is located bears a higher risk level under the current load conditions. Foot force data and joint driving torque data are spatially mapped to specific nodes and edges in the risk flow topology network, forming a risk flow injection distribution that evolves with gait changes over time.

[0056] Before constructing the cutset optimization model, a flow distribution analysis is performed on the risk flow topology network. Flow distribution analysis refers to determining the distribution of risk flow intensity propagating along each edge in the network, given the risk source injection intensity of nodes and edges, and the risk transmission impedance parameters and capacity parameters of each edge. Specifically, the propagation of risk flow in the network must satisfy two types of constraints: first, the risk conservation constraint, that is, for ordinary nodes other than risk source injection nodes, the sum of the risk flows from all incoming edges connected to that node and the risk flows from all outgoing edges is balanced; second, the capacity constraint, that is, the risk flow intensity on each edge must not exceed the capacity parameter of that edge. At the same time, the risk transmission impedance parameter on the edge is used to reflect the "cost" of propagating a unit risk flow on different paths, and in the flow distribution analysis, it is used to guide risk flow to propagate more tending to propagate through paths with lower risk transmission impedance.

[0057] Under the above constraints, a cut-set optimization model is constructed. A cut set refers to dividing the set of nodes in a risky flow topology network into two disjoint subsets, and the set of all edges pointing from one subset to the other is taken as the edge set corresponding to the cut set. The minimum risk cut set is the set of edges in the cut set that results in the most strained overall capacity parameter occupancy among all possible cut sets. To reflect the magnitude of load coupling risk, the cut-set optimization model uses the ratio of risk flow intensity to capacity parameter of each edge in the cut set, and the risk amplification of this ratio at downstream nodes, as evaluation indicators. This is used to measure the transmission bottleneck position in the network that first reaches risk saturation when the load on a certain path continues to increase. By solving this cut-set optimization model, the edge set corresponding to the minimum risk cut set, along with its risk flow intensity and capacity margin, characterizes the weakest transmission path when the load propagates between different rigid components.

[0058] In the same walking task, as the quadruped robot dog traverses terrain areas with varying slopes and roughness, the force distribution at its feet and the distribution of joint driving torques change, correspondingly adjusting the nodal and edge risk sources. By repeatedly executing S3 within various time windows, a series of time-evolving minimum risk cut sets can be obtained, along with curves showing the changes in capacity margin and risk flow intensity of edges within each cut set. The load coupling risk assessment results are generated based on the statistical characteristics of these minimum risk cut sets. For example, the edge with the smallest capacity margin in the minimum risk cut set within a certain time window is marked, and its corresponding rigid component and adjacent components are identified as areas with high load coupling risk. If a transmission path consistently appears within a minimum risk cut set across multiple consecutive time windows, the structure corresponding to that path is considered a bottleneck path with significant long-term load coupling risk.

[0059] Step S4: Based on the body posture and motion state data, identify the unstructured external force interference experienced by the quadruped robot dog, inject it as a risk source into the risk flow topology network, and analyze its temporal propagation characteristics based on the network feature values ​​to obtain the delay instability risk assessment results.

[0060] The acquisition of the delayed instability risk assessment results includes: performing time series analysis on the body posture and motion state data within a preset time window; calculating the deviation between the body posture and the preset reference posture trajectory and detecting the moment when the posture change rate exceeds a threshold; identifying the time and location of action of unstructured external force interference based on the deviation direction and amplitude; injecting unstructured external force interference as a risk source into the node corresponding to the disturbed rigid component and its adjacent edges in the risk flow topology network; after injecting the risk source, establishing a linear approximate dynamic model describing the propagation of the risk flow in the network based on the current risk flow topology network; constructing a matrix representing the network connection relationship and edge weights from the linear approximate dynamic model; solving for the eigenvalues ​​and corresponding characteristic modes of the matrix; and evaluating the cumulative impact of unstructured external force interference on the body posture stability within several subsequent time windows based on the magnitude of the real part of the eigenvalues, the trend of characteristic mode energy change over time, and the degree of modal coupling at different body motion state stages, thereby obtaining the delayed instability risk assessment results of the external force interference.

[0061] In one embodiment of the present invention, the unstructured external force disturbance refers to external forces outside the scope of the quadruped robot dog's motion planning and conventional model prediction, such as lateral collisions, sudden wind loads, or instantaneous pushing forces from irregular obstacles in the environment during walking. These external forces do not follow pre-set rules in terms of magnitude, direction, and timing of action; they are not the driving torque output by the robot's own motion control system, nor are they contact reaction forces directly derived from terrain geometry and planned gait. Instead, they are reflected in the robot's posture and motion state data as additional disturbances. By analyzing the time series of the robot's posture and motion state data, abnormal change segments corresponding to these additional disturbances can be separated, providing a basis for subsequently injecting corresponding risk sources into the risk flow topology network.

[0062] Within a preset time window, when performing time-series analysis on the robot's attitude and motion data, a reference attitude trajectory is first determined based on the quadruped robot's motion task and control strategy. This reference attitude trajectory is generated by the gait planning module under ideal terrain conditions and without external interference, reflecting the robot's expected roll angle, pitch angle, yaw angle, and center of mass trajectory within one or more gait cycles. The actual robot attitude and motion data are aligned with the reference attitude trajectory on the same time axis. The deviation between the actual attitude and the reference attitude is calculated for each sampling moment, and the attitude change between adjacent sampling moments is used as an estimate of the attitude change rate.

[0063] Based on the aforementioned time series analysis, the time intervals of unstructured external force interference are identified by setting attitude deviation thresholds and attitude change rate thresholds. When the deviation of the actual attitude from the reference attitude increases rapidly within a short period of time and the attitude change rate exceeds the preset threshold, this time period is determined to be the stage of external force interference. Based on the direction and magnitude of the deviation, combined with the spatial distribution relationship between the aircraft attitude and various rigid components, the main location of the external force is determined, such as acting on the side, back, or area of ​​a certain leg of the aircraft.

[0064] After identifying the location of the external force interference, the interference is represented in the risk flow topology network as an additional risk source applied to the node corresponding to the disturbed rigid component and its adjacent edges. Specifically, if the external force acts on the side of the main body, the node corresponding to the main body is set as the external force risk source node, and an additional edge risk source injection intensity is set on the edge directly connected to the node to represent the influence of the external force on the dynamic coupling path between the main body and each leg. If the external force is more concentrated near a certain leg or foot, corresponding risk source injections are set on the corresponding leg or foot node and the relevant joint edges. This risk source injection method is distinguished from load-type risk sources and is separately labeled as an external force interference risk source to differentiate different risk sources in subsequent coupled dynamics analysis.

[0065] After injecting risk sources, a linear approximate dynamic model is established based on the current risk flow topology to analyze the temporal propagation characteristics of external disturbances in the network. The linear approximate dynamic model is a state equation obtained by linearizing the originally nonlinear risk propagation process near the current operating point, used to characterize the propagation law of risk flow in the network under small disturbance conditions. In constructing the linear approximate dynamic model, the risk state of each node and the risk flow intensity of each edge in the network are used as state variables, and the injection intensity of external disturbance risk sources and the change in basic risk flow caused by load distribution are used as input terms. Combined with risk transmission impedance parameters and capacity parameters, a set of linear difference relationships describing the changes of state variables over time is obtained.

[0066] Based on the aforementioned linear approximation dynamic model, a matrix is ​​constructed to characterize network connectivity and edge weights. This matrix comprehensively reflects the limiting effect of the connection topology between nodes, the risk transmission impedance parameters of each edge, and the capacity parameters on risk flow propagation. By performing eigenvalue decomposition on this matrix, several eigenvalues ​​and corresponding characteristic modes are obtained. Eigenvalues ​​are used to characterize the decay or growth rate of different risk propagation modes over time. The closer the real part of the eigenvalue is to zero or the larger it is, the slower the corresponding mode decays over time or the more it tends to amplify. Characteristic modes describe the spatial distribution pattern of the risk state of each node and the risk flow intensity of each edge in the network under that mode.

[0067] Within several preset time windows following the occurrence of external disturbance, the energy changes of each characteristic mode over time are calculated based on the evolution law of the linear approximate dynamic model. This is combined with the current motion state of the quadruped robot dog (e.g., acceleration, constant speed, or deceleration) to assess the coupling degree between different characteristic modes. When the energy of certain characteristic modes related to the robot's attitude stability continuously increases after the external force is applied, and these modes exhibit strong coupling relationships at specific motion states, it indicates that the risk of instability due to the external disturbance increases significantly after a period of time.

[0068] By comprehensively analyzing the real part of the eigenvalues, the growth or decay trend of the characteristic mode energy over time, and the modal coupling strength, the delayed instability risk assessment results provide the cumulative impact of external disturbances on the body's attitude stability over several future time windows. For example, when a quadruped robot dog is carrying a load and encounters a lateral push, the delayed instability risk assessment results not only indicate whether the current attitude deviation is within an acceptable range, but also whether the risk state of certain key rigid components tends to saturate or reach the instability boundary in the subsequent gait cycles.

[0069] Step S5: When the robot dog is in a low-speed or micro-gait state, obtain the current support state and map it as a virtual stable subgraph, analyze the evolution characteristics of its connectivity and minimum cut capacity, and obtain the time-related static instability risk assessment results.

[0070] The acquisition of the static instability risk assessment results includes: determining the set of supporting feet currently in the support phase based on foot force data and body posture and motion state data, and constructing the current support polygon based on the projection of the contact position between the supporting feet and the ground onto a preset reference plane; extracting nodes corresponding to the supporting feet and nodes corresponding to the body center of gravity calculated from the risk flow topology network, constructing a virtual stable subgraph, and determining the edge weights of each edge in the virtual stable subgraph based on foot force data to characterize the static support capacity of each supporting foot for the body; performing connectivity analysis on the virtual stable subgraph, calculating the minimum cutting capacity from the node corresponding to the body center of gravity to the node corresponding to the boundary of the support polygon under the condition of including the node, and obtaining the static stability margin at the current moment; tracking the changing trend of the minimum cutting capacity within multiple consecutive time windows, and when the minimum cutting capacity continuously decreases and falls below a preset threshold, the static instability risk is judged to be increased, and a time-related static instability risk assessment result is generated based on the changing trend.

[0071] In one embodiment of the present invention, low speed or micro-gait refers to the movement phase in which the quadruped robot dog is near rest or walking slowly with a small stride. For example, when performing tasks such as fine observation, traversing narrow passages, or starting and stopping transitions, the linear velocity of the robot body is lower than a preset speed threshold and the gait cycle is prolonged. Under such conditions, the system is more concerned with the static or quasi-static equilibrium state of the robot body on the support structure, rather than the dynamic stability during high-speed running. Therefore, it is necessary to abstract and analyze the support structure separately on the risk flow topology network.

[0072] To determine the current support state, the force data at the foot ends is used to identify the support phase and swing phase of each leg. For each foot end, based on the relationship between its vertical support force and a preset contact threshold, feet with a vertical support force greater than the threshold and a duration exceeding the minimum contact duration are marked as being in the support phase, while feet with a vertical support force close to zero or significantly lower than the threshold are marked as being in the swing phase. All feet in the support phase constitute the current support foot set. Combining the spatial positions of the feet in the reference coordinate system recorded in the body posture and motion data, the projected coordinates of the support feet on a preset reference plane are extracted and sorted according to their geometric relationships on the plane, forming a closed polygon. This polygon is the current support polygon, used to describe the geometric range of the body's contact support area with the ground.

[0073] The virtual stabilization subgraph is a subnetwork constructed within the risk flow topology network for static support structures, specifically used to analyze the static support relationship between the robot's center of gravity and the supporting feet. Nodes in the virtual stabilization subgraph are selected from the node set of the risk flow topology network, including all nodes corresponding to the current supporting foot and the node corresponding to the robot's center of gravity calculated from the robot's attitude and motion state data. The position of the center of gravity in the robot's coordinate system is calculated using the current robot attitude, robot geometric parameters, and joint configuration, and then mapped to its spatial position in the reference coordinate system through coordinate transformation. Edges in the virtual stabilization subgraph represent load transfer paths between the supporting feet and the robot's center of gravity, or between supporting feet, under static conditions; its topology is inherited from the support-related parts of the risk flow topology network.

[0074] The edge weights in the virtual stability subgraph characterize the static support capability of each support foot for the aircraft. The edge weights are set based on the force data at the foot tip and the support geometry. For example, for an edge connecting a support foot node to the corresponding node of the aircraft's center of gravity, its weight is proportional to the vertical support force borne by that support foot at the current moment, and is also adjusted based on the positional relationship of the support foot within the support polygon. When a support foot bears a high proportion of the load and is close to the projected position of the aircraft's center of gravity, the corresponding edge weight is set to a higher value to reflect the strong static support effect of that support foot on the aircraft's center of gravity; when a support foot bears a low proportion of the load or is far from the projected position of the center of gravity, the corresponding edge weight is set to a lower value to reflect its weaker contribution to overall static stability. The edge weights between support feet are set based on the relative position of the foot tips and the load distribution, used to characterize the load redistribution path within the support structure.

[0075] After constructing a virtual stable subgraph containing support foot nodes and the aircraft's center of gravity node, connectivity analysis is performed on this subgraph to determine whether the aircraft's center of gravity is within a safe zone inside the support polygon under the current support structure and through which paths it transfers its weight to the support foot. A core metric for connectivity analysis is the set of paths in the virtual stable subgraph leading from the node corresponding to the aircraft's center of gravity to the node corresponding to the boundary of the support polygon. By constructing cut sets on the virtual stable subgraph, one side containing the node corresponding to the aircraft's center of gravity is separated from the other side containing the node corresponding to the boundary of the support polygon. The comprehensive cost in terms of static support capacity is calculated for the set of edges that need to be "cut" to sever the connectivity between these two sides. This comprehensive cost, known as the minimum cut capacity, is used to quantify the magnitude of the static stability margin of the aircraft's center of gravity.

[0076] A larger minimum cut capacity indicates that more static support capacity needs to be "weakened" under the current support condition to cut off the effective support path between the machine's center of gravity and the support foot. The farther the machine is from the static instability boundary, the higher the static stability margin. Conversely, a smaller minimum cut capacity indicates that there are fewer or weaker effective support paths between the machine's center of gravity and the support foot under the current support structure. Once external disturbances or load changes continue to act, the machine will be closer to the static instability boundary. By calculating the minimum cut capacity of the virtual stable subgraph at a sampling moment, the static stability margin at that moment can be obtained.

[0077] The minimum cut capacity of the virtual stable subgraph is tracked within multiple consecutive time windows. Specifically, as the quadruped robot dog slowly moves along uneven terrain or adjusts its posture in a narrow area with a low-speed gait, the steps of supporting foot determination, supporting polygon construction, virtual stable subgraph construction, and minimum cut capacity calculation are repeatedly executed within each preset time window to obtain a minimum cut capacity sequence that changes over time. When this sequence shows a continuous decreasing trend over a continuous period of time, and drops to near or below a preset threshold at a certain moment, it indicates that the static stability margin of the robot's center of gravity is gradually weakening and approaching the instability boundary. At this point, the static instability risk is judged to be increased, and the evolution trend of the static instability risk in the future is assessed based on the slope and fluctuation characteristics of the minimum cut capacity change.

[0078] The static instability risk assessment results are presented in a combination of quantitative indicators and qualitative levels. On one hand, they provide the current static stability margin value and its positional relationship relative to the threshold; on the other hand, they provide the static instability risk level and risk change trend, such as being labeled "stable," "decreasing stability margin," or "high static instability risk." For quadrupedal robotic dogs operating under special conditions such as heavy loads, steep terrain slopes, or a reduced number of supporting legs, the static instability risk assessment results provide static equilibrium constraints for subsequent risk flow guidance and motion strategy adjustments.

[0079] Step S6: Take the above three types of risk assessment results as different risk flows propagating in the risk flow topology network, and establish a coupled dynamic model of the risk flow; monitor the risk flow, and guide the risk flow to the preset risk pool or make it staggered in time by actively adjusting the motion strategy to reconstruct the network topology or adjusting the edge weights.

[0080] The coupled dynamics model includes: using the risk state of each node and the risk flow intensity of each edge in the risk flow topology network as state variables, and using the load-type risk flow, external force disturbance-type risk flow, and static instability risk flow corresponding to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result as input terms, and combining the risk transmission impedance parameter and capacity parameter to establish the state equation and constraints describing the evolution of the state variables over time, thus forming the coupled dynamics model.

[0081] The monitoring of the risk flow includes: calculating the risk state of each node and the risk flow intensity of each edge in the risk flow topology network in real time based on the coupled dynamics model; comparing the risk flow intensity with the capacity parameter of the corresponding edge; comparing the node risk state with the risk threshold set according to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result; identifying nodes and edges whose risk flow intensity exceeds the corresponding capacity parameter or whose node risk state exceeds the risk threshold, and recording them as target nodes and target edges respectively; and generating control parameters for actively adjusting the motion strategy based on the position of the target nodes and target edges in the risk flow topology network and their corresponding body posture and motion state.

[0082] The guidance of the risk flow includes: adjusting the movement strategy of the quadruped robot dog according to the control parameters; modifying the edge connection relationship between the nodes corresponding to the support feet in the risk flow topology network by changing the landing sequence of the support feet and the support combination; reconstructing the topology of the risk flow topology network; or changing the risk transmission impedance parameter by adjusting the edge weight corresponding to the joint driving torque, so that different types of risk flows are preferentially directed to the preset risk sink or staggered in time in the risk flow topology network.

[0083] In one embodiment of the present invention, the load coupling risk assessment results, delay instability risk assessment results, and static instability risk assessment results are uniformly represented on a unified risk flow topology network. Therefore, these three types of assessment results are respectively classified as load-related risk flows, external disturbance-related risk flows, and static instability-related risk flows. Within the same time window, combining the capacity margin and risk flow intensity of each bottleneck path in the aforementioned load coupling risk assessment results, the characteristic mode energy related to attitude stability in the delay instability risk assessment results, and the changing trend of the minimum cut capacity in the static instability risk assessment results, the injection intensity of different types of risk flows applicable to each node and each edge is determined. In the risk flow topology network, the risk state of each edge and each node not only reflects the influence of a single risk source but is also simultaneously affected by the spatial and temporal superposition of the three types of risk flows.

[0084] The coupled dynamics model is a unified description of the propagation of various types of risk flows over time in a risk flow topology network. This model uses the risk states of all nodes and the risk flow intensities of all edges in the risk flow topology network as state variables to characterize the risk distribution of the entire network under current load distribution, external disturbances, and support structure conditions. At each discrete time step, the input of load-related risk flows is determined by the load coupling risk assessment results within the current time window. For example, based on the capacity margin of each edge in the minimum risk cut set, load-related risk flows are injected into nodes and edges related to bottleneck paths. The input of external disturbance-related risk flows is determined by the delayed instability risk assessment results. For example, based on the predicted values ​​of attitude-related characteristic mode energy within several gait cycles after the external disturbance, external disturbance-related risk flows are injected into nodes corresponding to disturbed rigid components and edges strongly coupled with attitude modes. The input of static instability-related risk flows is determined by the static instability risk assessment results. For example, based on the degree to which the minimum cut capacity approaches or falls below the threshold, static instability-related risk flows are injected into nodes corresponding to the machine's center of gravity and support feet, and the edges between them.

[0085] When establishing the state equations, the coupled dynamics model updates the node risk state and edge risk flow intensity at each discrete time step based on the current state variables, the three types of risk flow inputs, and the risk propagation impedance and capacity parameters. The risk propagation impedance parameter describes the impact intensity of unit risk flow propagation on downstream nodes on different paths, while the capacity parameter limits the upper bound of risk flow intensity on each edge. This ensures that the model reflects not only the topological path of risk propagation but also the "smoothness" and "capacity limit" of different paths in risk propagation. Through the recursion of discrete time steps, the coupled dynamics model continuously characterizes the dynamic coupled propagation process of the three types of risk flows in the network when the quadruped robot dog is walking with a load, experiencing external force interference, and in a low-speed support phase during the same walking task.

[0086] In each control cycle, the coupled dynamics model is invoked to calculate the risk status of all nodes and the risk flow intensity of each edge in the network at the current time step. For each edge, the calculated risk flow intensity is compared with the dynamically updated capacity parameter of that edge. If the risk flow intensity of an edge is close to or exceeds the corresponding capacity parameter, that edge is judged to have a tendency to be overloaded. For each node, the risk status of that node is compared with a risk threshold set based on the current load coupling risk assessment results, delay instability risk assessment results, and static instability risk assessment results. The risk threshold is preset or adaptively adjusted based on factors such as the structural importance of the rigid component represented by the node, its location, acceptable attitude deviations, joint load limits, and static stability margins. When the risk status of a node exceeds the corresponding risk threshold, it indicates that the rigid component is in a high-risk state.

[0087] In the above comparison process, all edges whose risk flow intensity exceeds the corresponding capacity parameter are marked as target edges, and all nodes whose risk state exceeds the corresponding risk threshold are marked as target nodes. Target nodes and target edges centrally represent the key locations where risks are concentrated or overflow in the network under the current load coupling status, external force disturbance propagation status, and static support structure constraints. Based on the positional relationship of target nodes and target edges in the risk flow topology network, and their physical meaning in the quadruped robot dog structure (e.g., located in the main body, the thigh link of a leg, or a foot contact point), and combined with the current robot body attitude and motion state data, control parameters for actively adjusting the motion strategy are generated.

[0088] Control parameters are a set of structured adjustment instructions for the motion control output of a quadruped robot dog. These parameters are used to change the landing timing of the supporting leg, the combination of supporting phases, and the joint output form in subsequent gait cycles, thereby physically altering the topology or edge weight settings of the risk flow network. Control parameters include the expected landing time of each leg in the next one or more gait cycles, the duty cycle of the swing phase and the supporting phase, the sequence of single or multi-leg coordinated support, and the expected driving torque adjustment ratio or equivalent stiffness adjustment ratio corresponding to each joint. For example, when the target edge is concentrated in the connection path between a certain leg and the main body, the control parameters instruct to increase the supporting time of the opposite leg, advance the landing time of the opposite leg, or reduce the upper limit of the driving torque of some joints of that leg in the next few steps, thereby reducing the risk flow intensity on the original bottleneck path.

[0089] During risk flow guidance, the movement strategy is adjusted according to control parameters, mainly including two types of operations. One type of operation directly modifies the edge connections between nodes corresponding to the supporting feet in the risk flow topology network by changing the landing sequence and support combinations of the supporting feet, thereby reconstructing the network's topology. For example, when a quadruped robot dog traverses uneven terrain and there is a high risk of static instability near a certain hind foot, the support duration of the forefoot and the opposite foot is increased in subsequent gait cycles by controlling parameters, while the independent support time of the high-risk hind foot is shortened. In some gait phases, a three-legged support pattern may even be used instead of the original diagonal support pattern, changing the connection relationships of nodes and edges representing different support combinations in the risk flow topology network, thereby increasing the chances of risk flow diversion on other paths at the topological level.

[0090] Another type of operation involves adjusting the risk transmission impedance parameter by modifying the edge weights corresponding to the joint driving torque. Edge weight adjustment is reflected in the control strategy related to specific joints. For example, by reducing the stiffness setting of certain joints or reducing their output torque, the risk transmission impedance parameter of the edge corresponding to that joint is increased, thereby suppressing the transmission of risk flow through that path in the coupled dynamics model. Conversely, for edges connected to rigid components with redundant structures or high dissipation capabilities, the risk transmission impedance parameter is reduced by appropriately increasing the allowable torque output or enhancing the damping characteristics, thereby forming a "low-resistance channel" in the model through which risk flow is more easily passed.

[0091] A risk settling basin is a pre-selected or dynamically identified risk absorption area in the risk flow topology network, corresponding to rigid components with large capacity parameters and high risk dissipation capabilities, and their adjacent connecting paths. For example, certain installation locations with buffer structures inside the main body, specially designed vibration damping modules, or the locations of non-critical components serve as combinations of risk settling basin nodes and edges.

[0092] In a continuous mission where a quadruped robot dog carries a fixed load across undulating terrain and experiences a lateral force push, the aforementioned coupled dynamics modeling, risk monitoring, and risk guidance operations are continuously executed within each control cycle. When load coupling analysis shows that the connection path between a certain leg and the body is under high load for a long period, delayed instability analysis shows that the energy of attitude-related characteristic modes has an increasing trend in the future gait cycles, and static instability analysis shows that the static stability margin within the support polygon continues to decrease, the control parameters will focus on adjusting the support combination and joint control strategies that are simultaneously related to these risks. This will guide the risk flow from the weak path to the path where the risk sink is located, and stagger the peak values ​​of different risk sources in time.

[0093] Step S7: Based on the state prediction of the risk flow topology network after risk guidance, generate a comprehensive motion risk assessment result including the risk decoupling effect evaluation, and make closed-loop adjustments accordingly.

[0094] The generation of the comprehensive motion risk assessment result includes: using the risk status of each node and the risk flow intensity of each edge in the current risk flow topology network as initial conditions, predicting the state of the risk flow topology network within a preset future time window to obtain the node risk status and edge risk flow intensity at each time step; comparing the predicted load coupling risk assessment result, delay instability risk assessment result, and static instability risk assessment result with the assessment results at the corresponding historical moments before risk guidance, calculating the changes in coupling degree of different types of risk flows in spatial distribution and temporal evolution as an evaluation index of risk decoupling effect; and generating a comprehensive motion risk assessment result containing the current comprehensive risk level and risk evolution trend based on the evaluation index of risk decoupling effect and the comprehensive risk level predicted within the preset future time window.

[0095] In one embodiment of the present invention, the current state of the risk flow topology network is given by the network state after risk flow guidance is completed. This state includes the risk state of each node in the risk flow topology network and the risk flow intensity of each edge, reflecting the comprehensive distribution of risks caused by load, external interference, and static support structures in the network after guidance and redistribution. To evaluate the sustained effect of risk guidance measures over a future period, it is necessary to predict the state of the risk flow topology network within a preset future time window based on this prediction.

[0096] A preset future time window is used to describe the movement process of the quadruped robot dog within several gait cycles after the current control cycle. The length of this time window is set according to task requirements, gait cycles, and control frequency, for example, selecting a time range covering several subsequent foot-landing and foot-lifting processes. Within this time window, based on the coupled dynamics model and the current node risk state and edge risk flow intensity as initial conditions, the network state corresponding to each future time step is calculated recursively through discrete time steps, i.e., the predicted values ​​of the node risk state and edge risk flow intensity at each time step.

[0097] After obtaining the predicted network state at each time step within the future time window, feature information related to the three types of risk assessment results is extracted. Specifically, based on the relationship between the risk flow intensity and capacity parameters on the bottleneck path in each future time step, a corresponding load coupling risk assessment result prediction sequence is constructed; based on the risk state changes of nodes related to attitude stability (such as main body nodes and key connecting component nodes) and the energy change trend of delayed instability-related characteristic modes in S4, a delayed instability risk assessment result prediction sequence for external force disturbance is constructed; based on the risk state of nodes corresponding to the center of gravity and support feet in the virtual stability subgraph and the risk flow intensity changes of related edges, a static instability risk assessment result prediction sequence is constructed. After aligning the above three types of prediction results with the corresponding historical assessment results of S3, S4, and S5 before risk guidance in time, they are used to compare and analyze the differences in risk distribution and evolution patterns before and after risk guidance.

[0098] Risk coupling degree describes the degree of overlap between different types of risk flows in terms of spatial location and temporal evolution. Spatially, risk coupling degree is assessed based on whether the high-risk regions of the three types of risk flows overlap in the risk flow topology network. For example, it is calculated by counting the proportion of nodes or edges that simultaneously belong to the load coupling high-risk region, the delay instability high-risk region, and the static instability high-risk region at the same time step. Temporally, risk coupling degree is assessed based on the degree of overlap of the peak occurrence times of the three types of risk flows on the time axis. For example, it is analyzed by considering the time difference between the peak occurrence of load-related risk flows and the peak occurrence of external disturbance-related risk flows at the same node or edge, and the proportion of events where this time difference falls within a small range. By comparing the spatial and temporal overlap before and after risk guidance, the change in risk coupling degree is obtained, which serves as an evaluation indicator of the risk decoupling effect. A better risk decoupling effect means that high-intensity risk flows are more dispersed spatially and more staggered temporally.

[0099] Based on this, a comprehensive motion risk assessment result is generated by comprehensively calculating the risk decoupling effect evaluation indicators and the predicted comprehensive risk level within the future time window. The calculation of the comprehensive risk level considers the predicted values ​​of the three types of risk assessment results and their distribution range in the network. For example, based on the proportion of high-risk nodes or edges in the network, the risk state values ​​of key structural locations, and changes in static stability margin, the comprehensive risk is divided into multiple levels, such as low risk, medium risk, and high risk, with accompanying indicators indicating whether the risk will increase, remain basically stable, or decrease in the future. Therefore, the comprehensive motion risk assessment result includes two aspects: first, the comprehensive risk level at the current moment within the future time window; and second, the direction and rate of change of this comprehensive risk level over time.

[0100] Closed-loop adjustments are made based on the comprehensive motion risk assessment results. When the comprehensive motion risk level is low and the risk decoupling effect index shows that the coupling degree of different types of risk flows remains low in space and time, the motion control strategy maintains its current settings, only continuing to execute S6 and S7 in subsequent control cycles to monitor and predict risks. When the comprehensive motion risk level approaches the preset warning level or the risk decoupling effect index shows that risks are re-concentrating at certain key structures, the closed-loop adjustment process updates the motion strategy based on the comprehensive motion risk assessment results. For example, it may further reduce the overall walking speed, increase the number of supporting feet, shorten the single-foot support time, or appropriately reduce the load level, so that the new control strategy is input into the coupling dynamics model, driving the risk flow topology network towards a safer state.

[0101] Example 2, a quadruped robot dog motion risk assessment system, see [link / reference] Figure 1 As shown, it includes the following modules:

[0102] Data acquisition module: Acquires multi-source state data of the quadruped robot dog during its movement;

[0103] Topology Network Module: Constructs risk flow topology networks;

[0104] Load Coupling Module: Obtains load coupling risk assessment results;

[0105] Delay instability module: Obtains delay instability risk assessment results;

[0106] Static instability module: Obtains static instability risk assessment results;

[0107] Risk guidance module: Monitors and guides risk flows;

[0108] Comprehensive assessment module: Generates comprehensive sports risk assessment results and makes closed-loop adjustments accordingly.

[0109] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing the movement risk of a quadruped robot dog, characterized in that, include: Acquire multi-source state data of the quadruped robot dog during its movement, including foot force data, joint driving torque data, and body posture and motion state data; Construct a risk flow topology network and calculate the risk transmission impedance and capacity of each edge in the risk flow topology network based on real-time status data; The risk flow topology network includes a set of nodes formed by the rigid components of the quadruped robot dog, and a set of edges formed by the dynamic relationships between the rigid components. The rigid components include the main body, leg links, and foot contact components. The dynamic relationships include joint connections between rigid components, structural coupling, and the transmission of reaction forces generated by the foot contacting the ground. The risk transmission impedance parameter and capacity parameter of each edge in the risk flow topology network are calculated based on real-time state data. The risk transmission impedance parameter is obtained by sensitivity calculation of the state changes of the two nodes connecting the edge within a preset time window. The capacity parameter is calculated by statistically analyzing the peak values ​​of the foot force data and joint driving torque data corresponding to the edge within the preset time window, combined with the body posture and motion state of the corresponding node. The force data at the foot end and the driving torque data of the joint are mapped to the risk sources in the risk flow topology network. The load coupling risk assessment results are obtained by solving the minimum risk cut set in the risk flow topology network. Based on the body posture and motion state data, the unstructured external force interference experienced by the quadruped robot dog is identified and injected into the risk flow topology network as a risk source. Based on the network feature values, its time-series propagation characteristics are analyzed to obtain the delay instability risk assessment results. When the robot dog is in a low-speed or micro-gait state, the current support state is obtained and mapped as a virtual stable subgraph. The evolution characteristics of its connectivity and minimum cut capacity are analyzed to obtain the time-related static instability risk assessment results. The results of load coupling risk assessment, delay instability risk assessment, and static instability risk assessment are used as different risk flows propagating in the risk flow topology network. A coupling dynamics model of the risk flow is established. The risk flow is monitored, and the network topology is reconstructed by actively adjusting the motion strategy or adjusting the edge weights to guide the risk flow to the preset risk pool or to make it propagate in staggered time. Based on the state prediction of the risk flow topology network after risk guidance, a comprehensive motion risk assessment result including the evaluation of risk decoupling effect is generated, and closed-loop adjustment is carried out accordingly.

2. The method for assessing the movement risk of a quadruped robot dog according to claim 1, characterized in that, The acquisition of the load coupling risk assessment results includes: mapping the foot-end force data to the nodes associated with the corresponding supporting foot in the risk flow topology network; mapping the joint driving torque data to the edges connecting the corresponding nodes in the risk flow topology network; determining the risk flow injection intensity of each node and each edge using the foot-end force data and joint driving torque data; performing flow distribution analysis on the risk flow topology network under the constraints of risk transmission impedance parameters and capacity parameters of each edge; constructing a cut set optimization model with the risk flow intensity and capacity occupancy of the edges in the cut set as evaluation indicators; identifying the bottleneck transmission path of load propagation between different rigid components by solving the minimum risk cut set in the risk flow topology network; and generating the load coupling risk assessment results based on the capacity margin and risk flow intensity corresponding to the minimum risk cut set.

3. The method for assessing the movement risk of a quadruped robot dog according to claim 1, characterized in that, The acquisition of the delayed instability risk assessment results includes: performing time series analysis on the body posture and motion state data within a preset time window; calculating the deviation between the body posture and the preset reference posture trajectory and detecting the moment when the posture change rate exceeds a threshold; identifying the time and location of action of unstructured external force interference based on the deviation direction and amplitude; injecting unstructured external force interference as a risk source into the node corresponding to the disturbed rigid component and its adjacent edges in the risk flow topology network; after injecting the risk source, establishing a linear approximate dynamic model describing the propagation of the risk flow in the network based on the current risk flow topology network; constructing a matrix representing the network connection relationship and edge weights from the linear approximate dynamic model; solving for the eigenvalues ​​and corresponding characteristic modes of the matrix; and evaluating the cumulative impact of unstructured external force interference on the body posture stability within several subsequent time windows based on the magnitude of the real part of the eigenvalues, the trend of characteristic mode energy change over time, and the degree of modal coupling at different body motion state stages, thereby obtaining the delayed instability risk assessment results of the external force interference.

4. The method for assessing the movement risk of a quadruped robot dog according to claim 1, characterized in that, The acquisition of the static instability risk assessment results includes: determining the set of supporting feet currently in the support phase based on foot force data and body posture and motion state data, and constructing the current support polygon based on the projection of the contact position between the supporting feet and the ground onto a preset reference plane; extracting nodes corresponding to the supporting feet and nodes corresponding to the body center of gravity calculated from the risk flow topology network, constructing a virtual stable subgraph, and determining the edge weights of each edge in the virtual stable subgraph based on foot force data to characterize the static support capacity of each supporting foot for the body; performing connectivity analysis on the virtual stable subgraph, calculating the minimum cutting capacity from the node corresponding to the body center of gravity to the node corresponding to the boundary of the support polygon under the condition of including the node, and obtaining the static stability margin at the current moment; tracking the changing trend of the minimum cutting capacity within multiple consecutive time windows, and when the minimum cutting capacity continuously decreases and falls below a preset threshold, the static instability risk is judged to be increased, and a time-related static instability risk assessment result is generated based on the changing trend.

5. The method for assessing the movement risk of a quadruped robot dog according to claim 1, characterized in that, The coupled dynamics model includes: using the risk state of each node and the risk flow intensity of each edge in the risk flow topology network as state variables, and using the load-type risk flow, external force disturbance-type risk flow, and static instability risk flow corresponding to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result as input terms, and combining the risk transmission impedance parameter and capacity parameter to establish the state equation and constraints describing the evolution of the state variables over time, thus forming the coupled dynamics model.

6. The method for assessing the movement risk of a quadruped robot dog according to claim 5, characterized in that, The monitoring of the risk flow includes: calculating the risk state of each node and the risk flow intensity of each edge in the risk flow topology network in real time based on the coupled dynamics model; comparing the risk flow intensity with the capacity parameter of the corresponding edge; comparing the node risk state with the risk threshold set according to the load coupling risk assessment result, the delay instability risk assessment result, and the static instability risk assessment result; identifying nodes and edges whose risk flow intensity exceeds the corresponding capacity parameter or whose node risk state exceeds the risk threshold, and recording them as target nodes and target edges respectively; and generating control parameters for actively adjusting the motion strategy based on the position of the target nodes and target edges in the risk flow topology network and their corresponding body posture and motion state.

7. The method for assessing the movement risk of a quadruped robot dog according to claim 6, characterized in that, The guidance of the risk flow includes: adjusting the movement strategy of the quadruped robot dog according to the control parameters; modifying the edge connection relationship between the nodes corresponding to the support feet in the risk flow topology network by changing the landing sequence of the support feet and the support combination; reconstructing the topology of the risk flow topology network; or changing the risk transmission impedance parameter by adjusting the edge weight corresponding to the joint driving torque, so that different types of risk flows are preferentially directed to the preset risk sink or staggered in time in the risk flow topology network.

8. The method for assessing the movement risk of a quadruped robot dog according to claim 1, characterized in that, The generation of the comprehensive motion risk assessment result includes: using the risk status of each node and the risk flow intensity of each edge in the current risk flow topology network as initial conditions, predicting the state of the risk flow topology network within a preset future time window to obtain the node risk status and edge risk flow intensity at each time step; comparing the predicted load coupling risk assessment result, delay instability risk assessment result, and static instability risk assessment result with the assessment results at the corresponding historical moments before risk guidance, calculating the changes in coupling degree of different types of risk flows in spatial distribution and temporal evolution as an evaluation index of risk decoupling effect; and generating a comprehensive motion risk assessment result containing the current comprehensive risk level and risk evolution trend based on the evaluation index of risk decoupling effect and the comprehensive risk level predicted within the preset future time window.

9. A quadruped robot dog motion risk assessment system, characterized in that, The system applies the quadrupedal robot dog movement risk assessment method according to any one of claims 1 to 8, including: Data acquisition module: Acquires multi-source state data of the quadruped robot dog during its movement; Topology Network Module: Constructs risk flow topology networks; Load Coupling Module: Obtains load coupling risk assessment results; Delay instability module: Obtains delay instability risk assessment results; Static instability module: Obtains static instability risk assessment results; Risk guidance module: Monitors and guides risk flows; Comprehensive assessment module: Generates comprehensive sports risk assessment results and makes closed-loop adjustments accordingly.

Images