Method and system for risk sharing and collaborative path planning in multi-machine cluster environment and medium

Abstract

A method, system, and medium for risk sharing and collaborative path planning in a multi-robot swarm environment are disclosed. The method includes acquiring real-time motion state data of any robot in the swarm; when a slippage event, foot depth exceeding the limit event, or body posture change event is detected based on the real-time motion state data, calculating the physical characteristic parameters of the current area; the physical characteristic parameters include an estimated friction coefficient, terrain compliance coefficient, and instability index; encoding the physical characteristic parameters and the spatial location information of the current area into dynamic risk symbol nodes; associating the dynamic risk symbol nodes with a local neural symbol topology map and broadcasting them to the swarm; the neural symbol topology map storing dynamic risk symbol nodes from multiple robots in the swarm; and, in response to the path planning task, extracting risk terms from the neural symbol topology map and integrating the risk terms into the cost function of path planning to generate a planned path.

Description

Technical Field

[0001] This invention relates to the field of multi-machine collaborative path planning technology, and in particular to a method, system and medium for risk sharing and collaborative path planning in a multi-machine cluster environment. Background Technology

[0002] Multi-robot swarm collaborative operations have significant advantages in complex environments such as mines, disaster areas, and industrial sites. Existing swarm robots typically employ distributed mapping methods such as point cloud sharing and occupancy grid maps to synchronize environmental information and perform path planning based on geometric obstacle information. However, existing technologies have the following shortcomings: Firstly, dense point clouds or grid maps have a large data volume, which can easily cause transmission delays in scenarios with limited bandwidth and unstable communication, thereby affecting the real-time response capability of the cluster.

[0003] Secondly, geometric maps can only reflect the terrain shape and cannot perceive the physical properties of the ground. When a single robot experiences dynamic anomalies due to slipping, falling, or instability, this risk is difficult to quantify and transmit to other members of the cluster, causing subsequent robots to repeatedly encounter the same failure.

[0004] Third, the lack of a lightweight risk expression and propagation mechanism means that the cluster cannot efficiently integrate dynamic risks into the cost function of path planning, resulting in planning results that are feasible at the geometric level but pose safety risks at the dynamic level.

[0005] Therefore, existing technologies lack a collaborative path planning method that can achieve low-overhead expression and rapid sharing of dynamic risk in a multi-machine cluster environment. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for risk sharing and collaborative path planning in a multi-machine cluster environment, comprising the following steps: Obtain real-time motion status data of any robot in the cluster; When a slip event, foot sinking depth exceeding the limit event, or fuselage attitude change event is detected based on the real-time motion state data, the physical characteristic parameters of the current area are calculated; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index. The physical characteristic parameters and the spatial location information of the current area are encoded into dynamic risk symbol nodes; the dynamic risk symbol nodes include node coordinates, node type, risk level, and key physical parameters; The dynamic risk symbol nodes are associated with a local neural symbol topology map and broadcast to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster. In response to the path planning task, risk terms are extracted from the neural symbol topology map and integrated into the cost function of path planning to generate a planned path.

[0007] Optionally, the real-time motion state data includes the horizontal tangential force at the foot, the vertical normal force at the foot, the incremental sinking displacement of the foot in the vertical direction, the minimum distance from the zero moment point to the boundary of the supporting polygon, the safety and stability margin coefficient, and the measured angular velocity and reference angular velocity of the fuselage. The estimated coefficient of friction is calculated based on the ratio of the horizontal tangential force at the foot to the vertical normal force at the foot. The terrain compliance coefficient is calculated based on the ratio of the increment of vertical normal pressure at the foot to the increment of subsidence displacement; The instability index is calculated based on the minimum distance, the safety margin coefficient, and the deviation between the measured angular velocity of the fuselage and the reference angular velocity.

[0008] Optionally, the physical characteristic parameters and the spatial location information of the current region are encoded into dynamic risk symbol nodes, including: Using the spatial location information of the risk area as node coordinates, the risk type enumeration value obtained by mapping the friction coefficient or the terrain compliance coefficient as the node type, the instability index as the risk level, the friction coefficient estimate and / or the terrain compliance coefficient as key physical parameters, and generating a byte-level data structure according to a preset neural symbol mapping function, as a dynamic risk symbol node; the dynamic risk symbol node is configured to be directly embedded as a hard constraint into the optimization problem of the model predictive controller to constrain the robot's feasible state space.

[0009] Optionally, the dynamic risk symbol node is associated with a local neural symbol topology map and broadcast to the cluster, including: The dynamic risk symbol node is associated with the local neural symbol topology map, and the dynamic risk symbol node is broadcast to other robots in the cluster via the local area network; Receive dynamic risk symbol nodes broadcast from other robots in the cluster and insert them as new nodes into the neural symbol topology map; For each dynamic risk symbol node in the neural symbol topology map, risk influence factors are diffused to the surrounding area using Gaussian process regression or radial basis function, with the node coordinates of each dynamic risk symbol node as the center. The diffusion results are then superimposed to generate a continuously distributed virtual risk cost field.

[0010] Optionally, risk terms are extracted from the neural symbol topology map and integrated into the cost function of path planning to generate a planned path, including: Risk terms are extracted from the virtual risk cost field using at least one of the following methods: Calculate the line integral of the virtual risk cost field value of each path segment on the candidate path, and accumulate the line integral as a risk term to the cost function; or, calculate the expected value and variance of the virtual risk cost field of the region covered by the candidate path based on Gaussian process regression, and the product of the risk aversion coefficient and the variance value, and then accumulate the sum of the expected value and the product as a risk term to the cost function; or, extract the peak value of the virtual risk cost field in the region covered by the candidate path, and embed the peak value as a hard constraint into the optimization problem of the model predictive controller to limit the feasible state space in the cost function solution process; Path search or trajectory optimization is performed based on the cost function after integrating the aforementioned risk terms to generate a planned path.

[0011] Optionally, the method further includes: when the planned path cannot avoid the risk area, adjusting the gait control parameters according to the risk type in the dynamic risk symbol node; If the risk type is slippery ground, generate gait control instructions to reduce center of gravity height, decrease stride length, and increase stride frequency; If the risk type is soft ground, generate gait control commands to increase the contact area between the foot and the ground and reduce the stiffness coefficient of the impedance controller.

[0012] Optionally, the dynamic risk symbol node may also include a survival time attribute; The method further includes: after associating the dynamic risk symbol node with the local neural symbol topology map, starting a timer for the existing dynamic risk symbol node; When the lifetime of any existing dynamic risk symbol node expires and no update verification information is received from the cluster, the risk level of the existing dynamic risk symbol node is automatically reduced or it is removed from the neural symbol topology map.

[0013] Optionally, the cluster includes at least two robots with different kinematic models; The method further includes: obtaining the kinematic model of the current robot; Based on the kinematic model, the physical characteristic parameters in the same dynamic risk symbol node are converted into risk values ​​that match the current robot kinematic model.

[0014] Corresponding to the aforementioned multi-machine cluster environment risk sharing and collaborative path planning method, this invention provides a multi-machine cluster environment risk sharing and collaborative path planning system, comprising: The data acquisition module is used to acquire real-time motion status data of any robot in the cluster; The calculation module is used to calculate the physical characteristic parameters of the current area when a slip event, a foot sink-in depth exceeding the limit event, or a sudden change in fuselage attitude event is detected based on the real-time motion state data; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index; The encoding module is used to encode the physical feature parameters and the spatial location information of the current area into dynamic risk symbol nodes; the dynamic risk symbol nodes include location coordinates, risk type, risk level, and key physical parameters; The neural symbol topology map update module is used to associate the dynamic risk symbol nodes with the local neural symbol topology map and broadcast them to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster; The path planning module is used to respond to path planning tasks, extract risk items from the neural symbol topology map, and integrate the risk items into the cost function of path planning to generate a planned path.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a multi-machine cluster environment risk sharing and collaborative path planning program, wherein when the multi-machine cluster environment risk sharing and collaborative path planning program is executed by a processor, it implements the steps of the multi-machine cluster environment risk sharing and collaborative path planning method described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) By quantifying slip events, foot sinking depth exceeding limits, or body attitude change events detected by real-time motion state data into physical characteristic parameters including friction coefficient estimates, terrain compliance coefficients, and instability indices, and encoding them with the spatial location information of the current area into lightweight dynamic risk symbol nodes, the byte-level low-overhead expression of dynamic risks is realized, significantly reducing the cluster communication bandwidth requirements. By associating this node with the local neural symbol topology map and broadcasting it to the cluster, cluster members can share non-geometric ground physical property information, avoiding repeated slippage, sinking, or instability of the robot in the same area. By extracting risk items from the neural symbol topology map and integrating them into the cost function of path planning, the planned path is geometrically feasible while being constrained by the dynamic safety boundary, thereby directly solving the technical problems of transmission delay caused by the large amount of dense geometric map data, the inability of the geometric map to perceive ground physical properties, and the lack of lightweight risk propagation and integration mechanisms.

[0017] (2) By calculating the physical characteristic parameters of the current region, the physical characteristic parameters can be obtained directly from the mechanical quantities of the robot foot interacting with the ground and the dynamic response of the body posture, thus providing a numerical representation basis that directly corresponds to the robot-ground physical interaction process for the quantification of dynamic risks.

[0018] (3) By using the spatial location information of the risk area as node coordinates, the risk type enumeration value obtained by mapping the friction coefficient or terrain compliance coefficient as node type, the instability index as risk level, and the friction coefficient and / or terrain compliance coefficient as key physical parameters, and generating a byte-level data structure according to the preset neural symbol mapping function, the dynamic risk symbol node has a unified and extremely low data volume expression format, which can directly adapt to the bandwidth constraints of cluster broadcast; at the same time, by configuring the node to be directly embedded into the optimization problem of the model predictive controller as a hard constraint condition, the risk information can directly constrain the feasible state space of the robot without going through complex intermediate transformation, thus realizing the semantic direct communication between risk expression and underlying motion control.

[0019] (4) By associating dynamic risk symbol nodes with the local neural symbol topology map and broadcasting them to other robots in the cluster via the local area network, and simultaneously receiving nodes broadcast from other robots and inserting them into the map as new nodes, distributed aggregation and bidirectional synchronization of dynamic risk information within the cluster are realized. By using the node coordinates of each dynamic risk symbol node as the center, Gaussian process regression or radial basis function is used to diffuse risk influence factors to the surrounding area, and the diffusion results are superimposed to generate a continuously distributed virtual risk cost field, so that discrete symbolic risk nodes are transformed into risk distribution representations that cover continuous space and have spatial differentiability, thereby providing a field-based mathematical foundation for path planning that can perform integration, expectation and extremum operations.

[0020] (5) By extracting the virtual risk cost field as a risk term and integrating it into the cost function through at least one of the following methods: line integral, weighted sum of expected value and variance, or peak hard constraint, path planning can select the appropriate mathematical representation under different risk sensitivity requirements. The line integral method realizes the cumulative measurement of the overall risk exposure of candidate path segments, the sum of the product of expected value, risk aversion coefficient and variance value takes into account both the average level of risk and uncertainty avoidance, and the peak hard constraint method directly limits extreme risks to the feasible boundary of the model predictive control optimization problem. The synergistic integration of the above extraction methods and cost function enables path search or trajectory optimization to further screen out the planning path that meets the dynamic safety constraints in the geometrically feasible solution space.

[0021] (6) When the planned path cannot avoid the risk area, the gait control parameters are adaptively adjusted according to the risk type in the dynamic risk symbol node, so that the robot can maintain motion stability when it has to cross the risk area. Specifically, for slippery ground, gait control commands are generated to reduce the height of the center of mass, reduce the stride and increase the step frequency, which can effectively suppress foot slippage and shorten the instability window of the single foot support phase. For soft ground, gait control commands are generated to increase the contact area between the foot and the ground and reduce the stiffness coefficient of the impedance controller, which can reduce the ground pressure and avoid excessive foot sinking, so that the robot's gait in the risk area can actively adapt to the physical properties of the terrain.

[0022] (7) By setting a lifespan attribute for dynamic risk symbol nodes and starting a timer after associating them with the local neural symbol topology map, the risk information is subject to time constraints. When the lifespan of an existing node expires and no update verification information is received from the cluster, the risk level of the existing node is automatically reduced or it is removed from the neural symbol topology map. This avoids the continuous occupation of map storage space and interference with subsequent path planning decisions caused by the change of environmental physical properties over time. It ensures that the risk information retained in the neural symbol topology map remains dynamically consistent with the current state of the environment.

[0023] (8) By acquiring the kinematic model of the current robot in a cluster containing at least two different kinematic models, and converting the physical feature parameters in the same dynamic risk symbol node into risk values ​​that match the current robot kinematic model, the original physical quantities of dynamic risk collected and shared by robots of different configurations can be mapped by heterogeneous robots into personalized risk measures that conform to their own dynamic constraints, thereby realizing the semantic compatibility and reusable sharing of dynamic risk information between heterogeneous clusters. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a simplified flowchart of an embodiment of the multi-machine cluster environment risk sharing and collaborative path planning method of the present invention; Figure 2 This is a framework diagram of an embodiment of the multi-machine cluster environment risk sharing and collaborative path planning system of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] like Figure 1 As shown, the present invention provides a method for risk sharing and collaborative path planning in a multi-machine cluster environment, which includes the following steps: Obtain real-time motion status data of any robot in the cluster; When a slip event, foot sinking depth exceeding the limit event, or fuselage attitude change event is detected based on real-time motion state data, the physical characteristic parameters of the current area are calculated; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index. The physical characteristic parameters and the spatial location information of the current area are encoded into dynamic risk symbol nodes; the dynamic risk symbol nodes include node coordinates, node type, risk level, and key physical parameters; The dynamic risk symbol nodes are associated with the local neural symbol topology map and broadcast to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster; In response to the path planning task, risk terms are extracted from the neural symbol topology map and integrated into the cost function of path planning to generate the planned path.

[0027] This invention quantifies slip events, foot sinking depth exceeding limits, or sudden body posture changes detected by real-time motion data into physical characteristic parameters including friction coefficient estimates, terrain compliance coefficients, and instability indices. These parameters are then encoded with the spatial location information of the current area into lightweight dynamic risk symbol nodes, achieving byte-level low-overhead representation of dynamic risks and significantly reducing cluster communication bandwidth requirements. By associating these nodes with a local neural symbol topology map and broadcasting them to the cluster, cluster members can share non-geometric ground physical property information, preventing subsequent robots from repeatedly encountering slippage, sinking, or instability in the same area. By extracting risk terms from the neural symbol topology map and integrating them into the cost function of path planning, the planned path is geometrically feasible while being constrained by dynamic safety boundaries. This directly solves the technical problems of transmission delays caused by the large amount of dense geometric map data, the inability of the geometric map to perceive ground physical properties, and the lack of lightweight risk propagation and integration mechanisms.

[0028] In this embodiment, the real-time motion state data includes the horizontal tangential force at the foot, the vertical normal force at the foot, the incremental sinking displacement of the foot in the vertical direction, the minimum distance from the zero moment point to the boundary of the supporting polygon, the safety and stability margin coefficient, and the measured angular velocity and reference angular velocity of the robot body; preferably, the real-time motion state data is obtained by monitoring the motion state in real time through the robot's foot force sensor and IMU. The coefficient of friction is typically estimated based on the triaxial force conditions of a foot force sensor. The estimated coefficient of friction is calculated based on the ratio of the horizontal tangential force to the vertical normal force at the foot, as shown in the following formula: ;in, The horizontal tangential force at the foot end is fixed along the X-axis (e.g., the forward / backward direction) in the robot's body coordinate system. The horizontal tangential force at the foot end is fixed along the Y-axis (e.g., left-right direction) in the robot's body coordinate system. To fix the vertical normal force on the foot end along the Z-axis of the robot's body coordinate system, This is an estimated value for the coefficient of friction; Preferably, online smoothing and slip correction can be further performed on the friction coefficient estimate. To prevent transient noise interference, a first-order low-pass filter is usually used, combined with the center of mass (CoM) acceleration monitored by the IMU. The specific formula is as follows: ;in, This is the corrected estimate of the friction coefficient. Let be the resultant vector of all horizontal tangential force vectors at the foot ends in the supporting phase, i.e., the total horizontal tangential force after vector synthesis of the tangential forces at each supporting foot end in the horizontal plane, where m is the known total mass of the robot and g is the standard gravitational acceleration. The instantaneous linear acceleration of the robot body in the vertical direction (Z-axis) is obtained by measuring the robot body's IMU and eliminating the gravity component; the estimated friction coefficient reflects the physical limit of the support domain. If it is much smaller than the preset threshold, the system automatically marks it as "slippery", indicating that a slippage event exists. The topographic compliance coefficient reflects the softness of the ground and is mainly calculated by the relationship between the foot's sinking depth and the contact force. Based on an elastic foundation model, it assumes that the ground undergoes linear or nonlinear deformation under foot pressure. Specifically, the topographic compliance coefficient is calculated based on the ratio of the increment of vertical normal pressure at the foot to the increment of settlement displacement, as shown in the following formula: ;in, This represents the increment of vertical positive pressure at the foot (read directly from the force sensor). K is the subsidence displacement increment (the difference between the foot position calculated by the forward kinematics solution and the ground height predicted by the visual / odometry method), and K is the terrain compliance coefficient. The instability index measures the degree to which a robot deviates from its equilibrium state. It is typically based on the zero-torque point (ZMP) or a linear inverted pendulum model. Specifically, the instability index is calculated based on the minimum distance, the safety margin coefficient, and the deviation between the measured angular velocity of the robot body and the reference angular velocity. The specific formula is as follows: ;in, For short distances, For safety and stability margin coefficient, The preset attitude disturbance weighting coefficient is used to adjust the contribution ratio of the fuselage rotation motion to the overall instability uncertainty, with a preferred range of 0.2-0.5. This refers to the measured angular velocity of the fuselage (the angular velocity vector of the fuselage measured in real time by the IMU). The reference angular velocity (usually 0) It is an instability index.

[0029] This invention calculates the physical characteristic parameters of the current region, enabling these parameters to be directly decoupled from the mechanical quantities of the robot's foot-ground interaction and the dynamic response of the robot's posture. This provides a numerical representation basis for quantifying dynamic risks that directly corresponds to the robot-ground physical interaction process.

[0030] In this embodiment, physical characteristic parameters and spatial location information of the current region are encoded into dynamic risk symbol nodes, including: The spatial location information (GPS / SLAM topological coordinates) of the risk area is used as the node coordinates. The risk type enumeration value (e.g., 0: slippery, 1: soft, 2: steep slope) obtained by mapping the friction coefficient estimate or terrain compliance coefficient is used as the node type. The instability index is directly used as the risk level (0-1.0). The friction coefficient estimate and / or terrain compliance coefficient are used as key physical parameters. A byte-level data structure is generated according to the preset neural symbol mapping function as a dynamic risk symbol node. The dynamic risk symbol node is configured to be directly embedded into the optimization problem of the model predictive controller as a hard constraint condition to constrain the robot's feasible state space.

[0031] Preferably, the risk type enumeration value is hard-coded based on the preset event state machine inside a single robot. The specific determination criteria are as follows: when it is determined that only a slip event has been triggered, the risk type field is automatically assigned a preset first enumeration value (e.g., fixed as an unsigned integer value 0x01); when it is determined that only a foot depth exceeding the limit event has been triggered, the field is automatically assigned a preset second enumeration value (e.g., fixed as an unsigned integer value 0x02); when it is determined that only a body posture change event has been triggered, the field is automatically assigned a preset third enumeration value (e.g., fixed as an unsigned integer value 0x03); when multiple abnormal events are detected to occur simultaneously within the same time window, it is determined to be a multi-scenario superposition event, and the field is automatically assigned a preset fourth enumeration value (e.g., fixed as an unsigned integer value 0x04).

[0032] The enumerated values ​​output by the aforementioned event state machine, together with the terrain attribute enumerated values ​​obtained by mapping the friction coefficient estimate or terrain compliance coefficient, constitute the complete encoding space of the risk type field. The terrain attribute enumerated values ​​represent the inherent physical category of the terrain, while the event state machine enumerated values ​​represent the robot's current dynamic abnormal state. The two are jointly encoded in a single byte by distinguishing high bits and superimposing low bits. Through this preset discrete enumerated value mapping, the accurate transmission of highly complex terrain physical heterogeneity types can be completed in cluster communication with only 1 byte of data overhead.

[0033] This invention uses the spatial location information of the risk area as node coordinates, the risk type enumeration value obtained by mapping the friction coefficient estimate or terrain compliance coefficient as the node type, the instability index as the risk level, and the friction coefficient estimate and / or terrain compliance coefficient as key physical parameters. It generates a byte-level data structure based on a preset neural symbol mapping function, enabling the dynamic risk symbol nodes to have a unified and extremely low data volume expression format that can directly adapt to the bandwidth constraints of cluster broadcasting. Simultaneously, by configuring this node to be directly embedded as a hard constraint into the optimization problem of the model predictive controller, risk information can directly constrain the robot's feasible state space without complex intermediate transformations, achieving semantic direct communication between risk expression and underlying motion control.

[0034] In this embodiment, the "spatial location information of the risk area" refers to the absolute three-dimensional spatial coordinates of the robot's center of mass or the triggering foot at the instant when the robot is monitored in real time by the foot force sensor and IMU to trigger a slip event, a foot depth exceeding the limit event, or a sudden change in body posture event.

[0035] To ensure high-speed communication for the cluster in weak network environments, each robot only compresses and encodes the absolute three-dimensional spatial coordinates along with the corresponding dynamic physical characteristic parameters into a byte-level symbol node for broadcast. However, when subsequent robots "consume" (apply) this symbol node for collaborative path planning, they are not required to precisely step on this coordinate point to avoid danger. Instead, using this coordinate point as the center, and utilizing its carried physical characteristic parameters (such as instability index and terrain compliance coefficient), a pre-defined diffusion algorithm renders a radius region with a continuous physical risk gradient online. This achieves dynamic adaptive risk avoidance where "transmission is a point, planning is a surface."

[0036] In this embodiment, associating dynamic risk symbol nodes with a local neural symbol topology map and broadcasting it to the cluster includes: Associate the dynamic risk symbol node with the local neural symbol topology map, and broadcast the dynamic risk symbol node to other robots in the cluster via the local area network; Receive dynamic risk symbol nodes broadcast from other robots in the cluster and insert them as new nodes into the neural symbol topology map; For each dynamic risk symbol node in the neural symbol topology map, risk influence factors are diffused to the surrounding area using Gaussian process regression or radial basis function, centered on the node coordinates of each dynamic risk symbol node. The diffusion results are then superimposed to generate a continuously distributed virtual risk cost field. The specific formula is as follows: ;in, For the surrounding area Diffusion risk influencing factors The risk level is represented by P, and the node coordinates are represented by P. The radius of risk impact.

[0037] This invention achieves distributed aggregation and bidirectional synchronization of dynamic risk information within the cluster by associating dynamic risk symbol nodes with a local neural symbol topology map and broadcasting it to other robots in the cluster via a local area network. Simultaneously, it receives nodes broadcast from other robots and inserts them as new nodes into the map. By using Gaussian process regression or radial basis functions to diffuse risk influence factors to the surrounding area with the node coordinates of each dynamic risk symbol node as the center, and superimposing the diffusion results to generate a continuously distributed virtual risk cost field, discrete symbolic risk nodes are transformed into risk distribution representations that cover a continuous space and have spatial differentiability. This provides a field-based mathematical foundation for path planning that can perform integration, expectation, and extremum calculations.

[0038] In this embodiment, when the robot needs to plan a path, its cost function considers not only geometric obstacles but also the physical risk term shared by the cluster; the specific formula for the cost function J is as follows: Where V is the real-time actual centroid linear velocity vector of the robot. In actual operation, this state is estimated by high-frequency extended Kalman filter (EKF) fusion of the acceleration integral of the robot's body IMU and the forward kinematics of the foot. The reference expected velocity vector issued by the task layer (directly given by the upper-level global scheduling command or the operator's remote control input, used as the reference tracking target for the robot's movement); This is the speed tracking weighting coefficient, used to control the stiffness of the robot when tracking the desired speed. Its preferred engineering range is 0.5-1.5. The geometric obstacle avoidance cost term is calculated using the robot's onboard LiDAR or depth camera. It measures the shortest spatial distance *d* from the robot's current center of mass to the nearest 3D geometric obstacle boundary, and is obtained by taking its reciprocal or exponential repulsive field form. This cost is applied when the robot is extremely close to the obstacle. This is the geometric obstacle avoidance weight coefficient, used to adjust the intervention of known static / dynamic geometric obstacle repulsion fields on path planning. Its preferred engineering range is 2.0-5.0. This is a risk item. The corresponding weight for the risk item is determined by the timestamp of the received symbol node. With current system time The difference (i.e., spatiotemporal dissipation characteristics) is dynamically adjusted online. The specific calculation formula is as follows: ,in The initial basic risk weight is set at 10.0-25.0 for engineering projects. Preferably, risk terms are extracted from the neural symbol topology map and integrated into the cost function of path planning to generate a planned path, including: Extract risk terms from the virtual risk cost field using at least one of the following methods: The line integral of the virtual risk cost field value of each path segment on the candidate path is calculated, and the line integral is accumulated as a risk term into the cost function for global path decision-making. If a path is very short but has extremely high risk density (cost field strength), its integral value will quickly exceed that of a longer but safer detour path, thus driving the algorithm to choose a detour. Alternatively, the expected value and variance of the virtual risk cost field covering the area of ​​the candidate path can be calculated based on Gaussian process regression, along with the product of the risk aversion coefficient and the variance. The sum of the expected value and the product is then added to the cost function as a risk term for spatiotemporal prediction. Since the virtual cost field is essentially a probability distribution due to noise in sensor identification, the expected value based on Gaussian process regression is used for extraction. This not only considers the magnitude of the risk but also the confidence level of the data returned by the preceding machine. If the confidence level of the risk points marked by the preceding machine is low (large variance), the subsequent machine will be more cautious in planning and maintain a larger safety margin. Alternatively, the peak value of the virtual risk cost field within the area covered by the candidate path can be extracted and embedded as a hard constraint into the optimization problem of the model predictive controller to limit the feasible state space in the cost function solution process. This method is used for the underlying dynamic kernel solution. Even if the average risk of the entire path is very low, as long as the risk peak value of one point (such as a small piece of ice) exceeds the robot's dynamic compensation limit, the system must force a gait switch or replan. Path search or trajectory optimization is performed based on the cost function after integrating risk terms to generate a planned path.

[0039] In this embodiment, the risk aversion coefficient is determined based on the switching of task modes: when the cluster is in a high-time-sensitive search and rescue / assault task mode, the system tolerates a higher risk of physical instability, and at this time the risk aversion coefficient is automatically lowered, preferably reduced to [value missing]. This causes the robot to exhibit risk-prone characteristics in exchange for shorter travel time and higher path straightness; when the cluster is in high-value material transportation / fine inspection mode, the system has zero tolerance for instability and rollover, at which point the risk aversion coefficient is automatically increased (preferably raised to ). This causes the robot to exhibit extreme risk aversion.

[0040] Alternatively, the risk aversion coefficient can be determined based on the differences in the throughput capabilities of individual hardware: in heterogeneous clusters, for heavy-duty quadruped robots with a lower chassis center of gravity, stronger foot grip, and high-output torque joints, their own dynamic boundaries against slippage and rollover are wider, and the system assigns them a smaller risk aversion coefficient. For lightweight, high-precision inspection robots, a larger risk aversion coefficient is assigned. .

[0041] Alternatively, the risk aversion coefficient can be determined based on a correction for environmental density: the system uses the density of risk symbol nodes within a local spatiotemporal window in the aforementioned topology map to calculate the severity of environmental heterogeneity. When the surrounding red and orange risk points are extremely dense, the system automatically adjusts the density using a preset monotonically increasing function. Increase the benchmark value This forces an amplification of the penalty for unknown variance, driving the planner to choose an extremely conservative, safe path.

[0042] This invention extracts the virtual risk cost field as a risk term and integrates it into the cost function using at least one of the following methods: line integral, weighted sum of expected value and variance, or peak hard constraint. This allows path planning to select appropriate mathematical representations under different risk sensitivity requirements. The line integral method realizes the cumulative measurement of the overall risk exposure of candidate path segments, while the sum of the expected value, risk aversion coefficient, and variance takes into account both the average risk level and uncertainty avoidance. The peak hard constraint method directly limits extreme risks to the feasible boundary of the model predictive control optimization problem. The synergistic integration of the above extraction methods and cost function enables path search or trajectory optimization to further screen out planning paths that meet dynamic safety constraints in the geometrically feasible solution space.

[0043] In this embodiment, the method further includes: when the planned path cannot avoid the risk area, adjusting the gait control parameters according to the risk type in the dynamic risk symbol node; If the risk type is slippery ground, generate gait control instructions to reduce center of gravity height, shorten stride and increase stride frequency; If the risk type is soft ground, generate gait control commands to increase the contact area between the foot and the ground and reduce the stiffness coefficient of the impedance controller.

[0044] When the planned path cannot avoid risk areas, the robot can maintain motion stability when it must traverse risk areas by adaptively adjusting gait control parameters according to the risk type in the dynamic risk symbol nodes. Specifically, for slippery ground, gait control commands are generated to reduce the center of gravity height, shorten the stride, and increase the stride frequency, which can effectively suppress foot slippage and shorten the instability window of the single-foot support phase. For soft ground, gait control commands are generated to increase the contact area between the foot and the ground and reduce the stiffness coefficient of the impedance controller, which can reduce the ground pressure and avoid excessive foot sinking, so that the robot's gait in risk areas can actively adapt to the physical properties of the terrain.

[0045] In this embodiment, the dynamic risk symbol node also includes a survival time attribute; The method also includes: after associating the dynamic risk symbol node with the local neural symbol topology map, starting a timer for the existing dynamic risk symbol node; When the lifetime of any existing dynamic risk symbol node expires and no update verification information is received from the cluster, the risk level of the existing dynamic risk symbol node is automatically reduced or it is removed from the neural symbol topology map.

[0046] This invention sets a lifespan attribute for dynamic risk symbol nodes and starts timing after associating them with the local neural symbol topology map, thus giving risk information a time constraint. When the lifespan of an existing node expires and no update verification information is received from the cluster, the risk level of the existing node is automatically reduced or it is removed from the neural symbol topology map. This avoids outdated risk information caused by changes in environmental physical properties over time from continuously occupying map storage space and interfering with subsequent path planning decisions, ensuring that the risk information retained in the neural symbol topology map remains dynamically consistent with the current state of the environment.

[0047] In the above embodiments, a quadruped robot is used as an example for illustration. In other embodiments, the cluster may simultaneously include at least two heterogeneous types of quadruped robots, wheeled robots, or tracked robots; The method further includes: obtaining the kinematic model of the current robot; Based on the kinematic model, the physical characteristic parameters of the same dynamic risk symbol node are converted into risk values ​​that match the current robot's kinematic model. For example, a drone uses visual perception from high altitude to initially label the geometric attributes of topological nodes, while a quadruped robot supplements the physical attributes such as friction coefficient and hardness through ground exploration, and propagates the symbolic risk to other ground robots, forming a three-dimensional layout of "high-altitude early warning - ground verification - global sharing". As another example, for the same slippery terrain, the system classifies it as "extremely high risk (impassable)" based on the kinematic model of a wheeled robot, while classifying it as "low risk" for a quadruped robot with jumping ability, thus achieving semantic compatibility and reusable sharing of risk information among heterogeneous clusters.

[0048] This invention obtains the kinematic model of the current robot in a cluster containing at least two different kinematic models, and converts the physical feature parameters in the same dynamic risk symbol node into risk values ​​that match the current robot's kinematic model. This allows the original physical quantities of dynamic risk collected and shared by robots of different configurations to be mapped by heterogeneous robots into personalized risk measures that conform to their own dynamic constraints, thereby achieving semantic compatibility and reusable sharing of dynamic risk information between heterogeneous clusters.

[0049] The invention is illustrated below through a brief three-robot embodiment: Leading robot trigger: Robot No. 1 slipped slightly at point A due to oil accumulation on the ground, and real-time motion status data of Robot No. 1 was acquired.

[0050] Symbol generation: When a slip event is detected in robot No. 1, the physical characteristic parameters of the current area are calculated, and a symbol [node coordinates: point A, node type: slip, risk level: 0.8, estimated friction coefficient: 0.12] is immediately generated. The data volume is only a few tens of bytes.

[0051] Instant sharing: The dynamic risk symbol node is associated with the local neural symbol topology map and broadcast to the cluster. Robots 2 and 3 synchronize the node within 100 milliseconds, and point A in their local neural symbol topology map is marked as "red high risk". The specific marking color can be set according to actual needs.

[0052] Early risk avoidance: Robot No. 2's original planned route passed through point A. At this point, the cost function was recalculated, and the system automatically guided it to detour to the left and reach dry ground 2 meters before point A.

[0053] Forced passage: Robot No. 3 could not detour due to an obstacle on its left. The system recognized the risk at point A and switched its gait from "Trot (diagonal trot)" to "Crawl (crawling)" in advance, allowing it to pass through the risk point smoothly without slipping.

[0054] In the above embodiments, each robot in the cluster broadcasts dynamic risk symbol nodes via a local area network (such as ROS2 DDS or 5G). In other embodiments, a fully decentralized point-to-point (P2P) communication mode can be adopted: when individual robots meet at close range using Bluetooth Low Energy or Ultra Wideband (UWB), they exchange incremental symbol nodes in the topology map through a handshake protocol, much like virus propagation, thereby achieving risk diffusion in areas without network coverage. Alternatively, fixed edge computing nodes can be deployed in the work area: robots upload the perceived neural symbols to the nearest edge node, which is responsible for globally optimizing and persistently storing the topology map of the area and pushing it to other robots entering the area.

[0055] In the above embodiments, the present invention can modify the path planning of subsequent robots based on a shared risk map. In other embodiments, a robot can be assigned to perform physical marking tasks (such as spraying markings or placing electronic road signs) at risk points to provide physical warnings for non-intelligent devices or personnel. Alternatively, the optimal gait parameter sequence recorded when the preceding robot successfully traversed the same risk area can be embedded in the dynamic risk symbol node, and subsequent robots can directly retrieve this sequence to imitate it when they arrive at the area, thus achieving "skill sharing".

[0056] All of the above-described embodiments are based on the core concept of neural symbol topology map and dynamic risk symbol node disclosed in this invention, and can be arbitrarily combined according to actual operational needs, which will not be elaborated here.

[0057] like Figure 2 As shown, the present invention also provides a multi-machine cluster environment risk sharing and collaborative path planning system, which includes: The data acquisition module 10 is used to acquire real-time motion status data of any robot in the cluster; The calculation module 20 is used to calculate the physical characteristic parameters of the current area when a slip event, foot sink-in depth exceeding the limit event, or fuselage attitude change event is detected based on real-time motion state data; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index; Encoding module 30 is used to encode physical feature parameters and spatial location information of the current area into dynamic risk symbol nodes; the dynamic risk symbol nodes include location coordinates, risk type, risk level and key physical parameters; The neural symbol topology map update module 40 is used to associate dynamic risk symbol nodes with the local neural symbol topology map and broadcast them to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster; The path planning module 50 is used to respond to path planning tasks, extract risk terms from the neural symbol topology map, and integrate the risk terms into the cost function of path planning to generate a planned path.

[0058] This invention also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the memory described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores at least one instruction, which is loaded and executed by a processor to implement... Figure 1 The method for risk sharing and collaborative path planning in a multi-machine cluster environment is shown. The computer-readable storage medium can be a read-only memory, a hard disk, or an optical disk, etc.

[0059] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments and storage medium embodiments, since they are basically similar to method embodiments, the descriptions are relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0060] Furthermore, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0061] The foregoing description illustrates and describes preferred embodiments of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept by means of the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for risk sharing and collaborative path planning in a multi-machine cluster environment, characterized in that, Includes the following steps: Obtain real-time motion status data of any robot in the cluster; When a slip event, foot sinking depth exceeding the limit event, or fuselage attitude change event is detected based on the real-time motion state data, the physical characteristic parameters of the current area are calculated; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index. The physical characteristic parameters and the spatial location information of the current area are encoded into dynamic risk symbol nodes; the dynamic risk symbol nodes include node coordinates, node type, risk level, and key physical parameters; The dynamic risk symbol nodes are associated with a local neural symbol topology map and broadcast to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster. In response to the path planning task, risk terms are extracted from the neural symbol topology map and integrated into the cost function of path planning to generate a planned path.

2. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 1, characterized in that, The real-time motion state data includes the horizontal tangential force at the foot, the vertical normal force at the foot, the incremental sinking displacement of the foot in the vertical direction, the minimum distance from the zero moment point to the boundary of the supporting polygon, the safety and stability margin coefficient, and the measured angular velocity and reference angular velocity of the fuselage. The estimated coefficient of friction is calculated based on the ratio of the horizontal tangential force at the foot to the vertical normal force at the foot. The terrain compliance coefficient is calculated based on the ratio of the increment of vertical normal pressure at the foot to the increment of subsidence displacement; The instability index is calculated based on the minimum distance, the safety margin coefficient, and the deviation between the measured angular velocity of the fuselage and the reference angular velocity.

3. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 2, characterized in that, Encoding the physical characteristic parameters and the spatial location information of the current region into dynamic risk symbol nodes includes: Using the spatial location information of the risk area as node coordinates, the risk type enumeration value obtained by mapping the friction coefficient or the terrain compliance coefficient as the node type, the instability index as the risk level, the friction coefficient estimate and / or the terrain compliance coefficient as key physical parameters, and generating a byte-level data structure according to a preset neural symbol mapping function, as a dynamic risk symbol node; the dynamic risk symbol node is configured to be directly embedded as a hard constraint into the optimization problem of the model predictive controller to constrain the robot's feasible state space.

4. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 1, characterized in that, Associating the dynamic risk symbol nodes with the local neural symbol topology map and broadcasting it to the cluster includes: The dynamic risk symbol node is associated with the local neural symbol topology map, and the dynamic risk symbol node is broadcast to other robots in the cluster via the local area network; Receive dynamic risk symbol nodes broadcast from other robots in the cluster and insert them as new nodes into the neural symbol topology map; For each dynamic risk symbol node in the neural symbol topology map, risk influence factors are diffused to the surrounding area using Gaussian process regression or radial basis function, with the node coordinates of each dynamic risk symbol node as the center. The diffusion results are then superimposed to generate a continuously distributed virtual risk cost field.

5. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 4, characterized in that, Extracting risk terms from the neural symbol topology map and integrating these risk terms into the cost function of path planning to generate a planned path includes: Risk terms are extracted from the virtual risk cost field using at least one of the following methods: Calculate the line integral of the virtual risk cost field value of each path segment on the candidate path, and accumulate the line integral as a risk term to the cost function; or, calculate the expected value and variance of the virtual risk cost field of the region covered by the candidate path based on Gaussian process regression, and the product of the risk aversion coefficient and the variance value, and then accumulate the sum of the expected value and the product as a risk term to the cost function; or, extract the peak value of the virtual risk cost field in the region covered by the candidate path, and embed the peak value as a hard constraint into the optimization problem of the model predictive controller to limit the feasible state space in the cost function solution process; Path search or trajectory optimization is performed based on the cost function after integrating the aforementioned risk terms to generate a planned path.

6. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 1, characterized in that, Also includes: When the planned path cannot avoid the risk area, the gait control parameters are adjusted according to the risk type in the dynamic risk symbol node; If the risk type is slippery ground, generate gait control instructions to reduce center of gravity height, decrease stride length, and increase stride frequency; If the risk type is soft ground, generate gait control commands to increase the contact area between the foot and the ground and reduce the stiffness coefficient of the impedance controller.

7. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 1, characterized in that, The dynamic risk symbol node also includes a survival time attribute; The method further includes: after associating the dynamic risk symbol node with the local neural symbol topology map, starting a timer for the existing dynamic risk symbol node; When the lifetime of any existing dynamic risk symbol node expires and no update verification information is received from the cluster, the risk level of the existing dynamic risk symbol node is automatically reduced or it is removed from the neural symbol topology map.

8. The method for risk sharing and collaborative path planning in a multi-machine cluster environment according to claim 1, characterized in that, The cluster includes robots with at least two different kinematic models; The method further includes: obtaining the kinematic model of the current robot; Based on the kinematic model, the physical characteristic parameters in the same dynamic risk symbol node are converted into risk values ​​that match the current robot kinematic model.

9. A multi-machine cluster environment risk sharing and collaborative path planning system, characterized in that, include: The data acquisition module is used to acquire real-time motion status data of any robot in the cluster; The calculation module is used to calculate the physical characteristic parameters of the current area when a slip event, a foot sink-in depth exceeding the limit event, or a sudden change in fuselage attitude event is detected based on the real-time motion state data; the physical characteristic parameters include the estimated value of the friction coefficient, the terrain compliance coefficient, and the instability index; The encoding module is used to encode the physical feature parameters and the spatial location information of the current area into dynamic risk symbol nodes; The dynamic risk symbol node includes location coordinates, risk type, risk level, and key physical parameters; The neural symbol topology map update module is used to associate the dynamic risk symbol nodes with the local neural symbol topology map and broadcast them to the cluster; the neural symbol topology map stores dynamic risk symbol nodes from multiple robots in the cluster; The path planning module is used to respond to path planning tasks, extract risk items from the neural symbol topology map, and integrate the risk items into the cost function of path planning to generate a planned path.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a multi-machine cluster environment risk sharing and collaborative path planning program. When executed by a processor, the multi-machine cluster environment risk sharing and collaborative path planning program implements the steps of the multi-machine cluster environment risk sharing and collaborative path planning method as described in any one of claims 1 to 8. A cluster environment risk sharing and collaborative path planning method based on neural symbol topology maps.

Images