A petrochemical explosion-proof humanoid robot trajectory planning method, device and equipment

Abstract

The application discloses a petrochemical explosion-proof humanoid robot trajectory planning method, device and equipment, collects petrochemical environment gas concentration data, robot joint motion parameters and target position information, calibrates a dangerous area of the gas concentration data to generate a forbidden map, determines a safety node and a joint activity range; based on the target position information and the safety node distribution, an initial path skeleton is constructed, local dangerous sections and major dangerous sections are identified, motion constraint conditions and poison avoidance paths are generated; based on the motion constraint conditions and the joint activity range, an action primitive library is generated, stability filtering is performed to generate stable action primitives, and a marching scheme set is formed; a joint load peak value is acquired to construct a dynamics envelope, execution fitness is determined to form an execution probability field; based on the poison avoidance path, the marching scheme set and the execution probability field, a three-dimensional decision space is constructed, an optimal convergence point is found, and a final planning trajectory is generated, thereby providing a safe and efficient trajectory planning scheme for the petrochemical explosion-proof humanoid robot.

Description

Technical Field

[0001] This invention relates to the field of robot trajectory planning technology, and in particular to a method, apparatus and equipment for trajectory planning of a petrochemical explosion-proof humanoid robot. Background Technology

[0002] The petrochemical industry environment, as a typical high-risk workplace, presents multiple safety threats, including toxic gas leaks, flammable gas accumulation, and corrosive substance distribution, placing extremely high demands on robot safety navigation. Traditional robot path planning methods are mainly designed for general industrial environments and lack in-depth consideration of the special hazards of the petrochemical environment, making it difficult to effectively handle complex constraints such as changes in gas concentration, static electricity accumulation, and explosion-proof requirements.

[0003] Existing path planning technologies for hazardous environments mostly employ static maps and fixed constraints, failing to adjust paths based on dynamic changes in environmental hazard levels and lacking a synergistic consideration of the robot's own dynamic characteristics and environmental safety requirements. In particular, they exhibit significant shortcomings in handling multiple overlapping hazards, balancing path safety and execution efficiency, and balancing motion stability and explosion-proof requirements. This results in planned paths that are either overly conservative, impacting operational efficiency, or lack sufficient safety margins, posing potential risks. Therefore, a method is urgently needed to address at least one of these problems. Summary of the Invention

[0004] This invention discloses a trajectory planning method, apparatus, and equipment for petrochemical explosion-proof humanoid robots, aiming to solve the technical problems of insufficient safety and poor adaptability of traditional trajectory planning methods in hazardous petrochemical environments. By establishing a coupled analysis model of path skeleton features and robot motion characteristics, a comprehensive assessment of multiple path risk factors, such as abrupt changes in path safety levels, geometric risk concentration, and connection fragility, is conducted to construct a safety constraint system. Using motion primitive decomposition and stability filtering techniques, combined with dynamic envelope analysis and execution probability field construction, a unified assessment of trajectory safety and execution feasibility is achieved, generating a final planned trajectory that balances safety, feasibility, and executability.

[0005] The first aspect of this invention proposes a trajectory planning method for a petrochemical explosion-proof humanoid robot, comprising the following steps:

[0006] Collect petrochemical environment gas concentration data, robot joint motion parameters, and target position information; use the gas concentration data to identify hazardous areas and generate a restricted area map; determine safe nodes from the restricted area map; and determine the joint range of motion based on the joint motion parameters.

[0007] Based on the target location information and the distribution of safety nodes, a path search is performed to construct an initial path skeleton. A safety analysis is then performed on the initial path skeleton to identify local dangerous sections. Based on the local dangerous sections, motion constraints and attitude balance requirements are generated.

[0008] Based on the initial path skeleton, major hazard segments are identified, and a threat level assessment is performed on the major hazard segments to generate risk weights. The risk weights are then used for path replanning to generate toxicity avoidance paths.

[0009] Based on the motion constraints and the joint range of motion, a motion primitive library is generated. According to the posture balance requirements, the motion primitive library is subjected to stability filtering to generate stable motion primitives. The stable motion primitives are then woven into paths between the safe nodes to form a set of travel schemes.

[0010] The peak values ​​of joint loads in the movement scheme are obtained, a dynamic envelope is constructed based on the peak values ​​of joint loads, and the execution fitness is determined according to the dynamic envelope. The execution fitness is then projected onto spatial coordinates to form an execution probability field.

[0011] A three-dimensional decision space is constructed based on the poison avoidance path, the set of travel schemes, and the execution probability field. The optimal convergence point is found in the three-dimensional decision space, and the final planned trajectory is generated based on the optimal convergence point.

[0012] A second aspect of this invention provides a trajectory planning device for a petrochemical explosion-proof humanoid robot, comprising:

[0013] The environmental perception module is used to collect petrochemical environment gas concentration data, robot joint motion parameters and target position information, mark dangerous areas from the gas concentration data to generate a restricted area map, determine safe nodes from the restricted area map, and determine the joint range of motion based on the joint motion parameters.

[0014] The path analysis module is used to perform path search and construct an initial path skeleton based on the target location information and the distribution of safety nodes, perform safety analysis on the initial path skeleton to identify local dangerous sections, and generate motion constraints and attitude balance requirements based on the local dangerous sections.

[0015] The risk avoidance module is used to identify major hazard segments based on the initial path skeleton, perform threat level assessment on the major hazard segments to generate risk weights, and use the risk weights to perform path replanning to generate toxicity avoidance paths.

[0016] The motion generation module is used to generate a motion primitive library based on the motion constraints and the joint range of motion, perform stability filtering on the motion primitive library according to the posture balance requirements to generate stable motion primitives, and weave the stable motion primitives into a travel scheme set by path weaving between the safe nodes.

[0017] The fitness evaluation module is used to obtain the peak value of the joint load in the movement scheme, construct a dynamic envelope based on the peak value of the joint load, determine the execution fitness according to the dynamic envelope, and project the execution fitness into spatial coordinates to form an execution probability field.

[0018] The trajectory decision module is used to construct a three-dimensional decision space based on the poison avoidance path, the set of travel schemes and the execution probability field, find the optimal convergence point in the three-dimensional decision space, and generate the final planned trajectory based on the optimal convergence point.

[0019] A third aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a trajectory planning method for a petrochemical explosion-proof humanoid robot disclosed in the first aspect.

[0020] The beneficial effects of this invention are reflected in the following aspects: First, it establishes a correlation mechanism between gas concentration in petrochemical environments and robot safe navigation. Through real-time gas concentration monitoring and hazardous area calibration, combined with electrostatic accumulation tendency analysis and joint friction avoidance strategies, it helps prevent ignition and explosion risks during robot movement, improving the inherent safety level of robot operations in petrochemical environments. Second, it constructs a complete technical link from hazardous segment identification to toxicity avoidance path generation. Through path risk feature quantification and threat level assessment, a standardized risk weight system is established, enabling trajectory planning to accurately respond to different toxicity threats. Simultaneously, through the construction of an action primitive library and stability filtering, the execution stability and safety reliability of avoidance actions are ensured. Finally, by employing three-dimensional decision space fusion and multi-level search technology, safety, feasibility, and executability are uniformly incorporated into the decision framework. Through dynamic envelope analysis and execution probability field construction, a deep integration of robot dynamic characteristics and environmental safety requirements is achieved. The generated final planned trajectory effectively balances execution efficiency and action stability while ensuring high safety standards.

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

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

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

[0024] Figure 1 This is a flowchart illustrating a trajectory planning method for a petrochemical explosion-proof humanoid robot according to the present invention.

[0025] Figure 2 This is a structural block diagram of a petrochemical explosion-proof humanoid robot trajectory planning device according to the present invention.

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

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

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

[0029] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0030] The technical solutions of the embodiments of this application will be described below.

[0031] like Figure 1 As shown, this embodiment of the invention provides a trajectory planning method for a petrochemical explosion-proof humanoid robot, including the following steps S110-S160:

[0032] Step S110: Collect petrochemical environment gas concentration data, robot joint motion parameters and target position information, mark dangerous areas from gas concentration data to generate a restricted area map, determine safe nodes from the restricted area map, and determine the joint range of motion based on joint motion parameters.

[0033] Specifically, the system collects petrochemical environment gas concentration data, robot joint motion parameters, and target position information. Petrochemical environment gas concentration data acquisition is achieved through multiple types of gas sensors deployed around the robot. The sensor configuration includes three main types of equipment: combustible gas detectors, toxic gas detectors, and oxygen meters. The combustible gas detectors use catalytic combustion sensors with a detection range of 0-100% LEL (lower explosive limit), a response time of less than 10 seconds, and an accuracy of ±3% LEL. The toxic gas detectors target common harmful petrochemical gases such as hydrogen sulfide, carbon monoxide, and benzene compounds, employing electrochemical principles and achieving ppm-level accuracy. The oxygen meter measures from 0-25% Vol with an accuracy of ±0.1% Vol, used to monitor oxygen deficiency in confined spaces. The sensors are arranged in a three-dimensional array, with sensor nodes installed in front of, to the left and right sides, and above the robot, forming a 360-degree omnidirectional monitoring coverage. The data acquisition frequency is set to 1Hz to ensure timely response to environmental changes. Gas concentration data preprocessing includes temperature compensation, humidity correction, and baseline drift correction to eliminate the influence of environmental factors on measurement accuracy. Robot joint motion parameters are acquired through a built-in encoder and inertial measurement unit (IMU), including four dimensions: joint angle, angular velocity, angular acceleration, and joint torque. The encoder has a resolution of 0.1° and a sampling frequency of 100Hz, providing high-precision position feedback. The IMU integrates a three-axis gyroscope and a three-axis accelerometer to measure the dynamic characteristics of joint motion. Target position information is obtained through a total station, GPS, or indoor positioning tags, with a positioning accuracy requirement better than 0.5m, providing accurate spatial reference for path planning.

[0034] Hazardous areas are identified and restricted zones are generated from gas concentration data. A tiered standard is used, classifying areas into four levels based on gas concentration levels: safe zone, warning zone, danger zone, and restricted zone. The safe zone is where all gas concentrations are below the safe threshold: combustible gas concentrations are less than 10% LEL, toxic gas concentrations are below occupational exposure limits, and oxygen concentrations are between 19.5% and 23.5%. The warning zone is where some gas concentrations are close to but do not exceed the danger threshold; enhanced monitoring is required, but passage is permitted. The danger zone is where gas concentrations exceed the safe limit: combustible gas concentrations are between 10% and 25% LEL, or toxic gas concentrations exceed short-term exposure limits. The restricted zone is where gas concentrations are severely exceeded: combustible gas concentrations are greater than 25% LEL, or toxic gas concentrations reach levels that immediately threaten life and health. The zone identification algorithm uses contour interpolation to generate a continuous concentration distribution map based on discrete sensor data. Kriging interpolation is selected as the interpolation technique, establishing a spatial autocorrelation model of gas concentrations. Interpolation accuracy is ensured through cross-validation to guarantee reliability. The restricted area map is generated using a rasterized representation with a map resolution of 1m×1m. Each grid cell records the hazard level and main hazard factors of the location.

[0035] Safe nodes are identified from the restricted area map. First, based on hazard level constraints, only grid locations marked as safe zones are considered as candidate nodes. Spatial constraints require that the area within a 3m × 3m radius around each node be a safe zone, ensuring sufficient safety margin for the robot at that location. Connectivity analysis employs graph theory, modeling the safe zone as an undirected graph, where nodes represent safe grids and edges represent connectivity between adjacent grids. Connectivity component identification is achieved using a depth-first search algorithm to determine each independent safe zone block. Node importance assessment comprehensively considers node centrality, connectivity, and spatial distribution characteristics. Centrality reflects the node's central position in the connected graph, while connectivity represents the number of other nodes reachable from that node. Spatial distribution analysis ensures that safe nodes are evenly distributed across the map, avoiding overly concentrated or sparse nodes. Node optimization uses a greedy algorithm to select the optimal node set, with the objective function comprehensively considering node coverage, connectivity, and distribution uniformity. Node labeling adopts a hierarchical numbering scheme, with primary safe nodes labeled S + numeric number and secondary nodes labeled SS + numeric number.

[0036] The range of motion of joints is determined based on joint motion parameters. The range of motion is the permissible variation range of each joint angle while ensuring operational efficiency and safety. Parameter analysis first involves statistically analyzing historical joint motion data to calculate basic statistics such as the mean, variance, maximum, and minimum values ​​of each joint angle. Motion pattern recognition uses cluster analysis to classify joint motion into four typical modes: walking, operating, obstacle avoidance, and standby. Each mode corresponds to different joint motion characteristics and range requirements. The range of motion is calculated using probabilistic statistical methods, with a 95% confidence interval as the normal range of motion for the joint: θ_range=[μ-1.96σ,μ+1.96σ], where μ is the mean of the joint angle and σ is the standard deviation. A safety margin is set to consider the special requirements of the petrochemical environment, leaving a 20% safety margin on top of the statistical range to avoid joint motion reaching extreme positions. Joint coupling analysis studies the motion coordination relationship between different joints, identifying strongly coupled joint pairs through correlation analysis. Workspace calculation is based on forward kinematics to determine the reachable space of the robot's end effector within a given joint range. Range optimization employs a multi-objective optimization method to minimize joint motion amplitude while meeting operational requirements, thereby reducing mechanical wear and energy consumption. Constraints include multiple constraints such as joint physical limits, collision avoidance, and singular configuration avoidance.

[0037] Step S120: Based on the target location information and the distribution of the safety nodes, a path search is performed to construct an initial path skeleton. The initial path skeleton is then subjected to safety analysis to identify local dangerous segments. Based on the local dangerous segments, motion constraints and attitude balance requirements are generated.

[0038] Specifically, an initial path skeleton is constructed based on the target location information and the distribution of safety nodes. The path search is built upon a connected graph of safety nodes, with safety nodes as vertices and safe passages between nodes as edges. Edge weights comprehensively consider distance cost and safety index. The search algorithm employs the A* algorithm, with a heuristic function combining Euclidean distance and hazard assessment. The cost function is designed as: f(n) = g(n) + h(n) + s(n), where n is the search node, g(n) is the actual cost, h(n) is the heuristic cost, and s(n) is the safety cost. The safety cost is calculated based on the safety level of the safety nodes; the lower the level, the higher the cost. The path search process uses a bidirectional search strategy, starting simultaneously from the beginning and end points, converging at the intermediate position to form a complete path. The search space is limited to the safe area, avoiding paths passing through danger zones and restricted areas. Path smoothing employs spline curve fitting technology to eliminate jagged corners in the search path, generating a smooth trajectory suitable for robot movement. The initial path skeleton is represented by a sequence of key points, including the path start point, end point, turning points, and important safety nodes. The skeleton data structure records the coordinates, arrival time, direction of movement, and local security level of each key point. Path segmentation divides the long path into multiple sub-segments, with each sub-segment's length controlled within 50m to facilitate subsequent analysis.

[0039] Safety analysis is performed on the initial path skeleton to identify local hazardous sections. Path safety is assessed based on the geometric features of the initial path skeleton, the safety node levels traversed, and path connection relationships. Path geometric analysis identifies geometric features such as sharp turns, long straight sections, and densely populated node areas. Sharp turns may increase movement risk, long straight sections may lack alternative paths, and densely populated node areas may have path conflicts. Safety node level analysis assesses the distribution of safety levels of nodes along the path direction, identifying road sections with low safety levels and areas with drastic changes in safety levels. Path connection relationship analysis checks the stability of the connection between the path and safety nodes, identifying vulnerable road sections with weak connections or those dependent on a single node. Hazardous section identification uses a risk value accumulation method, weighting and summing geometric risk, node risk, and connection risk: Risk = w1 × Geometry + w2 × NodeLevel + w3 × Connection, where Geometry is the geometric feature risk value, NodeLevel is the node level risk value, and Connection is the connection relationship risk value. The weighting coefficients are set according to path planning requirements: w1 = 0.3, w2 = 0.5, and w3 = 0.2. The risk threshold is set at the 80th percentile of the risk distribution; road segments exceeding the threshold are marked as locally hazardous segments. Hazardous segments are classified into three categories: high-risk, medium-risk, and special-risk segments, each with its own corresponding processing strategy. Geometric feature extraction of hazardous segments includes information such as length, width, center location, and primary risk type.

[0040] In some embodiments, generating motion constraints and attitude balance requirements based on the local hazardous segment includes: identifying regions prone to electrostatic accumulation from the local hazardous segment; establishing a charge flow path map based on the electrostatic accumulation region; setting joint friction avoidance points using the charge flow path map; and establishing motion constraints and attitude balance requirements based on the joint friction avoidance points.

[0041] Identifying regions prone to static electricity accumulation within local hazardous sections. A local hazardous section feature analysis method is employed, comprehensively considering key parameters such as hazardous section type, geometric complexity, risk level, and spatial distribution density. Hazard type analysis categorizes local hazardous sections into three types based on risk nature: geometric risk, node risk, and connection risk. Geometric risk sections, due to their complex and tortuous paths, frequent robot directional changes, and increased frictional contact, exhibit a high tendency for static electricity accumulation. Node risk sections, passing through low-safety-level nodes, have smaller environmental safety margins, resulting in greater safety hazards from static electricity accumulation. Connection risk sections, due to unstable path connections, may require frequent adjustments to the movement trajectory, increasing the chance of static electricity generation. Geometric complexity assessment is based on the length, width, turning frequency, and angle variations of the hazardous section; higher complexity corresponds to a higher tendency for static electricity accumulation. Straight hazardous sections exceeding 30 meters in length have a higher accumulation tendency due to continuous friction; narrow hazardous sections less than 2 meters in width increase contact opportunities due to spatial constraints; and tortuous hazardous sections with a turning frequency greater than 3 times / 10 meters increase friction due to motion complexity. The risk level mapping maps the high, medium, and special risk levels of local hazardous sections to the strong, medium, and weak tendency levels of static electricity accumulation. Spatial distribution density analysis identifies areas with densely distributed hazardous sections; these densely distributed areas show a significantly enhanced tendency to accumulate static electricity due to the superposition of multiple risks. The accumulation tendency assessment employs a weighted scoring method, comprehensively calculating factors such as hazardous section type, geometric complexity, risk level, and distribution density according to a weight ratio of 4:3:2:1. The scoring threshold is set at the 70th percentile of the score distribution; local hazardous sections exceeding this threshold are marked as areas with a tendency to accumulate static electricity.

[0042] A charge flow path map is established based on regions prone to electrostatic accumulation. Charge generation sources are identified, including major sources such as robot joint friction, foot-to-ground contact, and relative motion between the robot body and air. The generation rate is estimated using the triboelectric formula, considering factors such as contact area, relative velocity, and material properties. The charge transport path is described using a resistive network model, dividing the space into resistive grids. The resistance value of each grid is calculated based on the material resistivity and geometric dimensions. The transport direction is determined by the potential gradient, with charge flowing from high-potential regions to low-potential regions. Accumulation nodes are identified through charge balance analysis; accumulation nodes are formed when the charge inflow rate exceeds the outflow rate. Dissipation paths include natural dissipation mechanisms such as leakage to the ground, air ionization, and corona discharge. The path map data structure uses a directed graph, where nodes represent charge accumulation points, edges represent charge flow paths, and edge weights represent resistance magnitudes. The resolution of the path map is set to the spatial grid size, typically 1m × 1m to ensure computational accuracy.

[0043] For example, setting joint friction avoidance points using the charge flow path diagram includes: extracting high-risk charge convergence nodes from the charge flow path diagram; establishing friction contact exclusion zones based on the high-risk charge convergence nodes; generating joint motion offset points using the friction contact exclusion zones; and converting the joint motion offset points into joint friction avoidance points.

[0044] High-risk charge accumulation nodes are extracted from the charge flow path diagram. Identification of high-risk charge accumulation nodes is based on charge density distribution and potential gradient analysis to determine the spatial locations with the most severe charge accumulation and the highest risk of discharge. First, the charge density of each node in the path diagram is calculated using the charge balance equation: Where D is the electric displacement vector, This refers to charge density. The criteria for identifying high-risk nodes include three conditions: charge density exceeding the safety threshold, potential gradient greater than the critical value, and discharge probability higher than the risk limit. The charge density threshold is set according to the explosion-proof requirements of the petrochemical environment, generally taking the charge density corresponding to the minimum ignition energy. The critical value of the potential gradient is determined based on the air breakdown strength; the breakdown strength of dry air is approximately 3 × 10⁻⁶. 6 V / m. The discharge probability is calculated using statistical methods, taking into account factors such as ambient humidity, temperature, and air pressure. Node hazard level assessment employs a fuzzy comprehensive evaluation method, converting quantitative indicators into qualitative risk levels. High-risk nodes are ranked according to their comprehensive risk value from highest to lowest, with the highest-risk nodes being addressed first. The impact range of a node is determined through electric field distribution calculations; all spaces within the impact range require special protection. Temporary stability analysis assesses the temporal stability of high-risk nodes; short-lived nodes require temporary protection, while long-term nodes require permanent protection measures.

[0045] Frictional contact exclusion zones are established based on high-risk charge convergence nodes. These zones are spatial areas where robot joints should not engage in frictional contact to avoid electrostatic discharge and potential explosion risks. The exclusion zones are established using a safety distance envelope method, creating spherical or ellipsoidal spatial exclusion zones centered on the high-risk charge convergence nodes. Safety distance calculations are based on the attenuation law of electric field strength and minimum safety distance requirements. Where Q is the node charge, E_threshold is the safe electric field strength, and k is the safety factor. The shape of the restricted area is determined based on the electric field distribution characteristics; isotropic electric fields correspond to spherical restricted areas, and anisotropic electric fields correspond to ellipsoidal restricted areas. Restricted areas are classified into three levels: absolute restricted areas, restricted areas, and warning areas. Absolute restricted areas strictly prohibit any contact; restricted areas allow contact under special protection; and warning areas require enhanced monitoring. Multiple high-risk node restricted areas are merged using a union operation, with overlapping areas processed according to the strictest restriction level. The fuzzy processing of restricted area boundaries uses a gradient function to avoid abrupt constraints at the boundaries. Restricted area parameters include geometric descriptions such as center coordinates, radius or axis length, and direction angle. The restricted area data structure records information such as center coordinates, geometric parameters, restriction level, and effective time. Visualization uses a three-dimensional transparent volume, with different colors corresponding to different restriction levels. Joint motion offset points are generated using friction contact restricted areas. These joint motion offset points are new target points set to avoid friction contact restricted areas based on the original motion trajectory. Offset points are generated using geometric calculation methods. Under the premise of satisfying restricted area avoidance constraints, the alternative trajectory with the smallest deviation from the original trajectory is sought. The calculation objective is to minimize the weighted sum of squared trajectory deviations, with weights allocated according to joint importance and motion accuracy requirements. Constraints include restricted area avoidance constraints, joint limitation constraints, motion continuity constraints, and task completion constraints. Avoidance constraints are expressed in inequality form, requiring that the distance between the offset point and the center of the restricted area be greater than a safe distance. The offset direction is selected using a gradient method, offsetting away from the center of the restricted area. The offset magnitude is determined through a binary search to find the minimum offset amount that satisfies the constraints. Multi-joint coordinated offset considers the kinematic coupling relationship between joints, using the Jacobian matrix to analyze the impact of joint movement on the end effector position. The timing of the offset points is synchronized with the overall motion planning to ensure smooth execution of the offset action. Trajectory reconstruction uses spline interpolation to generate smooth motion trajectories between offset points.

[0046] The joint motion offset points are converted into joint friction avoidance points. First, inverse kinematics calculations are performed to convert the offset point coordinates in Cartesian space into angle values ​​in joint space. A numerical iterative method is used, solving a system of nonlinear equations through Newton-Raphson iteration. For multiple solutions, an optimality criterion is used to select solutions that are close to the current joint angle and satisfy physical constraints. Reachability analysis checks whether the inverse solution results are within the joint's physical limits; solutions exceeding these limits are marked as unreachable. Singularity detection uses condition number analysis of the Jacobian matrix to prevent the robot from entering singular configurations. Avoidance point accuracy is evaluated through forward kinematics calculations to determine the deviation between the end effector position and the target position corresponding to the inverse solution results. Avoidance points that do not meet accuracy requirements are iteratively improved by adjusting joint angles to enhance accuracy. Avoidance point time stamps record the execution time and duration of each avoidance point, providing timing information for motion control. The avoidance strategy storage includes complete information such as the original trajectory, offset trajectory, avoidance point coordinates, and execution timing. Coordinate transformation uses standard robot kinematic transformation matrices to ensure transformation accuracy and consistency.

[0047] Motion constraints and posture balance requirements are established based on the joint friction avoidance points. The motion constraints employ a hierarchical constraint structure, including both hard and soft constraints. Hard constraints are strict safety requirements that must be met, including avoidance point position constraints, velocity limit constraints, and acceleration limit constraints. Avoidance point position constraints require joint movements to strictly adhere to the avoidance trajectory, with a positional deviation not exceeding ±2cm. Velocity limit constraints restrict joint angular velocities to a safe range, with angular velocities within the avoidance area not exceeding 50% of their normal value. Acceleration limit constraints prevent abrupt changes in motion, limiting angular acceleration to 30% of its normal value. Soft constraints are performance requirements, including motion smoothness constraints, energy consumption control constraints, and task efficiency constraints. Posture balance requirements consider the impact of avoidance actions on the robot's overall stability, and corresponding center of gravity control and support strategies are formulated. Center of gravity offset compensation is achieved by adjusting the posture of non-avoidance joints, maintaining the overall center of gravity within the support polygon. Support strategies include increasing the support base, lowering the center of gravity height, and adjusting gait parameters. Balance control parameters include key indicators such as zero-torque point location, stability margin, and posture angles. The mathematical representation of the constraints uses a combination of linear and nonlinear inequalities, and feasible motion control parameters are determined through constraint-solving algorithms. Constraint priority is set with safety constraints having the highest priority, followed by balance constraints, and performance constraints having the lowest priority.

[0048] Step S130: Identify major hazard segments based on the initial path skeleton, conduct threat level assessment on major hazard segments to generate risk weights, and use the risk weights to perform path replanning to generate toxicity avoidance paths.

[0049] Specifically, critical hazard sections are identified based on the initial path skeleton. These critical hazard sections are the most dangerous segments with the highest threat level and the widest impact range among local hazard sections. Following the trajectory of the initial path skeleton, critical hazard sections are identified based on the spatial distribution characteristics of the path skeleton and the safety node level information. Critical hazard sections are those sections on the path that simultaneously meet multiple high-risk conditions and whose threat level exceeds the safety threshold; these sections pose a serious safety threat to the robot and its surrounding environment. The identification method employs multi-factor comprehensive analysis, primarily considering four key factors: abrupt changes in path safety level, concentration of geometric risks along the path, path connectivity vulnerability, and proximity of the path to restricted areas. Safety level abrupt change analysis identifies sections on the path where the safety node level drops sharply; sections with a drop of more than two levels are marked as high-risk. Path geometric risk concentration analysis identifies dangerous sections with dense sharp turns and long distances without alternative paths; sections with turning angles greater than 90 degrees and turning intervals less than 20 meters are geometrically high-risk areas. Connectivity vulnerability analysis identifies sections relying on a single connection or with unstable connections; sections with a connectivity degree less than 2 or a connection weight less than 50% of the average are considered connectivity vulnerable areas. The proximity analysis of restricted areas identifies road segments that are too close to dangerous areas on the restricted area map identified by S110. Road segments less than 10 meters from the boundary of the restricted area are considered to be approaching high-risk zones. Threat concentration calculation uses spatial overlay analysis, spatially superimposing multiple path risk factors. Areas where the overlay value exceeds a set threshold are designated as major danger sections. Danger section feature extraction focuses on the maximum threat intensity and the main path risk types.

[0050] In some embodiments, the step of assessing the threat level of the major hazard section to generate risk weights includes: determining a quantitative scale based on the chemical hazard intensity assessment of the major hazard section; setting a hazard calibration coefficient according to the quantitative scale; and using the hazard calibration coefficient to convert the threat level of the major hazard section to generate risk weights.

[0051] The quantitative scale for assessing chemical hazard intensity based on major hazard sections was determined. A comprehensive analysis method using four key factors for major hazard sections was employed, with key parameters including the severity of safety level abrupt changes, geometric risk concentration, connectivity vulnerability strength, and proximity to restricted areas. The severity of safety level abrupt changes was assessed based on the magnitude and frequency of node level declines; sections with a decline exceeding three levels were marked as extremely high abrupt changes, two to three levels as high abrupt changes, one to two levels as medium abrupt changes, and less than one level as low abrupt changes. Geometric risk concentration was assessed considering sharp turn density and path complexity; sections with a sharp turn density greater than 5 per 100 meters and a turning angle greater than 120 degrees were classified as extremely high geometric risk. Connectivity vulnerability strength was assessed based on connectivity redundancy and connectivity weight distribution; sections with a connectivity degree of 1 and a weight below 70% of the average were classified as extremely vulnerable. Proximity to restricted areas was assessed based on the closest distance to the restricted area and its impact range; sections less than 5 meters away were classified as extremely high proximity risk. A normalization method was used to establish the quantitative scale, converting path risk data of different magnitudes into a unified quantitative scale. The scale range is set from 1 to 10, where 1 represents the lowest risk and 10 represents the highest risk. Quantitative conversion uses a piecewise linear function, with different conversion weights for different risk factors. Multi-factor synthesis employs a weighted arithmetic mean method, with weights allocated according to factor importance: safety level mutation 40%, geometric risk 25%, connectivity vulnerability 20%, and proximity 15%.

[0052] Hazard calibration coefficients are set according to the quantification scale. The hazard calibration coefficient K_calibration is used to adjust the weighting of path hazard intensity in risk assessment, establishing an accurate mapping relationship between the quantification scale and the actual risk level. The calibration coefficients are set based on the numerical distribution characteristics and statistical laws of the quantification scale, using a piecewise function approach to determine the calibration coefficients corresponding to different quantification scale intervals. The calibration coefficient for extremely high-hazard path segments (quantification scales 8-10) is set to 1.2-1.5, reflecting its extraordinary hazard. The calibration coefficient for high-hazard path segments (quantification scales 6-8) is 1.0-1.2, and for moderate-hazard path segments (quantification scales 4-6) it is 0.8-1.0. The coefficient for low-hazard path segments (quantification scales 2-4) is 0.6-0.8, and for slightly hazardous path segments (quantification scales 0-2) it is 0.4-0.6. The coefficient distribution uses a linear interpolation method to ensure a smooth transition between different quantification scale intervals. The stability of the calibration coefficients is evaluated through analysis of variance; coefficients with a variance less than 0.1 are considered stable and usable.

[0053] Risk weights are generated by converting the threat level of major hazard sections using hazard calibration coefficients. Threat level conversion transforms qualitative hazard descriptions and quantitative hazard indicators into standardized risk weight values. A multi-step calculation method is employed: first, a base risk value is calculated, and then adjustments are made using calibration coefficients. The base risk value R_base is calculated based on the comprehensive characteristics of the major hazard section: R_base = W1 × S_level + W2 × G_risk + W3 × C_weak + W4 × P_close, where S_level is the safety level abrupt change value, G_risk is the geometric risk value, C_weak is the connectivity vulnerability value, and P_close is the proximity value. The weight coefficients W1-W4 are consistent with the weights in the quantitative scale calculation. The adjusted risk value R_adjusted is obtained by correcting with calibration coefficients: R_adjusted = R_base × K_calibration, where K_calibration is the hazard calibration coefficient. Threat levels are classified based on adjusted risk values, using a five-level classification standard: extremely high threat (R≥8.0), high threat (6.0≤R<8.0), medium threat (4.0≤R<6.0), low threat (2.0≤R<4.0), and minor threat (R<2.0). The risk weight W is generated through normalization, mapping the risk value to the 0-1 interval: W=R_adjusted / R_max, where R_max is the system's maximum risk value.

[0054] In some embodiments, the step of using the risk weights to perform path replanning to generate a toxicity avoidance path includes: constructing a threat avoidance region and a safety guidance region based on the risk weights; generating a path rejection factor using the threat avoidance region, the path rejection factor including toxicity rejection degree and explosive / flammable rejection degree; and performing path fusion with the path rejection factor and the safety guidance region to generate a toxicity avoidance path.

[0055] Threat avoidance zones and safety guidance zones are constructed based on risk weights. The construction of these zones provides spatial constraints and guidance information for path replanning, enabling effective avoidance of hazardous areas and full utilization of safe areas. Threat avoidance zones are spatial areas where the risk weight exceeds a set threshold and the robot needs to avoid entering. Zone division uses the risk weight contour line method, converting the continuous risk weight distribution into discrete zone identifiers. The avoidance zone threshold is set at the 75th percentile of the risk weight to ensure effective identification of high-risk areas. Zone expansion considers the robot's physical size and safety margin, extending 2 meters outward from the original high-risk zone as a safety buffer. Safety guidance zones are spatial areas with lower risk weights suitable for robot passage, serving as preferred pathways for path planning. Guidance zone identification is based on reverse analysis of risk weights; areas with risk weights below the 25th percentile are marked as safety guidance zones. Regional connectivity analysis uses graph theory to identify connected safe zone blocks, providing a feasible pathway network for path search. The spatial distribution characteristics of threat avoidance zones and safety guidance zones affect the effectiveness of subsequent path planning, and the accuracy of zone boundaries determines the safety of the avoidance path.

[0056] Path rejection factors are generated using threat avoidance zones. The calculation of toxicity rejection considers the size and shape complexity of the threat avoidance zone; larger and more irregularly shaped zones have higher rejection rates. Rejection rate calculation uses a distance decay function; the closer to the center of the threat avoidance zone, the higher the rejection rate. Zone density analysis considers the number of threat avoidance zones per unit area; higher density zones have increased rejection rate weights. The calculation of combustion and explosion rejection considers the spatial distribution pattern and connectivity characteristics of the threat avoidance zones; large, interconnected avoidance zones have higher combustion and explosion rejection rates. Zone boundary complexity analysis assesses the tortuosity of the avoidance zone boundaries; complex boundaries increase navigation difficulty. Spatial gradient analysis calculates the risk weight gradient of the threat avoidance zone boundaries; larger gradients indicate drastic risk changes. The spatial distribution of the rejection factor uses a distance decay model; the closer to the threat center, the higher the rejection factor. Rejection factor standardization unifies the numerical range to 0-10, facilitating integrated calculations with other path planning parameters.

[0057] The path fusion algorithm generates a hazard avoidance path by fusing the path repulsion factor with the safety guidance area. It comprehensively considers the hindering effect of the repulsion factor and the guiding effect of the safety guidance area to generate an optimal path that effectively avoids danger and efficiently reaches the target. The fusion calculation uses the potential field method, converting the repulsion factor into a repulsive potential field and the safety guidance area into an attractive potential field. The strength of the repulsive potential field is proportional to the value of the repulsion factor, forming a strong repulsive potential in areas with high repulsion factors. The strength of the attractive potential field is related to the safety level and the distance from the target; areas with high safety levels and close proximity to the target generate a strong attractive potential. Potential field superposition uses a vector synthesis method, calculating the vector sum of the repulsive and attractive potentials at each spatial location. Path search proceeds along the gradient direction of the synthesized potential field, automatically avoiding high potential energy areas (dangerous areas) and tending towards low potential energy areas (safe areas). Path smoothing uses spline curve fitting technology to eliminate path discontinuities caused by potential field gradient changes. Path feasibility checks include geometric constraint checks, kinematic constraint checks, and safety constraint checks. Geometric constraints ensure the path does not cross physical obstacles, while kinematic constraints ensure the robot can execute path movements. The quality assessment of hazard avoidance routes employs a multi-objective evaluation method, comprehensively considering route length, safety level, travel time, and energy consumption. Route data output includes a route coordinate sequence, risk level distribution, recommended travel speed, and safety protection requirements.

[0058] Step S140: Generate a motion primitive library based on motion constraints and joint range of motion. Perform stability filtering on the motion primitive library according to posture balance requirements to generate stable motion primitives. Weave the stable motion primitives into a travel scheme set by weaving paths between safe nodes.

[0059] Specifically, a motion primitive library is generated based on the aforementioned motion constraints and joint range of motion. Motion primitives are the smallest motion units for a robot to perform a specific function, possessing characteristics such as independence, composability, and repeatability. A constraint-oriented generation method is used to construct the primitive library, categorizing motion primitives into four main types based on the type of motion constraints: avoidance constraint type, velocity-limited type, acceleration-limited type, and position constraint type. Avoidance constraint primitives are generated based on avoidance point position constraints, ensuring that joint movements strictly follow the avoidance trajectory and that positional deviations do not exceed a set range. Velocity-limited primitives are generated based on angular velocity limit constraints, ensuring that angular velocities within the avoidance area do not exceed a specified proportion of normal values. Acceleration-limited primitives are generated based on acceleration limit constraints, preventing abrupt changes in motion and limiting angular acceleration within a safe range. Position constraint primitives are generated based on joint range of motion constraints, ensuring that all joint movements are within a defined effective range. The parameterized representation of the primitives uses a joint space description method, with each primitive containing parameters such as joint angle sequence, angular velocity constraints, angular acceleration limits, and execution time. Constraint mapping applies hard and soft constraints to the primitive generation process, with hard constraints serving as mandatory boundary conditions and soft constraints as optimization objectives. Joint range of motion checks utilize the aforementioned θ_range parameter, automatically discarding actions exceeding the range. Primitive combination rules establish connection methods and transformation conditions between different primitives based on constraint compatibility, supporting the construction of complex action sequences. The primitive library data structure is organized using constraint classification: the top layer represents constraint types, the middle layer represents constraint strength, and the bottom layer represents specific action primitives. Each primitive contains complete information including constraint source, parameter list, execution conditions, and a list of compatible primitives.

[0060] In some embodiments, the step of generating stable motion primitives by performing stability filtering on the motion primitive library according to the attitude balance requirements includes: sorting the motion primitive library according to the attitude balance requirements to form a stability gradient; extracting stability peak points and stability valley points from the stability gradient; constructing a motion stability system by partitioning the motion stability system with the stability peak points and stability valley points as boundaries; and performing optimization judgment on the motion stability system to generate stable motion primitives.

[0061] The motion primitive library is ranked and sorted according to attitude balance requirements to form a stability gradient. Stability evaluation criteria are established based on key indicators such as zero-moment point location, stability margin, and attitude angle in the attitude balance requirements. The stability evaluation first transforms the attitude balance requirements into quantifiable scoring criteria. Zero-moment point location evaluation is based on the zero-moment point location requirements, using ZMP trajectory deviation to calculate the score. Stability margin evaluation is based on the stability margin parameters in the attitude balance requirements, using the ratio of actual margin to required margin to calculate the score. Attitude angle evaluation is based on the attitude angle constraints in the attitude balance requirements, using the ratio of angle deviation to the constraint range to calculate the score. The comprehensive stability score adopts the weighting method in the attitude balance requirements: S_total = W_zmp × S_zmp + W_margin × S_margin + W_posture × S_posture, where the weight coefficients W_zmp, W_margin, and W_posture are derived from the parameter settings of the attitude balance requirements. A stability score is calculated for each motion primitive, and the ZMP trajectory, stability margin changes, and attitude angle sequences during primitive execution are obtained through simulation or experimental data. The scoring results are standardized according to the evaluation criteria for attitude balance requirements, facilitating unified comparison and sorting. Sorting uses a quicksort algorithm, arranging all primitives from highest to lowest stability score. Hierarchical processing, based on the hierarchical criteria for attitude balance requirements, divides the sorted primitives into corresponding stability levels. Gradient construction uses a continuous score distribution to form a smooth stability variation curve, reflecting the overall stability characteristics of the primitive library. Primitives with the same score are further sorted using secondary indicators in the attitude balance requirements.

[0062] Stability peaks and valleys are extracted from the stability gradient. Peaks represent the primitive positions with the highest local stability, corresponding to local maxima of the gradient curve. Valleys represent the primitive positions with the lowest local stability, corresponding to local minima of the gradient curve. The extraction algorithm uses first-order derivative analysis to calculate the derivative sequence of the stability gradient; the position where the derivative sign changes corresponds to a peak or valley. Peaks are identified when the derivative changes from positive to negative, and valleys when it changes from negative to positive. Significance filtering retains only significant peaks and valleys by setting threshold conditions, eliminating minor fluctuations caused by noise. The threshold setting is based on statistical analysis of the gradient change amplitude, using the 90th quantile of the amplitude distribution as the significance threshold. Peak-valley correspondence analysis establishes pairing relationships between adjacent peaks and valleys, with each peak-valley pairing a stability change interval. Multi-peak and multi-valley processing considers the complex situation of multiple peaks and valleys in the gradient, using cluster analysis to merge similar peaks and valleys.

[0063] A stability system is constructed by partitioning areas based on stability peaks and valleys. Partition boundaries are set primarily using peaks and valleys as dividing lines, while also considering the continuity of stability scores and the reasonableness of intervals. The partitioning method employs a strategy combining threshold segmentation and boundary adjustment. Initial segmentation is performed based on peaks and valleys, and then the boundary positions are adjusted according to stability distribution characteristics. Intervals are named using stability level identifiers, such as high stability, medium-high stability, medium stability, medium-low stability, and low stability. The number of primitives in each partition is controlled within a reasonable range to avoid any partition being too large or too small, which could affect the system's balance. Partition feature analysis calculates the stability mean, variance, number of primitives, and typical representativeness of each partition. Interval boundary optimization uses boundary sensitivity analysis to adjust boundary positions to maximize the consistency of stability within intervals. The stability system is established with a complete framework including partition hierarchy, interval boundaries, feature parameters, and application rules. Partition standardization establishes unified partitioning standards and naming conventions to support the application of the stability system in different scenarios.

[0064] Stable action primitives are generated through an optimal selection process based on the motion stability system. The selection is based on the partition structure and stability evaluation results of the motion stability system. A multi-level selection mechanism is employed: first, partition-level selection selects partitions with higher evaluation levels in the stability system; then, primitive-level selection chooses primitives from the selected partitions according to their stability scores. The partition selection criteria are based on the partition level of the stability system, prioritizing primitives in high-stability and medium-to-high-stability regions. Primitive selection uses the stability system's score ranking, selecting primitives with higher scores. Constraint compatibility checks ensure that the selected primitives are consistent with the original constraints, maintaining the constraint integrity of the primitive library. Duplicate primitive elimination processes primitives with similar functions in the stability system, retaining the optimal version through stability score comparison. The optimization algorithm is based on the hierarchical structure of the stability system, prioritizing primitives with high stability and high partition levels. The selection quantity is determined according to the distribution characteristics of the stability system, maintaining a reasonable proportion of primitives at each level. The optimization result record includes a list of selected primitives, stability level, partition affiliation, and score information. A stable action primitive database is established, containing complete archives including primitive parameters, stability information, and system affiliation.

[0065] Stable action primitives are used to weave paths between safe nodes to form a set of travel plans. Path weaving is based on the constraint characteristics of stable action primitives and the spatial distribution characteristics of safe nodes. A combination of graph search and action sequence planning is used, with safe nodes as key control points of the path and stable action primitives as the connection method between nodes. The weaving algorithm first establishes a reachability analysis between nodes, determining which safe nodes can be connected through existing stable primitives based on the movement range and execution characteristics of stable action primitives. Connection feasibility assessment is based on the constraint type of the action primitives and the safety level of the safe nodes; only connections with compatible constraints and matching safety levels are considered feasible. Multi-primary combination processing is based on the constraint compatibility of primitives. When a single primitive cannot complete the connection, a constraint-compatible primitive sequence combination method is used. Sequence connection rules are based on the constraint continuity of primitives, ensuring the continuity of adjacent primitives in terms of constraint satisfaction, state transitions, etc. Path weaving uses a depth-first search algorithm, starting from the initial safe node and gradually expanding to adjacent nodes until the target node is reached. The branching strategy retains multiple branches when encountering multiple optional paths, generating diverse travel plans. After weaving, a set of travel plans containing all feasible paths is generated. Each plan records a complete sequence of action primitives and node paths. The number of plans is determined based on the number of actual reachable paths, and includes all feasible plans from the starting node to the target node that satisfy the constraints. The plan data structure records the sequence of action primitives, node paths, constraint satisfaction, and connection relationships.

[0066] Step S150: Obtain the peak value of joint load in the movement scheme, construct a dynamic envelope based on the peak value of joint load, determine the execution fitness based on the dynamic envelope, and project the execution fitness onto spatial coordinates to form an execution probability field.

[0067] The peak joint load of the movement scheme is obtained. The peak joint load is the point where the joint experiences the maximum torque or power consumption during the execution of the action sequence. The load calculation adopts the inverse dynamics method, which calculates the driving torque required by each joint based on the robot's dynamic model and motion parameters. The calculation formula is based on the Lagrange equation: τ=M(q)×q_ddot+C(q,q_dot)×q_dot+G(q), where τ is the joint torque vector, M is the inertia matrix, C is the Coriolis matrix, G is the gravity term, q is the joint angle, q_dot is the joint angular velocity, and q_ddot is the joint angular acceleration. The node path provides the robot's attitude information at different spatial positions, affecting the calculation of the gravity term G(q) and the inertia matrix M(q). The constraint satisfaction information determines the constraint boundary conditions of each scheme during execution, affecting the calculation range of joint torque and the safety margin assessment. The connection relationship information ensures the motion continuity between adjacent action primitives, ensuring a smooth transition of velocity and acceleration at the primitive connection points, and avoiding false load peaks caused by abrupt motion changes. Peak load identification is achieved by peak detection of time-series torque data, using a local maximum search algorithm to locate peak points. Peak selection sets a threshold condition, retaining only significant peaks exceeding 1.5 times the average load. Multi-scheme load analysis is based on the complete data structure of each scheme, extracting action primitive sequences one by one for dynamic modeling, calculating load distribution by combining the spatial location of node paths, considering the impact of constraint satisfaction on load boundaries, using connectivity relationships to ensure computational continuity, and establishing an accurate correspondence between schemes and load peaks.

[0068] In some embodiments, constructing a dynamic envelope based on the joint load peak includes: identifying torque distribution based on the joint load peak to generate a high torque concentration band and a low torque dispersion band; using the high torque concentration band to perform load balancing and allocation on the low torque dispersion band to form a balanced torque band; spatially recombining the balanced torque band to generate a recombined envelope unit; and contour locking the recombined envelope unit to construct a dynamic envelope.

[0069] Torque distribution identification based on joint load peaks generates high-torque concentration zones and low-torque dispersion zones. First, a torque distribution density function is established, and the torque distribution density along the joint space and time axis is calculated using kernel density estimation. High-torque concentration zones are regions with torque density exceeding twice the average density, representing areas with concentrated and high-intensity torque loads. Low-torque dispersion zones are regions with torque density less than 0.5 times the average density, representing areas with dispersed and low-intensity torque loads. Distribution identification employs clustering analysis to group load peaks with similar torque characteristics into the same concentration or dispersion zone. Multi-joint collaborative analysis considers the mutual influence of torque distributions across different joints, identifying torque transmission and compensation relationships between joints. Distribution boundary determination uses a threshold segmentation method, determining the boundaries of concentration and dispersion zones based on the rate of change of torque density. Spatiotemporal distribution mapping establishes a two-dimensional mapping of torque distribution in time and space, supporting multi-dimensional distribution analysis.

[0070] A balanced torque zone is formed by load balancing of a high-torque concentrated zone and a low-torque distributed zone. Load balancing is achieved by transferring a portion of the load from the high-torque concentrated zone to the low-torque distributed zone using torque redistribution technology, thus achieving spatial and temporal balance of torque load. The redistribution principle is based on the characteristics of redundant degrees of freedom in the robot, achieving load redistribution by adjusting the coordination of joint movements. The redistribution algorithm uses constrained least squares to minimize the non-uniformity of torque distribution while satisfying kinematic constraints. The objective function is designed to minimize the torque variance, with constraints including joint limit constraints, velocity constraints, and task completion constraints. Principal component analysis is used to decompose the load in the high-torque concentrated zone into multiple independent components. The load transfer strategy determines the transfer amount and direction based on the load-bearing capacity of the low-torque distributed zone. Transfer path planning establishes the load transfer path from the concentrated zone to the distributed zone, ensuring the smoothness of the transfer process. Balance evaluation is calculated using the statistical characteristics of the torque distribution, including variance, skewness, and kurtosis. The redistribution effect is measured by the degree of improvement in balance; a greater improvement indicates a better redistribution effect. Multiple rounds of adjustments are used to handle complex load distribution situations, and torque balance is gradually improved through iterative adjustments.

[0071] The equilibrium torque band is spatially enclosed and recombined to generate recombined envelope elements. A spatial segmentation and merging strategy is employed. First, the torque band is spatially segmented according to geometric features. Then, segments with similar features are merged to form envelope elements. The segmentation criteria are based on torque gradient and geometric continuity, with locations of abrupt gradient changes serving as segmentation boundaries. The merging criteria consider the shape regularity and functional integrity of the envelope elements, prioritizing the merging of elements with similar shapes and complementary functions. The geometry of the envelope elements is described using regular polyhedra, including basic geometric shapes such as cuboids, cylinders, and ellipsoids. Shape selection is based on the geometric characteristics of the torque distribution; slender distributions correspond to cylinders, while compact distributions correspond to ellipsoids. The element size is determined by the statistical characteristics of the torque distribution, with the envelope volume proportional to the total torque. The recombining process maintains the basic characteristics of the torque distribution, ensuring that the recombined envelope elements accurately reflect the original torque characteristics. Inter-element connections are established through adjacency analysis, examining the interface connection methods and transmission characteristics between adjacent elements.

[0072] A dynamic envelope is constructed by contour locking of recombined envelope elements. Discrete recombined envelope elements are connected to form a continuous dynamic envelope boundary through boundary extraction and surface fitting methods. First, the outer contour of each envelope element is extracted. Then, the contours of adjacent elements are connected using surface fitting techniques to form the overall envelope. Contour extraction employs an isosurface algorithm to generate equitorsion surfaces as contour lines at the boundaries of envelope elements. B-spline surfaces are selected for surface fitting, and smooth connections between different elements are achieved through control point and weight adjustments. Boundary continuity checks ensure geometric and torque continuity of adjacent element contours at the connection points. Envelope closure is achieved by identifying open boundaries of the envelope through topological analysis and using surface extension techniques to close the envelope. The geometric features of the dynamic envelope include the envelope volume and principal axis lengths; these parameters are used for subsequent stability margin calculations. The mathematical representation of the envelope boundary uses parametric equations or implicit functions to support analytical calculations of the envelope.

[0073] Execution fitness is determined based on the dynamic envelope. Execution fitness is a quantitative indicator of the degree of matching between the travel plan and the dynamic envelope. Fitness evaluation mainly considers four key dimensions: envelope geometry, envelope distribution uniformity, envelope boundary stability, and envelope volumetric efficiency. Envelope geometry is evaluated based on the shape regularity and symmetry of the dynamic envelope; the more regular the shape and the better the symmetry, the more reasonable the load distribution. Envelope distribution uniformity is calculated by the variance of the torque distribution within the envelope; the smaller the variance, the more uniform the load distribution. Envelope boundary stability is evaluated based on the curvature change of the envelope boundary; the smoother the boundary, the more stable the load transfer. Envelope volumetric efficiency is calculated by the ratio of envelope volume to total load; the smaller the ratio, the higher the load concentration and the better the efficiency. The overall fitness score (Adaptability) is calculated using a weighted average method: Adaptability = w1 × Geometry + w2 × Uniformity + w3 × Boundary + w4 × Volume, where Geometry is the geometric feature score, Uniformity is the distribution uniformity score, Boundary is the boundary stability score, and Volume is the volumetric efficiency score. Weighting coefficients are assigned based on the importance of the dynamic envelope. Score standardization maps fitness to a 0-1 range, facilitating comparisons between different solutions. A threshold is set to accept fitness scores; solutions exceeding this threshold are marked as executable. Multiple solutions are ranked from highest to lowest fitness score, providing a priority reference for solution selection.

[0074] The execution fitness is projected onto spatial coordinates to form an execution probability field. Spatial projection technology maps the execution fitness from the solution space to the physical space, constructing a continuous field distribution reflecting the execution probability at different spatial locations. Spatial coordinate projection establishes a mapping relationship between execution fitness and the robot's workspace position, with each spatial point corresponding to an execution probability value. The projection method uses radial basis function interpolation to expand the discrete fitness data into a continuous spatial distribution. A Gaussian function is chosen as the interpolation kernel function, and the kernel parameters are adaptively adjusted according to the data distribution characteristics. The execution probability P_exec is calculated based on the fitness normalization process: P_exec = Adaptability / Σ(Adaptability), where P_exec is the execution probability, ensuring that the probability value meets the probability distribution requirements. Spatial filtering technology is used for field distribution smoothing to eliminate discontinuities and outliers in the interpolation process. Probability field gradient calculation identifies the regions with the most drastic changes in execution probability, providing spatial guidance information for path planning. Multi-scale field representation supports probability field analysis at different resolutions; coarse scales are used for global planning, and fine scales are used for local path adjustment. The field data structure uses a regular grid for storage, recording the coordinates and probability value of each grid point.

[0075] Step S160: Construct a three-dimensional decision space based on the toxicity avoidance path, the set of travel schemes, and the execution probability field; find the optimal convergence point in the three-dimensional decision space; and generate the final planned trajectory based on the optimal convergence point.

[0076] Specifically, a three-dimensional decision space is constructed based on hazard avoidance paths, a set of travel plans, and an execution probability field. This three-dimensional decision space is a decision analysis framework encompassing three dimensions: safety, feasibility, and executability, providing comprehensive decision support for robot path planning. The space construction adopts a hierarchical architecture: the bottom layer represents physical space coordinates, the middle layer represents decision attributes, and the top layer represents evaluation indicators. The physical space uses a three-dimensional Cartesian coordinate system, covering the entire working area of ​​the robot, with a spatial resolution of 0.5m × 0.5m × 0.2m to ensure decision accuracy. The safety dimension, based on the risk assessment results of hazard avoidance paths, marks each location in the space with a corresponding safety level and risk weight. The feasibility dimension, based on the distribution of action primitives in the travel plan set, identifies the types of actions that can be executed and their execution difficulty at different locations. The executability dimension, based on the probability distribution of the execution probability field, reflects the robot's success rate at each location. Data fusion employs tensor operations to combine information from the three dimensions into a four-dimensional decision tensor. Where D(x,y,z,t) is a four-dimensional decision tensor, x, y, z are physical space coordinates, t is the time coordinate, Safety(x,y,z) is the safety value, Feasibility(x,y,z) is the feasibility value, and Execution(x,y,z) is the executability value. This represents the tensor product operation. Spatial discretization divides the continuous decision space into a regular grid, with each grid point recording decision attributes and evaluation indicators. Attribute weights are assigned based on task priority: safety weight 0.5, feasibility weight 0.3, and executability weight 0.2. Spatial constraints are established, including boundary constraints, obstacle constraints, and hazardous area constraints, to ensure the physical rationality of the decision space.

[0077] In some embodiments, finding the optimal convergence point in the three-dimensional decision space includes: determining a search depth based on an assessment of the safe convergence difficulty in the three-dimensional decision space, wherein the safe convergence difficulty includes the safety distribution density, the feasibility space connectivity, and the executability probability gradient; determining a priority configuration based on the search depth; formulating a hierarchical search execution plan using the priority configuration; and generating the optimal convergence point through the hierarchical search execution plan.

[0078] The search depth is determined by assessing the difficulty of safe convergence in a three-dimensional decision space. The convergence difficulty is evaluated based on the distribution characteristics and spatial complexity of the three dimensions within the three-dimensional decision space. A quantitative index for search complexity is established to assess the safe convergence difficulty, providing a basis for setting the parameters of the search algorithm. The evaluation method employs decision space feature analysis, primarily considering three key factors: safety distribution density, feasibility space connectivity, and executability probability gradient. Safety distribution density is calculated based on the variance and extreme value distribution of the safety dimension in the three-dimensional decision space; regions with drastic density changes have high convergence difficulty. Feasibility space connectivity is assessed by analyzing the number and size of connected regions in the feasibility dimension of the decision space; regions with low connectivity are difficult to search. The executability probability gradient is based on the gradient distribution of the execution probability field in the decision space; regions with large gradient changes require refined searching. The overall difficulty assessment uses a weighted scoring method: Difficulty = α × Safety_density + β × Feasibility_connectivity + γ × Execution_gradient, where Safety_density is the safety distribution density, Feasibility_connectivity is the feasibility connectivity, and Execution_gradient is the execution gradient, with weighting coefficients of α = 0.4, β = 0.3, and γ = 0.3, respectively. Search depth is determined based on a mapping relationship between difficulty levels; higher difficulty corresponds to deep search, and lower difficulty corresponds to shallow search. Depth levels are divided into five levels: extremely deep search (difficulty 0.8-1.0), deep search (difficulty 0.6-0.8), medium search (difficulty 0.4-0.6), shallow search (difficulty 0.2-0.4), and extremely shallow search (difficulty 0-0.2). Search depth parameters include technical indicators such as the number of search layers, grid density, number of iterations, and convergence accuracy.

[0079] Priority configuration is determined based on search depth. Differentiated priority strategies are developed according to search depth levels. In extremely deep search mode, safety is prioritized highest, with the search algorithm considering safety constraints first, followed by feasibility and executability. In deep search mode, safety and feasibility are given equal weight, ensuring a balance between the two through weight balancing. Medium search mode employs a balanced configuration, with equal weights for the three target dimensions. Shallow search mode focuses on executability, pursuing efficiency while ensuring basic safety. Extremely shallow search mode aims for rapid convergence, sacrificing accuracy requirements for search speed. Priority configuration parameters include target weight vectors, constraint importance ranking, search resource allocation, and termination condition settings. Resource allocation considers computation time, storage space, and search scope, with higher-priority tasks receiving more resources. A dynamic priority adjustment mechanism adjusts priority configurations based on search progress and intermediate results, improving search adaptability. Configuration effectiveness is evaluated through statistical analysis of historical search results, with configurations boasting high success rates receiving higher credibility. The priority configuration data structure records information such as configuration parameters, applicable conditions, performance metrics, and usage history.

[0080] A hierarchical search execution plan is formulated using priority configuration. A hierarchical design approach is adopted, decomposing the complex search problem into several relatively simple sub-problems. The search hierarchy is divided into three levels: global search layer, regional search layer, and local search layer. The global search layer performs a coarse-grained search across the entire decision space to identify potential high-quality regions. The regional search layer performs a medium-precision search within the identified high-quality regions to narrow the search scope. The local search layer performs a high-precision search within the most promising local regions to determine the precise convergence point. Inter-layer information transfer establishes a data flow that transmits search results from upper layers to lower layers, ensuring the continuity of the search process. The search strategy configuration sets the search algorithm, parameter settings, and termination conditions for each layer according to priority. The execution order adopts a serial execution method, executing each layer of search sequentially from global to local. Search parameters are adaptively adjusted for the next layer based on the intermediate results of each layer's search. An exception handling mechanism handles abnormal situations such as no solution, multiple solutions, or convergence failure during the search process. The plan execution monitoring records key information of the search process, including the search trajectory, intermediate results, resource consumption, and performance indicators. The execution plan data includes complete information such as hierarchical structure, algorithm configuration, parameter settings, execution process, and monitoring metrics.

[0081] The optimal convergence point is generated through a hierarchical search execution scheme. Based on the established hierarchical search execution scheme, the optimal position is sought in the three-dimensional decision space according to the predetermined search strategy and hierarchical structure. The hierarchical search algorithm is executed to search for the optimal position layer by layer in the three-dimensional decision space according to the established scheme. In the global search phase, a coarse-grained search is performed according to the global layer configuration of the execution scheme, using the search method and grid density specified in the scheme. In the region search phase, according to the region layer configuration of the execution scheme, the search algorithm and accuracy requirements specified in the scheme are used within the candidate regions. In the local search phase, according to the local layer configuration of the execution scheme, the precise positioning method determined in the scheme is used to quickly converge to the optimal point. Multi-point comparison selects the global optimal point from the local optimal points found in each candidate region as the final optimal convergence point. The convergence judgment criteria are based on the termination criteria of the hierarchical search execution scheme, including search accuracy requirements and position change thresholds. The optimal convergence point is verified by checking its performance in the original three dimensions (safety, feasibility, and executability) to ensure that each dimension meets the minimum requirements. The convergence point record includes information such as spatial coordinates, decision value, scores of each dimension, and search path.

[0082] The final planned trajectory is generated based on the optimal convergence point. A piecewise planning method is employed, with separate path planning for the starting point to the convergence point and for the convergence point to the target point. Path connections are based on the spatial characteristics and decision attributes of the convergence point, using smooth curve fitting techniques to ensure the continuity and differentiability of the trajectory at the convergence point. Trajectory constraints are based on the decision dimension information of the optimal convergence point, including safety requirements, feasibility limitations, and executability conditions. Kinematic constraints are determined based on the feasibility dimension of the convergence point, including maximum velocity and maximum acceleration limits near that point. Dynamic constraints are determined based on the executability dimension of the convergence point, ensuring the trajectory meets the load requirements at that point. Safety constraints are determined based on the safety dimension of the convergence point, requiring the trajectory to maintain the corresponding safety level. Trajectory parameterization uses B-spline curves, and trajectory shape control is achieved through control point and weight adjustments. Time allocation is based on the decision value and spatial location of the convergence point, comprehensively considering the time requirement for reaching the convergence point and execution efficiency. The trajectory resolution is set to a time step of 0.1 seconds and a spatial step of 0.05 meters to ensure control accuracy. Execution parameter labeling, based on the decision attributes of the convergence point, assigns corresponding execution parameters such as velocity, acceleration, and safety level to each point on the trajectory. Trajectory quality is evaluated using a comprehensive assessment of three indicators: shortest length, shortest time, and highest safety. Emergency response plan design outlines contingency plans for unforeseen circumstances during trajectory execution. The final planned trajectory output includes complete information such as trajectory coordinate sequence, time sequence, execution parameters, and safety measures.

[0083] To implement the above method embodiments, a trajectory planning method for a petrochemical explosion-proof humanoid robot is provided to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2This diagram illustrates a structural block diagram of a petrochemical explosion-proof humanoid robot trajectory planning device 200 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The petrochemical explosion-proof humanoid robot trajectory planning device 200 provided in this embodiment includes:

[0084] The environmental perception module 201 is used to collect petrochemical environment gas concentration data, robot joint motion parameters and target position information, mark dangerous areas from the gas concentration data to generate a restricted area map, determine safe nodes from the restricted area map, and determine the joint range of motion based on the joint motion parameters.

[0085] Path analysis module 202 is used to perform path search and construct an initial path skeleton based on the target location information and the distribution of safety nodes, perform safety analysis on the initial path skeleton to identify local dangerous sections, and generate motion constraints and attitude balance requirements based on the local dangerous sections.

[0086] Risk avoidance module 203 is used to identify major danger segments based on the initial path skeleton, perform threat level assessment on the major danger segments to generate risk weights, and use the risk weights to perform path replanning to generate toxicity avoidance paths;

[0087] The motion generation module 204 is used to generate a motion primitive library based on the motion constraints and the joint range of motion, perform stability filtering on the motion primitive library according to the posture balance requirements to generate stable motion primitives, and weave the stable motion primitives into a travel scheme set by path weaving between the safe nodes.

[0088] The fitness evaluation module 205 is used to obtain the peak value of the joint load in the movement scheme, construct a dynamic envelope based on the peak value of the joint load, determine the execution fitness according to the dynamic envelope, and project the execution fitness into spatial coordinates to form an execution probability field.

[0089] The trajectory decision module 206 is used to construct a three-dimensional decision space based on the poison avoidance path, the set of travel schemes and the execution probability field, find the optimal convergence point in the three-dimensional decision space, and generate the final planned trajectory based on the optimal convergence point.

[0090] The aforementioned petrochemical explosion-proof humanoid robot trajectory planning device 200 can implement the petrochemical explosion-proof humanoid robot trajectory planning method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0091] like Figure 3As shown, the third embodiment of the present invention also provides a computer device, including a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, characterized in that the processor 302 executes the program to implement the steps of the petrochemical explosion-proof humanoid robot trajectory planning method described in the first embodiment of the present invention.

[0092] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0093] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A trajectory planning method for a petrochemical explosion-proof humanoid robot, characterized in that, include: Collect petrochemical environment gas concentration data, robot joint motion parameters, and target position information; use the gas concentration data to identify hazardous areas and generate a restricted area map; determine safe nodes from the restricted area map; and determine the joint range of motion based on the joint motion parameters. Based on the target location information and the distribution of safety nodes, a path search is performed to construct an initial path skeleton. A safety analysis is then performed on the initial path skeleton to identify local dangerous sections. Based on the local dangerous sections, motion constraints and attitude balance requirements are generated. Based on the initial path skeleton, major hazard segments are identified, and a threat level assessment is performed on the major hazard segments to generate risk weights. The risk weights are then used for path replanning to generate toxicity avoidance paths. Based on the motion constraints and the joint range of motion, a motion primitive library is generated. According to the posture balance requirements, the motion primitive library is subjected to stability filtering to generate stable motion primitives. The stable motion primitives are then woven into paths between the safe nodes to form a set of travel schemes. The peak values ​​of joint loads in the movement scheme are obtained, a dynamic envelope is constructed based on the peak values ​​of joint loads, and the execution fitness is determined according to the dynamic envelope. The execution fitness is then projected onto spatial coordinates to form an execution probability field. A three-dimensional decision space is constructed based on the poison avoidance path, the set of travel schemes, and the execution probability field. The optimal convergence point is found in the three-dimensional decision space, and the final planned trajectory is generated based on the optimal convergence point.

2. The method according to claim 1, characterized in that, The generation of motion constraints and attitude balance requirements based on the local danger segment includes: Identify areas prone to static electricity buildup from the aforementioned localized hazardous sections; A charge flow path diagram is established based on the electrostatic accumulation tendency region; The charge flow path diagram is used to set joint friction avoidance points; Based on the joint friction avoidance points, establish motion constraints and posture balance requirements.

3. The method according to claim 1, characterized in that, The process of generating risk weights by assessing the threat level of the critically hazardous sections includes: Based on the assessment of the chemical hazard intensity of the aforementioned major hazard sections, a quantitative scale is determined. The risk level calibration coefficient is set according to the aforementioned quantification scale; The risk level of the major hazard section is converted using the aforementioned hazard rating coefficient to generate a risk weight.

4. The method according to claim 1, characterized in that, The step of generating stable action primitives by performing stability filtering on the action primitive library according to the attitude balance requirements includes: Based on the posture balance requirements, the motion primitive library is sorted and ranked according to stability to form a stability gradient. Extract the stability peak points and stability valley points from the stability gradient; A motion stability system is generated by partitioning the system with the stability peak points and stability valley points as boundaries. The stability system of the motion is optimized and a stable motion primitive is generated.

5. The method according to claim 1, characterized in that, The construction of the dynamic envelope based on the joint load peak includes: Based on the peak joint load, torque distribution is identified to generate a high torque concentration band and a low torque dispersion band; The high-torque concentrated band is used to balance the load of the low-torque distributed band to form a balanced torque band. The balanced torque band is spatially enveloped and recombined to generate a recombined envelope unit; A dynamic envelope is constructed by contour locking of the recombined envelope unit.

6. The method according to claim 1, characterized in that, The step of using the risk weights to perform path replanning and generate toxicity avoidance paths includes: Based on the aforementioned risk weights, a threat avoidance zone and a security-oriented zone are constructed. The threat avoidance zone is used to generate a path repulsion factor, which includes toxicity repulsion and flammability / explosive repulsion. The path exclusion factor is fused with the safety guidance area to generate a toxicity avoidance path.

7. The method according to claim 1, characterized in that, Finding the optimal convergence point in the three-dimensional decision space includes: The search depth is determined based on the safety convergence difficulty assessed by the aforementioned three-dimensional decision space, whereby the safety convergence difficulty includes safety distribution density, feasibility space connectivity, and executability probability gradient. Priority configuration is determined based on the search depth; Utilize the aforementioned priority configuration to formulate a hierarchical search execution plan; The optimal convergence point is generated through the hierarchical search execution scheme.

8. The method according to claim 2, characterized in that, The step of setting joint friction avoidance points using the charge flow path diagram includes: Extract high-risk charge convergence nodes from the charge flow path diagram; A frictional contact exclusion zone is established based on the aforementioned high-risk charge convergence nodes; The joint movement offset point is generated using the aforementioned frictional contact restricted area; The joint motion offset point is converted into a joint friction avoidance point.

9. A trajectory planning device for a petrochemical explosion-proof humanoid robot, characterized in that, include: The environmental perception module is used to collect petrochemical environment gas concentration data, robot joint motion parameters and target position information, mark dangerous areas from the gas concentration data to generate a restricted area map, determine safe nodes from the restricted area map, and determine the joint range of motion based on the joint motion parameters. The path analysis module is used to perform path search and construct an initial path skeleton based on the target location information and the distribution of safety nodes, perform safety analysis on the initial path skeleton to identify local dangerous sections, and generate motion constraints and attitude balance requirements based on the local dangerous sections. The risk avoidance module is used to identify major hazard segments based on the initial path skeleton, perform threat level assessment on the major hazard segments to generate risk weights, and use the risk weights to perform path replanning to generate toxicity avoidance paths. The motion generation module is used to generate a motion primitive library based on the motion constraints and the joint range of motion, perform stability filtering on the motion primitive library according to the posture balance requirements to generate stable motion primitives, and weave the stable motion primitives into a travel scheme set by path weaving between the safe nodes. The fitness evaluation module is used to obtain the peak value of the joint load in the movement scheme, construct a dynamic envelope based on the peak value of the joint load, determine the execution fitness according to the dynamic envelope, and project the execution fitness into spatial coordinates to form an execution probability field. The trajectory decision module is used to construct a three-dimensional decision space based on the poison avoidance path, the set of travel schemes and the execution probability field, find the optimal convergence point in the three-dimensional decision space, and generate the final planned trajectory based on the optimal convergence point.

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

Images