Humanoid robot control method, device and equipment based on vector entanglement

Abstract

The invention provides a humanoid robot control method, device and equipment based on vector entanglement. The method comprises the following steps: performing frequency statistical analysis on a control task to extract a core action sequence; constructing a basic action vector, and generating a spiral entanglement action vector chain by adopting a partial overlapping entanglement mode; constructing an action semantic map by utilizing dynamic semantic mapping and generating a deviation prediction path; identifying a non-full-period action through the deviation feature vector, and extracting overlapped nodes from the entanglement vector chain for borrowing fusion to generate a complete execution sequence; constructing a deviation propagation network to analyze a deviation diffusion path, identifying deviation amplification nodes and convergence nodes, and extracting available components to generate action enhancement factors; and the enhancement factor and the entanglement vector chain are subjected to collaborative optimization to generate an enhancement control sequence, so that the deviation is converted from passive elimination to active utilization, and the action continuity and energy efficiency of the humanoid robot are improved.

Description

Technical Field

[0001] This invention relates to the field of humanoid robot control technology, and in particular to a humanoid robot control method, apparatus and equipment based on vector entanglement. Background Technology

[0002] Humanoid robots, as highly biomimetic intelligent systems, have shown broad application prospects in fields such as industrial manufacturing, service assistance, and medical rehabilitation. However, issues such as multi-joint coordinated control, complex motion planning, and environmental adaptability have always been key technological bottlenecks restricting their practical application. Especially when performing precision tasks, even small control deviations can be amplified through the kinematic chain, leading to task failure or system instability.

[0003] Most existing humanoid robot control methods employ traditional deviation elimination strategies, treating any deviation from a predetermined trajectory as an error requiring correction. This approach ignores the potentially beneficial effects of certain deviations, such as gravity deviation aiding descent or inertial deviation accelerating turning. Furthermore, existing methods lack effective compensation mechanisms when handling non-complete cyclic movements, making it difficult to achieve continuity and smoothness in movement sequences. Therefore, a method is urgently needed to address at least one of the aforementioned problems. Summary of the Invention

[0004] This invention provides a humanoid robot control method, device, and equipment based on vector entanglement. It aims to achieve coupling and correlation between actions by constructing entangled action vector chains, analyze the diffusion and convergence characteristics of deviations using deviation propagation networks, transform deviations that traditionally need to be eliminated into usable control resources, and complete non-integer cycle actions through borrowed fusion mechanisms, ultimately realizing intelligent and adaptive control of the humanoid robot.

[0005] The first aspect of this invention proposes a humanoid robot control method based on vector entanglement, comprising the following steps: Receive a humanoid robot control task, perform frequency statistical analysis on the control task to extract core action sequences, construct basic action vectors based on the core action sequences, and generate an entangled action vector chain using the basic action vectors through partial overlap and entanglement. An initial action semantic graph is constructed by using the entangled action vector chain for dynamic semantic mapping. The initial action semantic graph is then updated based on real-time execution feedback. A deviation prediction path is generated using the updated action semantic graph. The humanoid robot performs actions according to the deviation prediction path to generate actual execution deviation data. The actual execution deviation data and the deviation prediction path are compared and analyzed to generate a deviation feature vector. The deviation feature vector is used to identify non-integer periodic actions. Entangled action vectors that overlap with the non-integer periodic actions are extracted from the entangled action vector chain. The entangled action vectors are then fused by borrowing to generate a complete execution sequence. A deviation propagation network is constructed by combining the complete execution sequence and the deviation feature vector. The propagation network is used to analyze the diffusion path of the deviation in the action chain to obtain deviation amplification nodes and deviation convergence nodes. Deviation regulation analysis is performed on the deviation amplification nodes and deviation convergence nodes to obtain usable components. Based on the usable components, an action enhancement factor is generated. The motion enhancement factor and the entangled motion vector chain are collaboratively optimized to generate an enhanced control sequence, which is then used to drive the humanoid robot.

[0006] A second aspect of the present invention provides a humanoid robot control device based on vector entanglement, comprising: The vector construction module is used to receive humanoid robot control tasks, perform frequency statistical analysis on the control tasks to extract core action sequences, construct basic action vectors based on the core action sequences, and generate entangled action vector chains using the basic action vectors through partial overlap and entanglement. The semantic mapping module is used to construct an initial action semantic graph by performing dynamic semantic mapping using the entangled action vector chain, to evolve and update the initial action semantic graph based on real-time execution feedback, and to generate a deviation prediction path through the updated action semantic graph. The deviation analysis module is used to execute humanoid robot actions according to the deviation prediction path to generate actual execution deviation data, and compare and analyze the actual execution deviation data with the deviation prediction path to generate a deviation feature vector. The action completion module is used to identify non-integer periodic actions using the deviation feature vector, extract entangled action vectors that overlap with the non-integer periodic actions from the entangled action vector chain, and perform borrowing fusion on the entangled action vectors to generate a complete execution sequence. The deviation enhancement module is used to construct a deviation propagation network by combining the complete execution sequence and the deviation feature vector, analyze the diffusion path of the deviation in the action chain through the deviation propagation network to obtain deviation amplification nodes and deviation convergence nodes, perform deviation regulation analysis on the deviation amplification nodes and deviation convergence nodes to obtain usable components, and generate action enhancement factors based on the usable components. The control optimization module is used to collaboratively optimize the motion enhancement factor and the entangled motion vector chain to generate an enhanced control sequence, and to use the enhanced control sequence to drive the humanoid robot.

[0007] 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 vector entanglement-based humanoid robot control method disclosed in the first aspect.

[0008] The beneficial effects of this invention are reflected in the following points: First, by constructing a helical entangled motion vector chain, this invention achieves a structured expression and dynamic coupling of motion sequences. The entanglement coefficient quantifies the correlation strength between adjacent actions, and the helical topology preserves temporal relationships and spatial proximity, enabling the humanoid robot to predict and adjust subsequent actions based on the current action state, thus improving the smoothness and accuracy of multi-joint coordinated control, especially maintaining motion continuity during complex action transitions. Second, this invention implements an active deviation utilization mechanism. By identifying deviation amplification nodes and convergence nodes through a deviation propagation network, it transforms factors that traditionally need to be eliminated, such as gravity deviation and inertial deviation, into beneficial resources to assist in action execution. This not only reduces energy consumption but also makes the humanoid robot's movements more natural and fluid, fully utilizing environmental forces and inertial forces in specific actions such as descent and turning. Finally, the borrowing fusion mechanism effectively solves the problem of action interruption and non-integer cycle execution. By dynamically extracting the surplus force, velocity and direction components from adjacent entangled nodes, it realizes intelligent completion of interrupted actions, ensuring the continuous execution of tasks. This enables humanoid robots to maintain the integrity of their actions when faced with sudden interference or abnormal situations, and enhances their task adaptability in complex environments.

[0009] 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

[0010] 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.

[0011] Unless otherwise specified or defined, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0012] Figure 1 This is a flowchart illustrating a humanoid robot control method based on vector entanglement according to the present invention.

[0013] Figure 2 This is a structural block diagram of a humanoid robot control device based on vector entanglement according to the present invention.

[0014] Figure 3 This is a schematic diagram of the structure of a computer device according to the present invention. Detailed Implementation

[0015] 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.

[0016] 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.

[0017] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0018] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0019] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] 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.

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

[0022] like Figure 1 As shown, this embodiment of the invention provides a humanoid robot control method based on vector entanglement, including the following steps S110-S160: Step S110: Accept the humanoid robot control task, perform frequency statistical analysis on the control task to extract the core action sequence, construct basic action vectors based on the core action sequence, and generate an entangled action vector chain using the basic action vectors through partial overlap and entanglement.

[0023] The system receives task instructions from the host control system via a humanoid robot control interface, constructing a multi-layered task parsing framework. Control tasks are defined using a structured description language, including key elements such as task type, target object, execution requirements, and constraints. Task types cover typical application scenarios such as grasping operations, assembly operations, material handling, and precision positioning. The task parsing process first performs semantic understanding, converting natural language or symbolic task descriptions into machine-executable instruction sequences. For example, upon receiving the task "Move workpiece A from position P1 to position P2 and place it precisely," the system identifies the material handling task type and extracts parameters such as the starting position P1 (x1, y1, z1), the target position P2 (x2, y2, z2), and workpiece attributes (mass m = 2.5 kg, dimensions 200 × 150 × 80 mm). The task verification mechanism checks the completeness and feasibility of the instructions, including workspace accessibility verification, load capacity verification, and path collision-free detection. Task priorities are set based on urgency and importance, using a four-level classification: Urgent and Important (Priority 1), Urgent and Moderate (Priority 2), Routine and Important (Priority 3), and Routine and Moderate (Priority 4). Time constraints extract the start time, end time, maximum allowed execution time, and other time-series requirements for each task. An abnormal task identification mechanism filters out unreasonable or dangerous instructions, such as load requirements exceeding the humanoid robot's capabilities or action sequences that violate safety regulations.

[0024] Frequency statistical analysis was performed on the control task to extract the core action sequence. Based on the received control task, a hierarchical decomposition strategy was adopted to refine the complex task into action sets of different granularities. The multi-granularity decomposition followed the progressive principle of "whole-part-detail". Coarse-grained decomposition divided the task into main action stages, each stage representing a relatively independent functional unit. For example, the handling task was decomposed into five stages: approach, grasp, transfer, place, and return. Medium-granularity decomposition further refined each coarse-grained action into specific motion sequences, and fine-grained decomposition reached the joint motion level. The granularity of decomposition was determined using the information entropy criterion H=-Σp(i)log(p(i)). Decomposition was stopped when the rate of change of H was less than 5%. For the three granularity action sets generated, statistical analysis methods were used to identify high-frequency repetitive action patterns. The coarse-grained frequency calculation formula is f_coarse(k)=n_k / N_total, where n_k is the number of occurrences of action k and N_total is the total number of tasks. It was found that the "approach-grab-transfer-place" sequence appeared in 85% of the handling tasks. Medium-granularity analysis identifies motion transition patterns by calculating conditional probabilities P(a_j|a_i). When P(a_j|a_i) > 0.7, it indicates a strong correlation between motions. Fine-granularity statistics reveal common characteristics of low-level control; cluster analysis of joint angle sequences found that 70% of grasping actions use similar trajectory patterns. Cross-granularity correlation analysis establishes a mapping relationship between three levels of motions, and multi-level core motion sequences are extracted by comprehensively considering the statistical results of each granularity.

[0025] Basic motion vectors are constructed based on the core motion sequence. Based on the extracted core motion sequence, multi-dimensional basic motion vectors are constructed to fully characterize the humanoid robot's motion characteristics. The basic motion vector is defined as a four-dimensional composite structure V_base=(F,V,D,M), where F is the force vector component, V is the velocity vector component, D is the direction vector component, and M is the memory component. The force vector component F=[f_x,f_y,f_z,τ_x,τ_y,τ_z] contains triaxial forces and triaxial torques, fully describing the mechanical characteristics of the motion; for example, the "precise grasping" motion is characterized by a vertically downward clamping force and a moderate wrist torque. The velocity vector component V=[v_linear,a_linear,v_angular,a_angular] covers the linear velocity, linear acceleration, angular velocity, and angular acceleration of the end effector, reflecting the dynamic characteristics of the motion. The direction vector component D is represented by quaternions D=[q_w,q_x,q_y,q_z] to avoid the gimbaling problem of Euler angles and ensure the continuity and uniqueness of the attitude representation. The memory component M = [t_duration, success_rate, energy_cost, stability_index] records the average execution time, historical success rate, energy consumption metrics, and stability score, enabling the system to learn and optimize from historical executions. Vector normalization ensures comparability across different dimensions, and timestamps associate each vector with its position in the core sequence. Through multi-dimensional feature extraction and vectorization, a basic action vector that comprehensively reflects the essential characteristics of the action was successfully constructed.

[0026] In some embodiments, generating an entangled action vector chain using the basic action vectors through partial overlap and entanglement includes: calculating the vector angle and magnitude difference between temporally adjacent vectors in the basic action vectors to generate a similarity matrix; determining the vector overlap region based on the similarity matrix; performing nonlinear weighted fusion on the vector components within the overlap region to generate entanglement coefficients; and constructing the entangled action vector chain using the entanglement coefficients through a spiral connection method.

[0027] Using the constructed basic action vectors, a similarity measure between temporally adjacent vectors is calculated to generate a similarity matrix. Temporally adjacent is defined as two consecutive actions in the core action sequence, and their correlation needs to be evaluated. The vector angle is calculated using cosine similarity: cos(θ)=(V_i·V_j) / (||V_i||×||V_j||), where V_i and V_j are adjacent basic action vectors, θ is the angle between the two vectors, · represents the vector dot product, and ||·|| represents the vector magnitude. Since the basic action vector contains multiple components, the angle between each component is calculated separately and then weighted and combined. The magnitude difference reflects the change in action intensity: Δ_mag=|||V_i||-||V_j||| / max(||V_i||,||V_j||), where Δ_mag is the normalized magnitude difference. For example, in the transition from "rapid approach" to "precise positioning," the velocity component decreases significantly, indicating a change in action pattern. The similarity index S_ij integrates two aspects: directional consistency and intensity matching. The similarity matrix S is a symmetric matrix, where each element S_ij represents the similarity between action i and action j, with values ​​ranging from [0,1]. The structural characteristics of the matrix reflect the inherent correlation patterns of the action sequences. The calculation of the similarity matrix considers not only the vector relationships at a single moment but also captures the dynamic trends over a short period using a sliding window approach. The sparsity of the matrix reflects the modularity of the action sequences; high sparsity indicates that the actions are relatively independent.

[0028] Based on the generated similarity matrix, overlapping regions between adjacent vectors are determined. The determination of overlapping regions is based on a similarity threshold and the duration of the action; overlap is considered to exist when the similarity exceeds a preset threshold. The temporal range of overlap is dynamically determined according to the action duration and similarity magnitude; the higher the similarity, the greater the overlap ratio. Spatial overlap is determined through the intersection of workspaces, calculating the intersection of the reachable ranges of two actions. The degree of overlap among vector components varies; the overlap characteristics of force, velocity, and direction components are different and require separate processing. For example, two consecutive actions may have highly consistent force directions but significantly different velocity curves. Features of overlapping regions include key parameters such as the overlap center and overlap gradient. Boundary condition processing ensures a smooth transition in overlapping regions, avoiding abrupt changes. Abnormal overlap detection identifies unreasonable overlap patterns, such as contradictory situations where directions are completely opposite but similarity remains high. Special attention is paid to critical moments of action transition during overlapping region identification, as these moments often contain rich coordination information. The data density within overlapping regions is usually higher than in non-overlapping regions, requiring more refined sampling and processing.

[0029] Within a defined overlapping region, the vector components are nonlinearly weighted and fused to generate entanglement coefficients characterizing the strength of the action association. The entanglement coefficients are calculated using a comprehensive evaluation model: C_ij = α × S_ij × O_ij × exp(-β × Δt_ij), where C_ij is the entanglement coefficient between vectors i and j, ranging from [0,1]; S_ij is the similarity index calculated above, obtained based on the difference in vector angle and magnitude; O_ij is the coverage degree of the overlapping region, reflecting the spatiotemporal overlap of the two actions, ranging from [0,1]; Δt_ij is the time interval |t_i - t_j|, representing the time difference between adjacent actions; α is the weighting coefficient, typically 0.9; and β is the time decay parameter, 0.1, controlling the influence of time distance on the entanglement strength. Nonlinear fusion employs exponential decay characteristics to avoid the limitations of simple linear combinations. Force component fusion considers the continuity constraint of force to ensure no abrupt changes in force during the transition. Velocity component fusion uses spline interpolation to ensure the smoothness of the motion. Orientation component fusion uses quaternion spherical interpolation to maintain the continuity of posture changes. Memory component fusion integrates historical execution information through probabilistic updates. Coefficient normalization ensures the comparability of different action pairs. The key to nonlinear fusion is maintaining the physical feasibility of the actions; the fusion result must satisfy the kinematic and dynamic constraints of the humanoid robot.

[0030] For example, the step of constructing an entangled action vector chain using the entanglement coefficients in a spiral connection manner includes: determining the radial distribution of the spiral based on the force vector components in the basic action vector; determining the tangential velocity of the spiral using the velocity vector components in the basic action vector; generating a spiral trajectory equation by combining the radial distribution and the tangential velocity; calculating the spatial position of each vector based on the spiral trajectory equation and the entanglement coefficients; and forming a spiral entangled action vector chain by weighted connection of the spatial positions using the entanglement coefficients.

[0031] The radial distribution of the helical structure is determined based on the force vector components in the basic motion vector. The magnitude of the force vector components directly affects the helical radius, establishing a mapping relationship r(t) = r_0 + k_F × ||F(t)||, where r(t) is the helical radius at time t, r_0 is the basic radius, k_F is the force-radius mapping coefficient, and ||F(t)|| is the force vector magnitude. The radial distribution reflects the change in mechanical intensity of the motion; high-force motions correspond to larger helical radii, while delicate operations correspond to smaller radii. For example, the helical radius of a heavy-duty handling motion is significantly larger than that of a precision assembly motion. The radial rate of change dr / dt reflects the dynamic adjustment process of the force; rapid changes indicate the switching of motion modes. The continuity of the radial distribution is ensured by a smoothing function to avoid structural instability caused by abrupt radius changes. Limitations on abnormal radial values ​​prevent extreme cases by setting maximum and minimum radius boundaries. Statistical analysis shows that the radial distribution of typical operational tasks exhibits a bimodal characteristic, corresponding to the two main modes of stress control and position control. The physical significance of the radial distribution lies in its direct reflection of the working intensity of the humanoid robot; operators can understand the current load state by observing the expansion and contraction of the helical component. Radial mutation points often correspond to critical transition moments in a task and require special attention.

[0032] The tangential motion characteristics of the helical spiral are determined using the velocity vector components in the basic motion vector. These velocity vector components control the spiral's winding speed, establishing the relationship v_θ(t) = ω_0 + k_V × ||V(t)||, where v_θ(t) is the tangential angular velocity, ω_0 is the basic angular velocity, k_V is the velocity mapping coefficient, and ||V(t)|| is the velocity vector magnitude. The tangential velocity determines the spiral's density; high-speed motion produces a sparse spiral, while low-speed motion produces a dense spiral. For example, the spiral spacing is larger during the rapid movement phase and tighter during the precise positioning phase. Changes in angular velocity reflect the urgency and precision requirements of the motion. The acceleration component affects the rate of change of the tangential velocity and is used to predict velocity trends. Upper and lower limits for the tangential velocity are set to prevent over-winding or loosening. The direction information of the velocity vector is used to determine the spiral's direction; positive velocity produces right-handed spiraling, and negative velocity produces left-handed spiraling. A dynamic adjustment mechanism corrects the tangential velocity based on real-time feedback. The variation pattern of tangential velocity is closely related to the task rhythm. Uniform tangential velocity indicates a stable execution process, while frequent changes may indicate task difficulty or environmental interference. The density variation of the spiral provides natural segmentation markers for the action sequence. The tangential velocity of the spiral is determined by converting the velocity vector into tangential motion.

[0033] By combining the determined radial distribution and tangential velocity, a complete helical trajectory equation is generated. The helical trajectory is parametrically represented as: x(t) = r(t)cos(θ(t)), y(t) = r(t)sin(θ(t)), z(t) = h(t), where x, y, and z are Cartesian coordinates, θ(t) is the cumulative phase angle, and h(t) is the axial height. The phase angle is obtained by integrating the tangential velocity: θ(t) = ∫v_θ(τ)dτ. The axial height h(t) is linearly related to the execution progress and is modulated by the memory component. The continuity of the trajectory is guaranteed by the differentiability of the parametric equations. Curvature and torsion calculations are used to evaluate the geometric properties of the trajectory. For example, a complete grasp-transfer-place sequence generates a helical trajectory that first expands and then contracts. Trajectory optimization reduces unnecessary bending through the principle of energy minimization. Boundary conditions set the start and end points of the trajectory to ensure consistency with task requirements. The discretization of the parametric equations supports numerical computation and visualization. The generation process of the spiral trajectory fully considers the motion constraints of the humanoid robot, ensuring that the generated trajectory is reachable within the joint space. The geometric characteristics of the trajectory are related to the task complexity; simple tasks correspond to regular spirals, while complex tasks are represented by irregular spatial curves.

[0034] Based on the generated spiral trajectory equation and entanglement coefficient, the spatial position of each basic action vector in three-dimensional space is calculated. The calculation of the entanglement coefficient comprehensively considers similarity, overlap, and time decay characteristics: C_ij = α × S_ij × O_ij × exp(-β × Δt_ij), where S_ij is the similarity index calculated above, O_ij is the coverage degree of the overlapping area, Δt_ij is the time interval, and α = 0.9 and β = 0.1 are empirical parameters. Spatial positioning is achieved by substituting discrete time points into the trajectory equation, and the spatial coordinates of the i-th vector at time t_i are P_i = (x_i, y_i, z_i). The entanglement coefficient affects the relative position between vectors; a high entanglement coefficient leads to close spatial positions, forming tight entanglement. The position adjustment algorithm ensures that the spacing between adjacent vectors conforms to the entanglement strength: d_ij = d_base × (1 - C_ij), where d_ij is the spatial distance between vectors i and j, and d_base is the baseline distance. For example, when C_ij = 0.8, the actual spacing d_ij = 0.2 × d_base, indicating that strongly entangled action pairs are nearly overlapping on the spiral; when C_ij = 0.2, d_ij = 0.8 × d_base, and weakly entangled action pairs maintain a clear interval. The uniformity of spatial distribution is achieved through iterative optimization, avoiding excessive local concentration. The spatial distribution pattern reflects the inherent structure of the action sequence; dense regions correspond to highly coordinated action groups, while sparse regions represent relatively independent actions. The transformation relationship between the humanoid robot's base coordinate system and the world coordinate system is considered during position calculation.

[0035] By establishing weighted connections between vectors based on spatial location and entanglement coefficients, a complete spiral entangled action vector chain is formed. Connection construction is based on neighborhood relationships and entanglement strength; vectors at adjacent positions are connected by a weighted entanglement coefficient C_ij, with the connection weight directly equal to the entanglement coefficient value. The connection topology adopts a directed graph structure, where nodes are basic action vectors in spatial location, edges are weighted connections, and the direction of the edges represents the temporal sequence of action execution. Strongly entangled connections (C_ij > 0.7) form the backbone links, reflecting tight coupling between actions; moderately entangled connections (0.3 ≤ C_ij ≤ 0.7) form auxiliary links, providing coordination support between actions; weakly entangled connections (C_ij < 0.3) serve as supplementary links, maintaining the integrity of the action sequence. The spiral connection is characterized by maintaining temporal continuity and spatial proximity; actions at adjacent moments are physically close on the spiral, facilitating information transmission and state sharing. The time-varying nature of the connection weights allows for dynamic adjustment of entanglement strength based on the execution state, achieving adaptive link management. For example, in precision assembly tasks, the high entanglement coefficient of 0.85 between the "alignment" and "insertion" actions forms a strong connection, ensuring close coordination between the two actions. Through weighted connections based on the entanglement coefficient and the organization of helical topology, a complete helical entangled action vector chain is finally constructed.

[0036] Step S120: An initial action semantic graph is constructed by using entangled action vector chains for dynamic semantic mapping. The initial action semantic graph is then evolved and updated based on real-time execution feedback. A deviation prediction path is generated using the updated action semantic graph.

[0037] Specifically, an initial action semantic graph is constructed using dynamic semantic mapping with entangled action vector chains. The spiral connections and entanglement coefficients in the entangled action vector chains reflect the temporal relationships and coupling strength of the actions, and this numerical information needs to be converted into an understandable and analyzable semantic network. Semantic mapping first identifies key nodes in the vector chains, i.e., positions with high entanglement coefficients, which typically correspond to transition points in action modes. The semantic labeling of nodes is determined based on the dominant component of the basic action vector: when the force component is dominant, it is labeled "force-controlled"; when the velocity component is dominant, it is labeled "motion-controlled"; and when the direction component is dominant, it is labeled "positioning-controlled". Edge construction directly utilizes the entanglement coefficients; when the entanglement coefficients of two nodes exceed a set threshold, a connection edge is established, and the edge weight is equal to the entanglement coefficient value. The graph adopts a directed graph structure G=(V,E), where V is the set of action nodes, E is the set of transition edges, and the direction represents the execution order of the actions. For example, in a precision assembly task, the high entanglement coefficients of the "alignment" node and the "insertion" node form a strongly correlated edge, reflecting the close cooperation between these two actions. The initial graph also contains hierarchical information about actions, with coarse-grained actions as parent nodes and fine-grained actions as child nodes, forming a multi-level semantic structure.

[0038] Based on feedback data from the actual execution process of the humanoid robot, the initial action semantic graph is dynamically updated to reflect the real execution characteristics. Real-time execution feedback includes multi-dimensional data such as action completion time, execution accuracy, force control error, and energy consumption. This data is used to adjust node attributes and edge weights in the graph. Node attribute updates employ an exponentially weighted moving average method, where new execution data is weighted and fused with historical data to ensure the graph can adapt to changes in execution characteristics. Edge weight adjustments reflect the actual effect of action transfer; when the execution effect of an action sequence is better than expected, the weight of the corresponding edge increases, and vice versa. A new node discovery mechanism identifies new action patterns during execution. When an unrecorded action combination recurs with stable results, it is added to the graph as a new node. Graph topology optimization includes removing long-unused nodes and low-weight edges to maintain a streamlined and efficient structure. An adaptive strategy is adopted for the evolution rate, with rapid learning in the early stages of system operation and a gradual reduction in update frequency as experience accumulates. Version control records the evolution history of the graph, supporting performance comparison and anomaly tracing. The update process maintains the connectivity and integrity of the graph, avoiding isolated nodes or broken paths. The evolution and updating of the action semantic graph were achieved through incremental updates and structural optimization driven by real-time feedback.

[0039] In some embodiments, generating a deviation prediction path using an updated action semantic graph includes: extracting historical deviation nodes from the updated action semantic graph to construct a deviation evolution subgraph; using the deviation evolution subgraph to calculate the deviation transition probability between nodes to generate a deviation state matrix; performing multi-step prediction based on the deviation state matrix to generate a temporal deviation sequence; and mapping the temporal deviation sequence to an execution time axis to form a deviation prediction path.

[0040] Nodes exhibiting significant deviations are identified from the updated action semantic graph, and a subgraph specifically designed for deviation analysis is constructed. Deviation node identification is based on error statistics in execution feedback; a node is marked as a deviation node when the positional error, force error, or time deviation of an action exceeds the permissible range. Deviations are categorized into systematic and random deviations. Systematic deviations manifest as continuous unidirectional errors, while random deviations exhibit fluctuating errors. The deviation subgraph includes not only the deviation node itself but also its predecessor and successor nodes, used to analyze the causes and propagation effects of deviations. The subgraph construction preserves the connection relationships and weight information between these nodes in the original graph while adding deviation-related attribute annotations. Deviation intensity is normalized to unify the units, facilitating comparative analysis of different types of deviations. Time stamps record typical times and durations of deviation occurrences, aiding in predicting deviation patterns.

[0041] Using historical data from the deviation evolution subgraph, the propagation pattern of deviations between nodes is statistically analyzed and quantified into transition probabilities. The deviation transition probability P_ij represents the probability that node j will also deviate given that node i has deviated, calculated using the conditional probability formula P_ij = N_ij / N_i, where N_ij is the number of times node j deviates after node i deviates, and N_i is the total number of times node i deviates. Deviation states are divided into multiple levels, from no deviation to severe deviation, with each node in one of these states at any given time. State transitions consider not only the direct influence between adjacent nodes but also the cumulative effect and attenuation characteristics of deviations. The matrix construction process ensures row and normalization, satisfying the mathematical properties of probability matrices. Sparsity handling preserves the main transition paths while ignoring transition relationships with extremely low probabilities. Time-varying characteristics are reflected through time-segmented statistics, as the transition probabilities may differ across different task stages. Matrix eigenvalue analysis verifies the system's stability; a maximum eigenvalue less than 1 indicates that the deviation will not amplify indefinitely.

[0042] Multi-step Markov prediction based on the deviation state matrix generates a temporal deviation sequence for future time periods. Prediction starts from the current deviation state, calculating the probability distribution of future states at each time step through exponentiation of the state transition matrix. The prediction step size is determined based on task characteristics, typically choosing a number of steps that cover a complete action sequence. The initial state vector is set based on the latest deviation detection results and can be a deterministic single state or a probability distribution. As the number of prediction steps increases, uncertainty gradually accumulates, the probability distribution tends to diffuse, and the confidence of long-term predictions decreases accordingly. The most probable path is determined using a dynamic programming algorithm, providing the state evolution trajectory with the highest probability. Branch prediction considers multiple possible evolution scenarios, providing comprehensive information for risk assessment. The prediction results include the probability values ​​of each state at each time step, forming a complete temporal deviation sequence. Identifying key inflection points helps determine the timing of intervention; adjustments need to be made in advance when predictions indicate a sharp increase in deviation.

[0043] Discrete temporal deviation sequences are mapped to a continuous execution time axis, forming a deviation prediction path that can be used for real-time control. The time mapping requires incorporating the typical execution duration of each action in the semantic graph, which is statistically analyzed during graph updates. The execution time of each action fluctuates, and the statistical average is used as the nominal duration. The distribution of deviation intensity on the time axis is made continuous using interpolation methods, with cubic spline interpolation commonly used to ensure curve smoothness. Key time points are labeled, including the deviation start time, peak time, and recovery time; this information is crucial for control decisions. The confidence interval of the path reflects the uncertainty of the prediction, represented by upper and lower bound curves. Sliding updates of the time window support online prediction, with the prediction path refreshed each control cycle based on the latest data. Abnormal events are specially marked in the path to alert the system to potential risks. Compressed storage of the path data retains only key inflection points and feature points, reducing storage and computational overhead.

[0044] Step S130: Perform humanoid robot actions according to the deviation prediction path to generate actual execution deviation data, and compare and analyze the actual execution deviation data with the deviation prediction path to generate a deviation feature vector.

[0045] Specifically, the humanoid robot's actions are executed according to the deviation prediction path to generate actual execution deviation data. The deviation prediction path provides the expected deviation value and key time nodes at each time point, including the deviation start time, peak time, and recovery time. These key moments require close monitoring during execution. The humanoid robot executes the action sequence defined in the entangled action vector chain, while simultaneously adjusting control parameters with reference to the prediction path. When the prediction path indicates a position deviation peak at a certain time, the control system adjusts the position loop gain in advance to prepare for compensation. Data acquisition during execution employs high-frequency sampling, with the sampling frequency of key parameters such as position, force, and velocity dynamically adjusted according to the action type. For example, during the precision assembly stage, the position sampling frequency is increased to 1kHz to capture minute deviations; during the rapid movement stage, the focus is on monitoring changes in velocity and acceleration. Real-time calculation of deviation data uses a reference trajectory comparison method: actual deviation Δ(t) = X_actual(t) - X_reference(t), where X represents the state vector, containing multi-dimensional information such as position, attitude, and force. A data caching mechanism ensures complete time-series records, including normal execution data and abnormal mutation data. The guiding role of predictive paths is reflected in the proactive adjustment of control strategies, rather than in passive responses.

[0046] The actual deviation data and the predicted deviation path are compared and analyzed to generate a deviation feature vector. The comparative analysis unfolds along two dimensions: time and magnitude. First, time alignment is performed to ensure that the predicted and actual data are compared on the same time reference. Key time points marked on the predicted path are matched with the actual deviation evolution process, and the time error Δt = t_actual - t_predicted is calculated. The comparison of deviation magnitude uses a combination of point-to-point difference calculation and statistical analysis, focusing not only on instantaneous errors but also analyzing the distribution characteristics of the errors. Prediction accuracy evaluation indicators include mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient. The comparison of deviation evolution patterns identifies differences between predicted and actual trends, such as a predicted gradual increase but an actual step-like abrupt change. Feature extraction is performed from multiple perspectives: dynamic features include deviation growth rate, peak characteristics, and duration; statistical features include mean, variance, skewness, and kurtosis; frequency domain features obtain the dominant frequency component through FFT analysis; and correlation features reflect the coupling relationship between different types of deviations. Anomaly deviation identification identifies sudden situations not covered by the predicted path; this information is of significant value for model improvement. The construction of the bias feature vector V_dev=[f_1,f_2,...,f_n] integrates all extracted features, where f_i represents the i-th feature component. Feature normalization ensures the comparability of features with different dimensions.

[0047] Step S140: Use the deviation feature vector to identify non-integer periodic actions, extract entangled action vectors that overlap with non-integer periodic actions from the entangled action vector chain, and perform borrowing fusion on the entangled action vectors to generate a complete execution sequence.

[0048] Specifically, non-integer cycle actions are identified using deviation feature vectors. These vectors contain multi-dimensional information, including dynamic, statistical, and frequency domain characteristics, which collectively reveal the integrity of the action's execution. The determination of non-integer cycle actions is based on multiple indicators: when the duration characteristic in the deviation feature vector is significantly shorter than the expected value, the deviation growth rate shows an abnormal truncation, and the frequency domain characteristics show a missing fundamental frequency component, it indicates that the action failed to complete a complete cycle. For example, in a precision insertion task, the standard execution time for the "insertion" action is 2.5 seconds, but the deviation feature vector shows that it actually only executed for 1.8 seconds before terminating. Phase analysis shows that the termination phase φ_end = 1.45π < 2π, and the force characteristic shows a sharp increase, indicating that abnormal resistance caused the action to be interrupted. Interruption cause analysis distinguishes between active protection interruptions and passive interference interruptions by identifying abrupt change patterns in the deviation features. Timestamps record the precise time of the interruption and the percentage of execution completed.

[0049] Entangled action vectors that overlap with non-integer periodic actions are extracted from the entangled action vector chain. The spiral structure of the entangled action vector chain preserves the spatial relationships and entanglement strength between nodes, providing a topological basis for finding overlapping nodes. The determination of overlap considers multiple dimensions: the entanglement coefficient reflects the coupling strength between nodes, with a high entanglement coefficient indicating a close functional association; spatial distance is measured by the arc length on the spiral trajectory, with nearby nodes more likely to share resources; functional similarity is evaluated by comparing the components of the basic action vectors, with similar vector compositions suitable for complementarity. After determining the position of the non-integer periodic action in the vector chain, the search algorithm searches for candidate overlapping nodes in its neighborhood. The search scope includes not only directly adjacent predecessor and successor nodes but also spatially adjacent nodes in the spiral structure. Temporal compatibility ensures that overlapping nodes do not conflict in time.

[0050] In some embodiments, the step of borrowing and fusing the entangled action vectors to generate a complete execution sequence includes: locating adjacent entangled nodes in the entangled action vector chain according to the non-integer periodic action; extracting borrowable vector components from the adjacent entangled nodes to generate a compensation vector set; aligning the compensation vector set and the non-integer periodic action in time to generate a fusion weight matrix; and performing vector fusion operation through the fusion weight matrix to generate a complete execution sequence.

[0051] The algorithm locates adjacent entangled nodes in an entangled action vector chain based on non-integer periodic actions. The entangled vector chain's data structure records the index, spatial coordinates, and connectivity of each node, allowing for rapid location of non-integer periodic action nodes through index lookup. The definition of adjacent nodes is not limited to sequential relationships but also includes geometric proximity in spiral space. The neighborhood search algorithm expands the search range centered on the non-integer periodic node, following an increasing distance order. First-order neighbors include direct predecessors and successors, directly accessed via a linked list structure; second-order neighbors are indirectly obtained through the connectivity of first-order neighbors; spatial neighbors are identified by calculating Euclidean distance along the spiral trajectory. Entanglement strength serves as the weight for adjacency relationships, with strongly entangled nodes prioritized for inclusion in the neighbor set. Node state checks ensure the availability of adjacent nodes. The size of the neighbor set is dynamically determined based on the degree of absence of non-integer periodic actions.

[0052] For example, the step of extracting borrowable vector components from the adjacent entangled nodes to generate a compensation vector set includes: analyzing the vector redundancy of the adjacent entangled nodes and identifying borrowable components with redundancy exceeding a threshold; determining borrowing priorities for the missing types of the non-integer cycle actions, with force components taking precedence over velocity components and velocity components taking precedence over direction components; constructing borrowing weights by combining the borrowing priorities and the entangled action vector chain, with greater borrowing weights for closer nodes; and dynamically extracting vector components from the borrowable components to form a compensation vector set based on the borrowing weights.

[0053] The system analyzes and locates adjacent entangled nodes, evaluating the usage status and redundancy of each component in their basic action vector V_base=(F,V,D,M). Redundancy is calculated based on the ratio of the current value of a component to its capacity limit; a component is considered to have available redundancy when its usage rate is below a set threshold. For example, if an adjacent "precise positioning" node has just completed execution, its force component F still retains 30% of its residual value available for borrowing, and its velocity component V retains 50% kinetic energy redundancy due to inertia; while the "force control adjustment" node to be executed has all its force components in a reserve state, with a redundancy of 90%. The dynamic characteristics of redundancy consider the decay law of components; force components decay faster, while directional components are relatively stable. Time-varying redundancy is updated in real time to ensure that borrowing decisions are based on the latest state. A safety margin is reserved to prevent excessive borrowing from affecting the node's own functionality. After a comprehensive evaluation of the resource status of adjacent nodes, borrowable components with redundancy exceeding the threshold are identified.

[0054] Based on the specific missing details of non-integer cycle actions, a borrowing priority strategy for vector components is formulated. Missing type analysis is based on the state at the time of interruption; if the interruption is due to insufficient force control, the force component has the highest demand; if the interruption is due to trajectory deviation, the direction component is more important. The basic principle of borrowing priority is that force components take precedence over velocity components, and velocity components take precedence over direction components, reflecting the decreasing importance of execution stability. Dynamic priority adjustment is based on task characteristics; precision assembly tasks increase the weight of force components, and rapid movement tasks increase the weight of velocity components. Memory components, as auxiliary information, are generally not involved in borrowing to maintain the independence of each action. Priority quantification uses normalized weight representation to ensure the comparability of different components. A feedback adjustment mechanism corrects priorities based on borrowing effects. Based on task-oriented analysis and dynamic adjustment, the borrowing priority is determined.

[0055] Based on borrowing priority and the topological relationship between nodes, quantified borrowing weights are obtained. The aforementioned borrowing priority "force component takes precedence over velocity component, velocity component takes precedence over direction component" needs to be converted into a numerical priority weight P_priority. This weight is assigned using an arithmetic progression, with the force component receiving the highest weight, followed by the velocity component, and the direction component receiving the lowest, ensuring that the weight difference reflects the priority relationship. Key parameters required for constructing the weights are extracted from the entangled action vector chain: for adjacent nodes, the entanglement coefficient C_ij is a value already determined during the construction using a spiral connection; for non-adjacent nodes, the Euclidean distance or spiral arc length distance d_ij between nodes is recalculated based on the spatial coordinates P_i and P_j recorded in the vector chain. The borrowing weight comprehensively considers three factors: priority weight reflects the importance of the component, distance decay ensures priority borrowing from neighboring nodes, and entanglement strength reflects the functional correlation between nodes. Distance decay uses an exponential form exp(-d_ij / σ), causing the contribution of distant nodes to decrease rapidly; entanglement strength amplifies the role of highly coupled nodes through a power function C_ij^γ. The comprehensive weight calculation formula is W_ij=P_priority×exp(-d_ij / σ)×C_ij^γ, which is then normalized within a standard interval to form a comparable quantitative index. The weight matrix is ​​organized according to a two-dimensional node-component structure, with rows corresponding to adjacent nodes and columns corresponding to vector component types.

[0056] Based on borrowing weights, the required vector components are extracted from the available components to form a compensation set. The extraction process employs a greedy algorithm, prioritizing extraction from the node-component combination with the highest weight. The extraction quantity is determined based on the degree of missing non-integer cycle actions and the node's surplus capacity, following the minimum sufficiency principle. Force component extraction considers directional consistency, velocity component extraction maintains motion continuity, and directional component extraction avoids abrupt changes. The extraction order follows a weighted queue, updating the remaining demand and node surplus status after each extraction. Conflict detection is implemented during extraction; when multiple non-integer cycle actions simultaneously request the same node resource, an arbitration mechanism coordinates the allocation. A multi-source extraction strategy allows obtaining different types of components from different nodes, achieving complementary advantages. Constraints in the extraction process ensure that the normal function of the providing nodes is not affected. The compensation vector set is organized in a structured form, recording the source, quantity, and timeliness of each component. After systematic extraction based on weights, the extracted vector components are arranged in time sequence to generate a compensation vector V_compensate(t), where t represents the time point of the compensation action. This vector contains information on force, velocity, and direction components borrowed from multiple adjacent nodes.

[0057] The extracted compensation vector set is precisely aligned with the non-integer cycle actions in the time dimension to construct a fusion weight matrix. The core of time alignment is to determine the effect of the compensation components on the time axis. The completed parts of the non-integer cycle actions remain unchanged, and only the missing parts after the interruption are compensated. The alignment process divides the compensation period into three stages: transition period, fusion period, and stabilization period. The design of the fusion weight matrix W(t) considers time-varying characteristics, and the weight function adopts an S-shaped curve w(t) = 1 / (1 + exp(-k(t-t_c))), where t_c is the switching center time, and k controls the transition speed. Synchronization of the time base is achieved through a unified clock stamp, ensuring that the original actions and compensation components are aligned under the same time reference system. The parameters of the weight function are adaptively adjusted according to the action type; a slow transition is used for precise actions, while a steeper transition is allowed for fast actions. The rows of the matrix correspond to different vector components, and the columns correspond to time sampling points. Boundary conditions ensure that the weights are continuous at the time of interruption.

[0058] A weighted fusion operation is performed on the original non-integer periodic vector and the compensation vector set using a fusion weight matrix. The fusion operation is in matrix form: V_complete(t) = W(t) × [V_original(t); V_compensate(t)], where V_original(t) is the completed part of the original non-integer periodic vector, V_compensate(t) is the compensation vector extracted from adjacent nodes, [V_original(t); V_compensate(t)] represents the vertical concatenation of the two vectors to form an extended input vector, and W(t) is the corresponding time-varying fusion weight matrix. The fusion process maintains the continuity of the action; the part before the interruption point completely retains the original vector, and the part after the interruption point achieves a smooth transition through weighted combination. Independent fusion at the component level ensures that the optimization of each dimension does not interfere with each other. The fusion calculation adopts an incremental update method, processing only the fusion at the current time step, reducing computational complexity. The fused vector sequence covers the complete action cycle, restoring the interrupted function. Based on the system's fusion operation and optimization processing, a complete execution sequence is finally generated.

[0059] Step S150: Construct a deviation propagation network by combining the complete execution sequence and deviation feature vector. Analyze the diffusion path of deviation in the action chain through the deviation propagation network to obtain deviation amplification nodes and deviation convergence nodes. Perform deviation regulation analysis on deviation amplification nodes and deviation convergence nodes to obtain usable components. Generate action enhancement factors based on usable components.

[0060] Specifically, a deviation propagation network is constructed by combining the complete execution sequence and the deviation feature vector. The complete execution sequence provides the action flow after borrowing fusion repair, where each action node is arranged in execution order. These actions are directly converted into a set of network nodes, with the total number of nodes equal to the number of actions in the sequence. The deviation feature vector V_dev records the quantitative characteristics of various deviations. Key components such as the deviation growth rate r_dev and the maximum deviation D_max are extracted from it to determine node attributes. r_dev is obtained by calculating the first derivative of the deviation time series, and D_max is the peak value in the deviation sequence. r_dev is mapped to the deviation sensitivity of the node, and D_max is mapped to the deviation tolerance. The edges of the network are established according to the temporal relationship of the execution sequence. Directed edges are established between adjacent actions, and the edge weight w_ij is determined by calculating the deviation correlation coefficient between adjacent time steps in the deviation feature vector, w_ij=ρ(V_dev(i),V_dev(j)), where ρ is the correlation coefficient. For indirect effects across actions, when the deviation feature vector shows that the deviations of two non-adjacent actions are highly correlated (ρ>0.7), a jump connection edge is added. For example, the three actions of "grabbing-moving-placing" in the complete execution sequence are converted into three network nodes. The deviation feature vector shows that the force deviation during grasping is correlated with the placement accuracy by 0.82. Therefore, a direct connection with a weight of 0.82 is established between the grasping and placement nodes.

[0061] In some embodiments, the step of analyzing the propagation path of the deviation in the action chain through the deviation propagation network to obtain deviation amplification nodes and deviation convergence nodes includes: inputting the deviation feature vector as an initial stimulus into the deviation propagation network; calculating the cumulative intensity and propagation speed of the deviation feature vector at each node using a network propagation algorithm; identifying nodes where the cumulative intensity continuously increases and the propagation speed accelerates as deviation amplification nodes; and identifying nodes where the cumulative intensity gradually decreases and the propagation speed slows down as deviation convergence nodes.

[0062] The deviation feature vector is injected as the initial excitation signal into the corresponding node of the deviation propagation network. Each component of the deviation feature vector is mapped to the source node in the network that generated the deviation, serving as the starting state for propagation analysis. The excitation intensity is determined according to the severity of the deviation; indicators such as peak deviation and growth rate in the deviation feature vector are directly converted into excitation amplitude. The injection method considers the spatiotemporal distribution of the deviation; instantaneous deviations are excited by pulses, while continuous deviations are excited by steps. In the case of multi-source excitation, different deviation sources may produce coherent superposition or destructive interference effects. The temporal characteristics of the excitation preserve the dynamic evolution information of the original deviation. The initial state setting of the network ensures that all non-excited nodes are in the baseline state. The boundary conditions for excitation propagation are set according to task constraints to prevent the analysis results from exceeding a reasonable range. The initial excitation calibration process verifies the integrity and accuracy of the input signal. After system mapping and parameter setting, the conversion of the deviation feature vector into network excitation is completed.

[0063] A network propagation algorithm is used to analyze the dynamic evolution of the deviation excitation at each node. Based on signal flow graph theory, the injected initial excitation is used as the input signal of the network. Initially, the deviation value D_in(i,0) of the excitation node is equal to the injected excitation intensity, and D_in(j,0) = 0 for other nodes. The propagation process proceeds in discrete time steps. The deviation output of node i at step k+1 is calculated as: D_out(i,k+1) = f_i(Σw_ji×D_out(j,k)), where D_out(j,k) is the output of node j at step k (which is also the input of the downstream node), f_i is the transfer function of node i, and w_ji is the connection weight from node j to i. In this way, the initial excitation propagates gradually through the network connections, forming the dynamic evolution of the deviation. The cumulative intensity is calculated using an integral form: S_i(t) = S_i(0) + ∫D_i(τ)dτ, where S_i(0) is the initial intensity of the excited node, equal to the corresponding component value of the deviation eigenvector, and S_i(0) = 0 for non-excited nodes. The propagation velocity v_i is determined by calculating the arrival time of the deviation peak: v_i = d_i / Δt_i, where d_i is the topological distance between nodes, and Δt_i is the time delay for the deviation peak to propagate from the upstream node to node i. Finally, the cumulative intensity and propagation velocity data for each node are obtained.

[0064] Based on the changing characteristics of cumulative intensity and propagation velocity, deviation amplification nodes in the network are identified. Deviation amplification nodes are characterized by output deviations exceeding input deviations, and this amplification effect intensifies over time. Identification criteria include: a consistently positive first derivative of the cumulative intensity (dS_i / dt > 0) indicates continuously increasing deviation; an acceleration of the propagation velocity (d²v_i / dt² > 0) indicates a rapidly expanding influence range; and a node transfer function gain (G_i = S_i(t_out) / S_i(t_in) > 1 indicates the node possesses inherent amplification characteristics. Typical deviation amplification nodes appear in gait transitions, center of gravity shifts, and bi-arm coordination. For example, during the dynamic walking process of a humanoid robot, a small angular deviation in the supporting ankle joint is transmitted upwards through the kinematic chain, amplified several times at the torso position, potentially leading to loss of balance; when both arms are working together to lift heavy objects, force control deviations in one arm can disrupt the coordination between the arms, causing load tilting and the risk of falling. The distribution of amplification nodes exhibits a clustering characteristic, typically concentrated at key transition points in the kinematic chain, such as the hip and shoulder joints, which are multi-degree-of-freedom coupling points. The time-varying amplification characteristics indicate that certain nodes only exhibit an amplification effect under specific conditions, such as the significantly enhanced deviation amplification effect of the waist node during rapid turns. Identifying deviation amplification nodes is significant because they represent weak points in the positioning system; these nodes are the source of uncontrolled deviations and require focused monitoring and optimization.

[0065] Similar analytical methods are used to identify convergence nodes in the deviation propagation network. Deviation convergence nodes possess the ability to absorb and attenuate deviations, manifested as output deviations being smaller than input deviations. Convergence characteristics include: cumulative intensity decreasing over time with an exponential decay trend, where λ is the attenuation coefficient; propagation speed gradually decreasing until it stops; and the node transfer function exhibiting low-pass filtering characteristics, with a gain G_i < 1. Convergence mechanisms originate from adaptive adjustment of actions, redundant degrees of freedom, and compliant control. Force control absorbs position deviations through compliance, while visual servoing eliminates accumulated errors through real-time correction. Convergence nodes frequently appear in feedback control loops, fault-tolerant mechanism activation points, and human-machine interfaces. The convergence rate reflects the efficiency of the node in eliminating deviations; fast convergence nodes can reduce deviations to acceptable levels within several control cycles. Stability analysis of convergence ensures that oscillations caused by over-correction do not occur. The capacity limitation of convergence nodes determines the maximum amount of deviation they can handle. The value of identifying deviation convergence nodes lies in discovering the system's self-healing capabilities; these nodes can interrupt the deviation propagation chain and are a key resource for improving system robustness.

[0066] For the identified amplification and convergence nodes, a regulatory mechanism analysis is conducted to obtain usable components. The goal of the regulatory analysis is to understand the deviation handling mechanism of the nodes and to uncover factors that can be used to improve system performance. For amplification nodes, the causes of amplification are analyzed: kinematic amplification due to structural constraints, such as the inverted pendulum structure of a humanoid robot amplifying bottom deviations at the top; over-response caused by improper control gain settings; and coupling resonance between environmental interference and the system's natural frequency. For convergence nodes, the convergence mechanism is studied: the active compensation strategy of the ankle joint eliminates deviations by adjusting the support torque in real time; the passive compliance characteristics of the flexible spine naturally absorb shocks; and the fusion of vision and inertial sensors provides accurate state estimation. These convergence mechanisms are valuable resources of the system and can be extended to other nodes. The exploration of the regulatory space identifies adjustable parameters, such as control gain K_p, damping coefficient ζ, and filtering parameter τ. Adjusting these parameters can change the deviation handling characteristics of the nodes. Sensitivity analysis reveals the degree of influence of each parameter on the node characteristics. ∂G / ∂K_p represents the sensitivity of the gain to the amplification rate; parameters with higher sensitivity have greater optimization potential. Constraints define the feasible domain of regulation, ensuring that adjustments do not compromise system stability. The optimal regulation strategy is determined through multi-objective optimization, balancing deviation suppression and system performance. Available components refer to all elements identified through the above analysis that can be used to improve the system, including adjustable parameter ranges, constraints, and optimized control strategies. The identified adjustable parameters, constraints, and control strategies are quantified and collected to obtain a multidimensional dataset of the regulation characteristics of each node, which serves as the available components.

[0067] In some embodiments, generating action enhancement factors based on the available components includes: performing principal component analysis on the available components to extract key deviation features; constructing a deviation enhancement mapping relationship using the key deviation features and the core action sequence; generating standard enhancement parameters based on the deviation enhancement mapping relationship; and generating action enhancement factors that act on action execution by transforming the standard enhancement parameters through vector space.

[0068] Principal component analysis (PCA) is performed on the available components to extract key deviation features. The available component dataset contains multidimensional parameters extracted from deviation amplification nodes and convergence nodes. First, the parameters of each dimension are standardized: X_norm = (X - μ) / σ, where X is the original data, μ is the mean, and σ is the standard deviation, eliminating the influence of dimensional differences. The covariance matrix of the standardized data is calculated, and the principal component directions are obtained through eigenvalue decomposition. The k principal components with cumulative contribution rates exceeding a set threshold are selected based on the magnitude of the eigenvalues. The principal component loading matrix W shows the contribution weights of the original parameters to the principal components. The first principal component usually reflects the overall control capability, the second principal component corresponds to the rapid response characteristics, and the third principal component is associated with the stability index. Key deviation features are obtained through projection transformation: Y = X_norm × W, where Y is the dimensionality-reduced feature matrix, achieving effective compression from the multidimensional parameter space to the key feature space. The feature selection process not only reduces computational complexity but also highlights the key factors that have the most significant impact on system performance.

[0069] A nonlinear mapping relationship between deviation and enhancement is established using extracted key deviation features and core action sequences. The core action sequence provides a standard action template A_core=[a1,a2,...,an], where ai represents the i-th action, and each action includes basic attributes such as position, velocity, and force. Key deviation features Y represent a quantitative description of the system's weak points, requiring the establishment of a mapping function from deviation features to enhancement strategies. Specialized mapping relationships are constructed based on deviation type classification: position deviation uses trajectory optimization mapping E_pos=k1·y1+k2·y1² / (1+α|y1|), where E_pos is the position enhancement amount, y1 is the position deviation feature, k1 and k2 are mapping coefficients, and α is a saturation parameter. Force deviation and velocity deviation use impedance adjustment mapping and dynamic compensation mapping, respectively, forming a complete family of mapping functions. Mapping coefficients are obtained through training with historical data, and least squares fitting is used to ensure mapping accuracy. The mapping relationship also considers the coupling effect between actions; the enhancement amounts of adjacent actions are connected through a smoothing function to avoid abrupt changes in the enhancement strategy. At key transition points such as attitude switching and contact establishment, an enhancement weight factor w_critical is set to increase the enhancement priority. The established mapping relationship forms a structured enhancement lookup table, enabling rapid conversion from deviation features to enhancement strategies.

[0070] Based on the deviation enhancement mapping relationship, the mapping output is converted into directly applicable standard enhancement parameters. These standard enhancement parameters include the position compensation vector Δp=[Δpx,Δpy,Δpz], the force control gain correction coefficient α_f, the velocity curve optimization parameter β_v, and the acceleration limit parameter a_max, where Δpx, Δpy, and Δpz are the position compensation components of the three coordinate axes, respectively. The parameter standardization process considers the actuator's physical constraints, limiting position compensation within the workspace: ||Δp||≤p_max, where p_max is the maximum allowable compensation amplitude. Force gain correction maintains system stability. The enhancement intensity grading design provides a multi-level selection mechanism, automatically selecting the appropriate enhancement level based on the severity of the deviation to ensure that the enhancement effect matches the degree of deviation. The parameter timing arrangement is described by a time function: P_std(t) = P_base + P_enhance·h(t - t_start), where P_std(t) is the standard parameter at time t, P_base is the base parameter, P_enhance is the enhancement amplitude, h(t) is the enhancement activation function, and t_start is the enhancement start time. A sigmoid function is used to achieve smooth start and end. Compatibility verification is performed through simulation to ensure that the enhancement parameters do not cause control conflicts or system oscillations. The verification process includes stability analysis and performance evaluation. The standardized parameter format ensures compatibility with different types of actuators, while the modular design of the parameters supports flexible combination and adjustment.

[0071] The standard enhancement parameters are mapped to the humanoid robot's execution space through vector space transformation, generating motion enhancement factors that can directly drive the actuators. Based on the humanoid robot's kinematics and dynamics model, the transformation matrix T consists of a Jacobian matrix J and an inertial matrix M: T = J^T·M, where J is the Jacobian matrix, M is the inertial matrix, and J^T is the transpose of the Jacobian matrix. The Jacobian matrix J describes the velocity mapping relationship from joint space to Cartesian space, and its transpose J^T provides the conversion from Cartesian forces to joint torques. The inertial matrix M reflects the mass distribution and inertial characteristics of each link of the humanoid robot, ensuring that the enhancement factor conforms to the dynamic constraints. The enhancement factor is calculated as: F_enhance = T·P_std = J^T·M·P_std, where F_enhance is the enhancement factor vector, P_std is the standard enhancement parameter vector, and the output F_enhance includes the torque enhancement components of each joint. During the transformation process, kinematic singularities are handled, and damped least squares method is used to avoid numerical instability when the determinant of the Jacobian matrix approaches zero. The time-varying enhancement factor generates continuous trajectories through interpolation, and cubic spline interpolation ensures acceleration continuity, guaranteeing the smoothness of the enhancement process. The generation of the motion enhancement factor enables proactive optimization and adaptive adjustment of system performance. Taking a precision assembly task as an example, when it is predicted that the "insertion" action may fail due to positional deviation, the enhancement factor improves the success rate of task execution by adjusting the torque distribution and velocity curves of relevant joints in advance. The enhancement factor not only compensates for the inherent deviation amplification characteristics of the system but also utilizes the control mechanism of deviation convergence nodes to extend local convergence capabilities to the global level, improving the overall robustness of the system. This proactive enhancement mechanism realizes the transformation from passive fault tolerance to proactive optimization, enabling humanoid robots to prevent potential performance degradation and maintain stable and efficient task execution capabilities in complex industrial environments.

[0072] Step S160: The motion enhancement factor and the entangled motion vector chain are co-optimized to generate an enhanced control sequence, and the enhanced control sequence is used to drive the humanoid robot.

[0073] Specifically, firstly, specific control adjustment instructions are extracted from the action enhancement factors, including key parameters such as torque compensation amplitude, trajectory offset, attitude adjustment angle, and response time window. These parameters are derived from in-depth analysis of deviation amplification nodes and convergence nodes. Simultaneously, the spiral topology information is read from the entangled action vector chain to obtain the spatial coordinates of each action node, the connection relationships between adjacent nodes, the entanglement strength values ​​between nodes, and the temporal execution order. This structured data provides a navigational foundation for the precise positioning of the enhancement factors. The collaborative optimization process matches the enhancement factors with the vector chain nodes one-to-one using action identifier codes. The target node identifier carried by each enhancement factor is compared with the node number stored in the vector chain. Upon successful matching, the enhancement parameters are injected into the corresponding basic action vector. For nodes in highly entangled regions, the enhancement process needs to consider the chain reaction of adjacent nodes. The influence range is determined by reading the entanglement coefficient values. When the entanglement coefficient is high, the enhancement of a single node will synchronously adjust the execution parameters of its associated nodes to ensure that the coordination of the action combination is not disrupted. The optimization strategy employs a hierarchical approach, decomposing the complex vector chain structure into target adjustment at the task level, parameter correction at the action level, and drive commands at the joint level. Each level receives enhancement information at its corresponding granularity, forming a complete control chain from macro-planning to micro-execution. For example, when handling precision assembly tasks, the force control accuracy improvement command in the enhancement factor is assigned to the "insertion" action node. Simultaneously, the highly entangled "alignment" node also receives a coordination enhancement command, ensuring synchronous optimization of the two actions. Ultimately, an enhanced control sequence with a fusion deviation utilization mechanism is generated.

[0074] The generated enhanced control sequences drive the humanoid robot to perform tasks, achieving proactive utilization of deviations and performance optimization. A real-time controller converts sequence commands into motor drive signals. The deviation utilization mechanism transforms deviations that would otherwise need to be eliminated into beneficial adjustment resources; for example, it uses gravity deviation to assist descent and inertial deviation to accelerate turning. A real-time monitoring module tracks the execution status, maintaining the current strategy when the actual deviation matches the predicted path and activating the compensation mechanism when deviations occur. The enhancements are reflected in reduced action completion time, improved accuracy, and reduced energy consumption. Particularly when handling non-integer cycle actions, the enhanced borrowing fusion mechanism allows previously interrupted actions to be completed smoothly. Execution verification confirms effective suppression of identified deviation amplification nodes and full utilization of convergence nodes, maintaining operational stability throughout the process. The precise execution of the enhanced control sequences and the ingenious transformation of deviations ultimately successfully drive the humanoid robot, achieving optimized control through deviation utilization.

[0075] To implement the vector entanglement-based humanoid robot control method corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2This diagram illustrates a structural block diagram of a humanoid robot control device 200 based on vector entanglement according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The humanoid robot control device 200 based on vector entanglement provided in this embodiment includes: The vector construction module 201 is used to receive the humanoid robot control task, perform frequency statistical analysis on the control task to extract the core action sequence, construct basic action vectors based on the core action sequence, and generate an entangled action vector chain using the basic action vectors through partial overlap and entanglement. Semantic mapping module 202 is used to construct an initial action semantic graph by performing dynamic semantic mapping using the entangled action vector chain, to evolve and update the initial action semantic graph based on real-time execution feedback, and to generate a deviation prediction path through the updated action semantic graph. Deviation analysis module 203 is used to execute humanoid robot actions according to the deviation prediction path to generate actual execution deviation data, and compare and analyze the actual execution deviation data with the deviation prediction path to generate a deviation feature vector; The action completion module 204 is used to identify non-integer periodic actions using the deviation feature vector, extract entangled action vectors that overlap with the non-integer periodic actions from the entangled action vector chain, and perform borrowing fusion on the entangled action vectors to generate a complete execution sequence. The deviation enhancement module 205 is used to construct a deviation propagation network by combining the complete execution sequence and the deviation feature vector, analyze the diffusion path of the deviation in the action chain through the deviation propagation network to obtain deviation amplification nodes and deviation convergence nodes, perform deviation regulation analysis on the deviation amplification nodes and deviation convergence nodes to obtain usable components, and generate action enhancement factors based on the usable components. The control optimization module 206 is used to collaboratively optimize the motion enhancement factor and the entangled motion vector chain to generate an enhanced control sequence, and to use the enhanced control sequence to drive the humanoid robot.

[0076] The aforementioned vector entanglement-based humanoid robot control device 200 can implement the vector entanglement-based humanoid robot control method of the above-described 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's embodiments can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0077] 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 humanoid robot control method based on vector entanglement described in the first embodiment of the present invention.

[0078] 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.

[0079] 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 humanoid robot control method based on vector entanglement, characterized in that, include: Receive a humanoid robot control task, perform frequency statistical analysis on the control task to extract core action sequences, construct basic action vectors based on the core action sequences, and generate an entangled action vector chain using the basic action vectors through partial overlap and entanglement. An initial action semantic graph is constructed by using the entangled action vector chain for dynamic semantic mapping. The initial action semantic graph is then updated based on real-time execution feedback. A deviation prediction path is generated using the updated action semantic graph. The humanoid robot performs actions according to the deviation prediction path to generate actual execution deviation data. The actual execution deviation data and the deviation prediction path are compared and analyzed to generate a deviation feature vector. The deviation feature vector is used to identify non-integer periodic actions. Entangled action vectors that overlap with the non-integer periodic actions are extracted from the entangled action vector chain. The entangled action vectors are then fused by borrowing to generate a complete execution sequence. A deviation propagation network is constructed by combining the complete execution sequence and the deviation feature vector. The propagation network is used to analyze the diffusion path of the deviation in the action chain to obtain deviation amplification nodes and deviation convergence nodes. Deviation regulation analysis is performed on the deviation amplification nodes and deviation convergence nodes to obtain usable components. Based on the usable components, an action enhancement factor is generated. The motion enhancement factor and the entangled motion vector chain are collaboratively optimized to generate an enhanced control sequence, which is then used to drive the humanoid robot.

2. The method as described in claim 1, characterized in that, Using the aforementioned basic action vectors, an entangled action vector chain is generated through partial overlap entanglement, including: Calculate the vector angle and magnitude difference between temporally adjacent vectors in the basic action vector to generate a similarity matrix; The vector overlap region is determined based on the similarity matrix; Within the overlapping region, the vector components are nonlinearly weighted and fused to generate entanglement coefficients; The entanglement coefficients are used to construct an entangled action vector chain through a spiral connection method.

3. The method as described in claim 1, characterized in that, The updated action semantic graph generates biased prediction paths, including: Historical deviation nodes are extracted from the updated action semantic graph to construct a deviation evolution subgraph; The deviation evolution subgraph is used to calculate the deviation transition probabilities between nodes and generate a deviation state matrix; Based on the aforementioned deviation state matrix, a multi-step prediction is performed to generate a time-series deviation sequence; The time-series deviation sequence is mapped to the execution time axis to form a deviation prediction path.

4. The method as described in claim 1, characterized in that, Borrowing and fusing the entangled action vectors to generate a complete execution sequence includes: Based on the non-integer periodic action, locate adjacent entangled nodes in the entangled action vector chain; Extract the borrowable vector components from the adjacent entangled nodes to generate a compensation vector set; The compensation vector set and the non-integer cycle action are time-aligned to generate a fusion weight matrix; The complete execution sequence is generated by performing vector fusion operations using the fusion weight matrix.

5. The method as described in claim 1, characterized in that, The deviation propagation network is used to analyze the propagation path of the deviation in the action chain to obtain the deviation amplification node and the deviation convergence node, including: The deviation feature vector is used as the initial excitation input to the deviation propagation network; The cumulative strength and propagation speed of the deviation feature vector at each node are calculated using a network propagation algorithm. Nodes whose cumulative intensity continuously increases and whose propagation speed accelerates are identified as deviation amplification nodes; Nodes whose cumulative intensity gradually decreases and whose propagation speed slows down are identified as deviation convergence nodes.

6. The method as described in claim 1, characterized in that, The generation of motion enhancement factors based on the available components includes: Principal component analysis was performed on the available components to extract key deviation features; A deviation enhancement mapping relationship is constructed using the key deviation features and the core action sequence; Standard enhancement parameters are generated based on the aforementioned deviation enhancement mapping relationship; The standard enhancement parameters are transformed into action enhancement factors that act on the execution of the action through vector space transformation.

7. The method as described in claim 2, characterized in that, The entangled action vector chain is constructed using the entanglement coefficients in a spiral connection manner, including: The radial distribution of the spiral is determined based on the force vector component in the basic motion vector, and the tangential velocity of the spiral is determined using the velocity vector component in the basic motion vector. The helical trajectory equation is generated by combining the radial distribution and the tangential velocity; The spatial position of each vector is calculated based on the spiral trajectory equation and the entanglement coefficient; A spiral entangled action vector chain is formed by weighting the connection of the spatial locations using entanglement coefficients.

8. The method as described in claim 4, characterized in that, Extracting borrowable vector components from the adjacent entangled nodes to generate a compensation vector set includes: Analyze the vector redundancy of the adjacent entangled nodes and identify the borrowable components whose redundancy exceeds a threshold; For the missing type of the non-integer cycle action, the borrowing priority is determined, with force component taking priority over velocity component, and velocity component taking priority over direction component. The borrowing priority and the entangled action vector chain are combined to construct the borrowing weight, and the closer the nodes are, the greater the borrowing weight. Based on the borrowing weight, vector components are dynamically extracted from the borrowable components to form a compensation vector set.

9. A humanoid robot control device based on vector entanglement, characterized in that, include: The vector construction module is used to receive humanoid robot control tasks, perform frequency statistical analysis on the control tasks to extract core action sequences, construct basic action vectors based on the core action sequences, and generate entangled action vector chains using the basic action vectors through partial overlap and entanglement. The semantic mapping module is used to construct an initial action semantic graph by performing dynamic semantic mapping using the entangled action vector chain, to evolve and update the initial action semantic graph based on real-time execution feedback, and to generate a deviation prediction path through the updated action semantic graph. The deviation analysis module is used to execute humanoid robot actions according to the deviation prediction path to generate actual execution deviation data, and compare and analyze the actual execution deviation data with the deviation prediction path to generate a deviation feature vector. The action completion module is used to identify non-integer periodic actions using the deviation feature vector, extract entangled action vectors that overlap with the non-integer periodic actions from the entangled action vector chain, and perform borrowing fusion on the entangled action vectors to generate a complete execution sequence. The deviation enhancement module is used to construct a deviation propagation network by combining the complete execution sequence and the deviation feature vector, analyze the diffusion path of the deviation in the action chain through the deviation propagation network to obtain deviation amplification nodes and deviation convergence nodes, perform deviation regulation analysis on the deviation amplification nodes and deviation convergence nodes to obtain usable components, and generate action enhancement factors based on the usable components. The control optimization module is used to collaboratively optimize the motion enhancement factor and the entangled motion vector chain to generate an enhanced control sequence, and to use the enhanced control sequence to drive the humanoid robot.

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