A humanoid robot remote control method, device and equipment based on wireless communication

Abstract

The application provides a kind of human-shaped robot remote control method, device and equipment based on wireless communication, generates control intention data by multi-modal semantic analysis, establishes two-way expected calibration channel to generate expected consistency parameter and deviation tolerance threshold;Based on the expected consistency parameter, the expected execution result is obtained, the optimal control sequence is obtained by reverse causal reasoning, and the prohibited execution action set is identified by negative space analysis;Combine control sequence and prohibited action to build paradox verification scene to generate control boundary parameter;Based on the boundary parameter, the interlocking verification condition is established to generate a safe unlocking sequence;Parallel monitoring is carried out on the control sequence to obtain execution verification data to generate a unified instruction stream;The instruction stream is transmitted to the robot end for verification and execution, and the actual execution state data is obtained;Based on the state data, convergence analysis is carried out, the causal feedback chain is obtained for predictive correction to complete remote control, and intelligent and safe human-shaped robot remote control is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot control, in particular to a humanoid robot remote control method, device and equipment based on wireless communication. BACKGROUND

[0002] As an important development direction in the field of intelligent manufacturing and service robots, humanoid robots have shown great application potential in the fields of elderly care, medical rehabilitation, home service, etc. With the acceleration of population aging and the rapid growth of remote service demand, the remote control technology of humanoid robots based on wireless communication has become a key technical support for realizing intelligent nursing service and fine auxiliary operation. Remote control technology can break through geographical space limitations, enabling professional operators to remotely guide robots to complete complex nursing tasks, providing an effective way to solve the problem of uneven distribution of nursing resources and shortage of professionals.

[0003] However, the existing remote control method has significant technical limitations: the control instruction processing lacks intelligent semantic understanding and intent recognition mechanism, and cannot accurately grasp the real needs and expected targets of the controller; the control sequence planning mostly uses forward planning method, lacks reverse reasoning and safety verification ability, and is difficult to effectively avoid potential risks while ensuring execution effect; the safety protection mechanism is simple and rough, lacks multiple interlocking verification and dynamic boundary detection, and is prone to security vulnerabilities and control failures; the execution monitoring system is imperfect, lacks real-time state sampling and convergence analysis ability, and cannot timely find deviations and adjust strategies. These problems seriously restrict the control accuracy, safety and reliability, and intelligence level of humanoid robots in complex scenarios. SUMMARY

[0004] The present application provides a humanoid robot remote control method, device and equipment based on wireless communication, aiming to solve the key technical problems of existing remote control technology in control understanding, safety verification and execution monitoring. The technical solution integrates multi-modal semantic analysis and reverse causal reasoning technology to realize intelligent conversion from control instruction to execution sequence, integrates paradox verification analysis and interlocking safety control technology to realize multiple safety protection mechanisms, and integrates predictive feedback correction technology to realize dynamic optimization of control strategy, building a complete intelligent remote control technology system.

[0005] The present application provides a humanoid robot remote control method, device and equipment based on wireless communication, aiming to solve the key technical problems of existing remote control technology in control understanding, safety verification and execution monitoring. The technical solution integrates multi-modal semantic analysis and reverse causal reasoning technology to realize intelligent conversion from control instruction to execution sequence, integrates paradox verification analysis and interlocking safety control technology to realize multiple safety protection mechanisms, and integrates predictive feedback correction technology to realize dynamic optimization of control strategy, building a complete intelligent remote control technology system.

[0006] Receiving remote control instructions from the controller, performing semantic analysis on the remote control instructions to generate control intent data, establishing a bidirectional expectation calibration channel based on the control intent data, and generating expectation consistency parameters and deviation tolerance thresholds based on the bidirectional expectation calibration channel;

[0007] obtain an expected execution result based on the expected consistency parameter, deduce an optimal operation sequence from the expected execution result, and perform negative space analysis on the optimal operation sequence to identify a set of prohibited execution actions;

[0008] construct a paradox verification scenario in combination with the optimal operation sequence and the set of prohibited execution actions, and perform a contradiction conflict test through the paradox verification scenario to generate an operation boundary parameter;

[0009] establish an interlocking verification condition based on the operation boundary parameter and the deviation tolerance threshold, and generate a safe unlocking sequence based on the interlocking verification condition;

[0010] perform parallel monitoring on the optimal operation sequence based on the safe unlocking sequence to obtain execution verification data, determine a target execution branch according to the execution verification data, and generate a unified instruction stream based on the target execution branch;

[0011] transmit the unified instruction stream to the robot end through wireless communication, perform instruction verification at the robot end based on the safe unlocking sequence to generate a verification pass signal, and drive the robot to execute based on the verification pass signal to obtain actual execution state data;

[0012] perform convergence analysis based on the actual execution state data and the deviation tolerance threshold to generate a convergence monitoring result, obtain a causal feedback chain according to the convergence monitoring result, and use the causal feedback chain to predictively correct subsequent operation behaviors to complete remote operation.

[0013] The second aspect of the present application proposes a humanoid robot remote operation device based on wireless communication, comprising:

[0014] An instruction analysis module is configured to receive remote operation instructions of an operator, perform semantic analysis on the remote operation instructions to generate operation intention data, establish a bidirectional expected calibration channel based on the operation intention data, generate an expected consistency parameter and a deviation tolerance threshold based on the bidirectional expected calibration channel.

[0015] An inverse reasoning module is configured to obtain an expected execution result based on the expected consistency parameter, deduce an optimal operation sequence from the expected execution result, and perform negative space analysis on the optimal operation sequence to identify a set of prohibited execution actions.

[0016] A paradox verification module is configured to construct a paradox verification scenario in combination with the optimal operation sequence and the set of prohibited execution actions, and perform a contradiction conflict test through the paradox verification scenario to generate an operation boundary parameter.

[0017] A safety control module is configured to establish an interlocking verification condition based on the operation boundary parameter and the deviation tolerance threshold, and generate a safety unlocking sequence based on the interlocking verification condition;

[0018] An execution monitoring module is configured to monitor the optimal operation sequence in parallel based on the safety unlocking sequence to obtain execution verification data, determine a target execution branch according to the execution verification data, and generate a unified instruction stream based on the target execution branch;

[0019] A communication execution module is configured to transmit the unified instruction stream to a robot end through wireless communication, perform instruction verification at the robot end based on the safety unlocking sequence to generate a verification pass signal, and drive the robot to execute to obtain actual execution state data based on the verification pass signal;

[0020] A feedback optimization module is configured to perform convergence analysis based on the actual execution state data and the deviation tolerance threshold to generate a convergence monitoring result, obtain a causal feedback chain according to the convergence monitoring result, and predictively correct a subsequent operation behavior by using the causal feedback chain to complete remote operation.

[0021] The third aspect of the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the remote operation method of the humanoid robot based on wireless communication when executing the program.

[0022] The beneficial effects of the present application are embodied in the following points: first, through the multi-modal semantic analysis technology, the real intention of the operator can be accurately understood by simultaneously processing multiple input signals such as voice, gesture, and brain electricity. Through the bidirectional expectation calibration mechanism, the operation demand is matched with the actual ability of the robot, avoiding the problems of "expectation too high to execute" or "ability waste". The reverse causal reasoning method is adopted to plan the execution path from the target result, which is more reasonable and efficient than the traditional forward planning method. Second, a three-layer progressive safety protection system is established. The dangerous action set that is prohibited to execute is identified in advance through the negative space analysis technology to realize preventive safety protection; the system bearing capacity is tested under extreme conditions through the paradox verification scene technology to determine the operation boundary parameters; the progressive safety unlocking of the function is realized through the multiple interlocking verification conditions to ensure that each operation is within the verified safe range. The active protection mechanism of "prevention-test-verification" can more effectively prevent potential risks than the traditional passive protection method. Third, a complete remote operation closed-loop control system is realized. The multi-dimensional execution state data is obtained through the parallel monitoring mechanism, and the real-time instruction transmission and verification are realized based on wireless communication. The problem root is identified through the hierarchical and speed causal feedback chain, including the fast and strong correlation core control layer, the slow and strong correlation steady-state regulation layer, the fast and weak correlation disturbance compensation layer, and the slow and weak correlation trend prediction layer. The control parameters are adjusted in advance through the predictive correction technology, which can reduce the operation deviation compared with the traditional reactive control, so that the robot can maintain stable operation performance in complex environments.

[0023] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0024] The drawings herein show specific examples of the technical solutions described in the present application, and constitute part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0025] Unless specifically stated or defined otherwise, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0026] Figure 1 is a flowchart of a remote operation method of a humanoid robot based on wireless communication according to the present application.

[0027] Figure 2 is a structural block diagram of a remote operation device of a humanoid robot based on wireless communication according to the present application.

[0028] Figure 3 is a structural diagram of a computer device according to the present application. DETAILED DESCRIPTION

[0029] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0030] It is to be understood that the terminology "including", "comprising", "consisting" and "consisting essentially of" used in the specification and the appended claims, indicates the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0031] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment, but can refer to one or more but not all embodiments. Furthermore, the terms "a" or "an", as used herein, mean "one or more" unless otherwise explicitly stated.

[0032] The technical solutions of the embodiments of the present application are introduced as follows.

[0033] As shown in Figure 1 The embodiment of the present application provides a humanoid robot remote control method based on wireless communication, which comprises the following steps S110-S170:

[0034] S110, receiving the remote control instruction of the operator, performing semantic analysis on the remote control instruction to generate control intention data, establishing a bidirectional expected calibration channel based on the control intention data, and generating expected consistency parameters and deviation tolerance thresholds based on the bidirectional expected calibration channel.

[0035] The remote control instruction receiving adopts a multi-modal input fusion architecture, integrating voice instructions, gesture recognition, brain-computer interface, and tactile feedback and other multi-control input methods for parallel processing. The voice instruction receiving configures a high directivity microphone array, adopts a 16 kHz sampling frequency for real-time audio acquisition, and ensures the clarity of the instructions through noise suppression and echo cancellation technology. The gesture recognition system integrates an RGB-D depth camera and an IMU sensor to capture the three-dimensional hand movement trajectory of the operator in real time at a frame rate of 30 fps. The brain-computer interface module adopts a 64-channel EEG electrode array to record the brain electrical activity signal at a high frequency of 1000 Hz, and uses a convolutional neural network to extract motor imagery features. The tactile feedback device configures a pressure sensor and a vibration actuator to establish a bidirectional force perception interaction channel between the operator and the robot. All input signals are accurately time-aligned through hardware clock synchronization and software timestamp correction, and the distributed time protocol is used to ensure the synchronous fusion of multi-modal data. For example, in the home care scene, when the operator issues the "help me help the old man sitting on the sofa" instruction through voice, the multi-modal fusion system synchronously analyzes the urgency in the voice tone, the specific position of the gesture pointing, and the intensity of the control intention reflected by the brain electrical signal, and generates a complete remote control instruction.

[0036] The semantic analysis adopts a hierarchical semantic understanding architecture, which converts the remote control instruction into structured control intention data understandable by the robot through a multi-level processing flow of morphological analysis, syntactic analysis, semantic reasoning, and intent recognition. The natural language processing engine is based on a pre-trained language model of the Transformer architecture, which is trained through domain adaptation of human-computer interaction corpus to establish a professional semantic knowledge base covering human robot control scenarios. The voice signal processing adopts an end-to-end speech recognition model to realize the direct mapping conversion from audio waveform to semantic intention. The gesture semantic analysis establishes a three-dimensional space gesture dictionary and a time sequence action mode library to decompose continuous hand movement sequences into discrete semantic primitives. The EEG signal semantic decoding adopts a hybrid architecture of convolutional neural network and long short-term memory network to extract stable motor intention representations from high-dimensional EEG feature space. The semantic fusion module uses attention mechanism and graph neural network technology to allocate weights and associate reasoning for semantic fragments from different modalities, solving the conflict and ambiguity problems among multi-modal information. The context understanding mechanism maintains dynamic dialogue state and task context, and uses memory network and knowledge graph reasoning technology to complete incomplete instruction expression. For example, when the operator issues the ambiguous instruction "robot come over", the semantic analysis system automatically infers the specific care needs and execution plan by combining the current room layout, the old man's physical condition, and the historical care records, generating complete structured control intention data.

[0037] The manipulation intention data generated based on semantic parsing establishes a two-way expectation calibration channel, realizing the intelligent matching framework of manipulation demand and robot capability. The manipulation intention data adopts a structured representation, including key information such as task type, target object, execution action, and constraint condition, such as the task of helping an old person, represented as I = {task: "assist_standing", object: "elderly", action: "lift", force_range: [50N, 150N], speed: "slow", position: "sofa"}. The two-way calibration channel is built with two core components: expectation vector generation and capability vector generation. The expectation vector extracts quantitative requirements from the manipulation intention data, including expected assistive force, expected speed, expected lifting angle, and expected completion time. The robot capability vector reflects the actual technical indicators, including maximum output force, maximum speed, joint range of motion, and fastest response time. The two-way calibration channel establishes a matching analysis mechanism, using a multi-dimensional similarity evaluation method to design a two-way message passing architecture with forward and reverse transmission. The forward transmission path evaluates the degree of satisfaction of robot capability to manipulation expectation, and the reverse transmission path evaluates the rationality of manipulation expectation relative to robot capability. Through structured analysis based on manipulation intention data and two-way matching mechanism design, a two-way expectation calibration channel is successfully established.

[0038] Based on the established two-way expectation calibration channel, the expectation consistency parameter and the deviation tolerance threshold are generated through an iterative negotiation mechanism. The expectation consistency parameter is generated through the iterative negotiation process of two-way message passing, and is dynamically updated according to the priority weight of each dimension in the manipulation intention data. In the iterative process, the forward transmission calculates the degree of satisfaction of robot capability to expectation, and the reverse transmission evaluates the rationality of expectation, and through multiple iterations, the final expectation consistency parameter value is converged. For the task of helping an old person, safety priority is the highest, and comfort is the second, and through the iterative negotiation of the two-way calibration channel, the expectation consistency parameter is converged to 0.83. The deviation tolerance threshold is also dynamically set based on the negotiation mechanism of the two-way calibration channel. First, the basic constraint conditions are extracted from the manipulation intention data, and then combined with the actual precision capability of the robot for two-way negotiation. When the robot precision capability is high, the threshold is appropriately tightened, and when the capability is limited, the threshold range is appropriately relaxed. The expected execution result is generated using a parameter-driven mapping method: R_expect = EC × R_ideal + (1-EC) × R_safe, which is dynamically adjusted between the ideal state and the safe state according to the expectation consistency parameter. Finally, the deviation tolerance threshold set is generated through the negotiation mechanism of the two-way calibration channel, including force deviation ±20N, speed deviation ±0.05m / s, and angle deviation ±15°, ensuring that the execution process meets the intention requirements and ensures safety margin. Through the two-way expectation calibration channel, the expectation consistency parameter and the deviation tolerance threshold are finally generated.

[0039] Step S120, obtaining an expected execution result based on the expected consistency parameter, and inversely deducing an optimal manipulation sequence from the expected execution result, and performing negative space analysis on the optimal manipulation sequence to identify a set of prohibited execution actions.

[0040] Specifically, the expected execution result is obtained based on the expected consistency parameter. The expected consistency parameter EC is used as a matching measure of the manipulation expectation and the robot capability, and directly determines the setting standard of the expected execution result. The expected execution result adopts a parameter-driven mapping method: R_expect=EC×R_ideal+(1-EC)×R_safe, where R_expect is the expected execution result, R_ideal is the ideal result state, R_safe is the safe conservative result, and EC is the expected consistency parameter. When EC=1.0, the expected execution result is equal to the ideal state; when EC decreases, the result automatically deviates to the safe state. The deviation tolerance threshold T is used as a hard constraint to limit the acceptable range of the result, and ensures that |R_expect-R_actual|≤T. The expected execution result is represented as a state vector R=[p_target,θ_target,t_exec,f_max], which includes the target position p_target, the target attitude θ_target, the execution time t_exec, and the maximum acting force f_max. For example, in a transfer assistance task, when EC=0.83, the expected execution result is: target position accuracy ±5 cm, attitude accuracy ±10°, execution time 8 seconds, and maximum assistance force 150 N. The ideal state requires higher accuracy (±2 cm, ±5°), while the safe state allows greater margin (±10 cm, ±20°). Through linear mapping of the expected consistency parameter and threshold constraint, the expected execution result is obtained.

[0041] In some embodiments, the optimal manipulation sequence is inversely deduced from the expected execution result, including: constructing a result state mapping table based on the expected execution result; inversely analyzing the result state mapping table to obtain a causal association path; and inversely deducing and generating the optimal manipulation sequence according to the causal association path.

[0042] A result state mapping table is constructed based on the desired execution result, which decomposes the target requirements in the desired execution result into an executable state sequence. The desired execution result provides end-point constraints and process limitations of the task, and a complete mapping from the current state to the target state needs to be constructed. The result state mapping table is represented as a directed graph G=(V,E), where the nodes V_i represent key states in the execution process, and the edges E_ij represent feasible transitions between states. Each state inherits constraints from the desired execution result, ensuring that S_i.f≤f_max and the execution time satisfies Σt_ij≤t_exec. The state is defined as S_i=[p_i,θ_i,f_i,t_i], where the position p_i and the pose θ_i change gradually from the initial to p_target and θ_target through interpolation. The state discretization density is adjusted according to the accuracy requirements of the desired execution result, with dense sampling for high accuracy requirements (EC>0.8) and sparse sampling for low accuracy requirements. The mapping weight w_ij reflects the priority of state transition, with higher weight for states closer to the desired execution result. For example, the mapping table constructed based on the desired execution result for the lifting task includes: initial approach, contact establishment, force lifting, position adjustment, stable holding, and force release, each state satisfying the force constraint f≤150N and the total time constraint of 8 seconds. Through decomposition and constraint propagation of the desired execution result, the result state mapping table is generated.

[0043] The result state mapping table is analyzed in reverse to identify the causal association path from the target state in the mapping table. The result state mapping table G=(V,E) provides all possible state transition relationships, and reverse analysis needs to find the optimal path that satisfies the causal logic. The dynamic programming algorithm starts from the target node V_goal and calculates the optimal value V(s_i)=min{c(s_i,s_j)+V(s_j)} to each node. The transition cost c(s_i,s_j) directly uses the weight w_ij in the mapping table and considers the physical feasibility of state transition. Causal constraints are extracted from the topological structure of the mapping table, such as the node V_contact (contact establishment) must be before V_lift (force lifting), which is represented as the existence of directed edges in the mapping table. During path search, only paths that satisfy all constraints in the mapping table are retained. The backtracking algorithm starts from V_goal and selects the predecessor node that minimizes the cumulative cost until the initial state V_start is reached. Path verification checks whether the extracted path covers all necessary nodes in the mapping table. For example, the causal path extracted from the mapping table of the lifting task is: V_start→V_approach→V_contact→V_lift→V_adjust→V_stable→V_goal, each transition corresponds to an edge in the mapping table. Through reverse traversal and constraint verification of the result state mapping table, the causal association path is obtained.

[0044] The optimal manipulation sequence is generated by backpropagation along the causal path, converting the discrete state sequence in the path to continuous control commands. The causal path V_0→V_1→...→V_n provides the logical order and timing constraints of state transitions, each state V_i contains target values of position, orientation, force, etc. Trajectory generation interpolates between adjacent states V_i and V_i+1, ensuring smooth transitions. For each transition, a trajectory segment of corresponding length is generated according to the time allocation t_i in the causal path. Key transitions (e.g. contact establishment at V_approach→V_contact) are allocated more control points to ensure precise execution. Interpolation methods are chosen according to path characteristics, cubic spline for position to guarantee continuity, linear interpolation for force to avoid abrupt changes. Joint space transformation maps Cartesian states S_i.p in the causal path to joint configurations q_i through inverse kinematics. Timing constraints ensure that the execution time of each trajectory segment strictly follows the planning in the causal path. Dynamics verification checks whether the generated trajectory satisfies the physical constraints implied in the causal path. For example, based on the causal path "approach→contact→lift→steady", the generated manipulation sequence maintains low speed during the contact segment (following the causal constraints) and constant contact force during the lift segment (satisfying the path requirements). Through segmented trajectory generation and constraint satisfaction based on the causal path, the optimal manipulation sequence is generated.

[0045] In some embodiments, the negative space analysis of the optimal manipulation sequence identifies a set of forbidden execution actions, including: constructing a full action space matrix corresponding to the optimal manipulation sequence; excluding the optimal manipulation sequence from the full action space matrix to obtain a negative space region; analyzing the negative space region to generate a dangerous action label; and obtaining a set of forbidden execution actions based on the dangerous action label.

[0046] A full action space matrix corresponding to the optimal manipulation sequence is constructed, and a complete set of reachable actions is determined based on the motion range of the optimal manipulation sequence. The optimal manipulation sequence q_opt(t) defines the standard trajectory for task execution, and the full action space needs to include all possible actions that deviate from this trajectory. Matrix construction is centered on the optimal sequence and expands outward to the physical limits of the robot. Discretization of the space samples possible configurations within the neighborhood of each time instant t_k along the optimal trajectory, forming the matrix A[i,j,k]. The sampling density is adjusted according to the local characteristics of the optimal sequence, with increased sampling points in areas with large curvature of the trajectory. Reachability determination considers not only single-point reachability but also whether it can return after deviating from the optimal sequence. The boundary is determined by considering joint limits, collision constraints, and task space limitations. The time dimension of the matrix corresponds to the discrete points of the optimal sequence, ensuring time consistency. Sparse representation only stores regions associated with the optimal sequence, improving storage efficiency. Through space expansion and constraint checking based on the optimal manipulation sequence, the full action space matrix is constructed.

[0047] The negative space region is obtained by excluding the optimal manipulation sequence from the full action space matrix. The full action space matrix A contains all possible actions, and the optimal manipulation sequence q opt(t) occupies one trajectory in it. The remaining negative space needs to be identified. The exclusion operation not only removes the optimal trajectory itself, but also considers its safety envelope. The safety envelope is dynamically adjusted according to the speed v opt and acceleration a opt of the optimal sequence, and a larger safety margin is needed for high-speed segments. The trajectory inflation adopts a variable radius method, r(t) = r base + k v|v opt(t)| + k a|a opt(t)|, to ensure that the safety distance changes with the motion state. The inflated trajectory region O expand is subtracted from the matrix A to obtain the negative space N = A - O expand. The structural analysis of the negative space identifies the topological relationship with the optimal sequence, including completely separated regions, adjacent regions, and surrounded regions. Connectivity analysis determines whether the actions in the negative space can reach each other without passing through the optimal trajectory. The volume ratio calculation V N / V A quantifies the occupation degree of the optimal sequence to the action space. For example, in the task of lifting up, the optimal sequence and its safety envelope occupy about 30% of the full action space, and the negative space is mainly distributed in the fast lateral movement and large amplitude swing regions.

[0048] Risk analysis is performed on the negative space region to evaluate the danger of actions in it based on its characteristics. The negative space region N contains all actions that deviate from the optimal sequence, and each of them needs to be evaluated for risk level. Risk evaluation first analyzes the deviation of actions from the optimal sequence, and the larger the deviation, the higher the risk. Collision risk is evaluated by calculating the swept volume of actions in the negative space, and the larger the volume, the higher the collision probability. Speed risk checks whether the speed of negative space actions exceeds the safety threshold v safe. Torque risk analyzes whether the joint torque required to perform negative space actions is close to the limit. Stability risk is determined by checking whether the center of gravity trajectory exceeds the support range. Combined risk considers the superposition effect of multiple risk factors, and uses fuzzy logic to handle uncertainty. The risk level is divided into three levels of high, medium, and low according to the comprehensive score. The risk distribution map shows that the high-risk region in the negative space is mainly concentrated near the motion limit. For example, it is found that fast swing actions (located at the far end of the negative space) have high collision risk, while actions that deviate slightly from the optimal trajectory have lower risk. Through multi-dimensional risk evaluation of the negative space region, dangerous action markers are generated.

[0049] The forbidden action set is obtained based on the dangerous action label, and the action labeled as high risk is included in the forbidden list. The dangerous action label assigns a risk level to each action in the negative space, and the forbidden set extracts the part exceeding the safety threshold. The set construction rule classifies according to the risk level: high-risk actions are directly listed as absolute prohibitions, medium-risk actions are listed as conditional prohibitions, and low-risk actions are only prompted. The absolute prohibition class implements hard limits through the controller bottom layer to ensure that it will not be executed under any circumstances. The conditional prohibition class decides whether to allow execution according to the specific scene and safety measures. The parameterized description of forbidden actions includes joint configuration range, speed limit, acceleration limit, etc. The compact representation of the set removes redundancy, and similar dangerous actions are merged into a class. The real-time update mechanism dynamically adjusts the forbidden boundary according to the task execution feedback. The index structure supports fast query to determine whether a given action is in the forbidden set. For example, the forbidden set formed in the elderly care scene includes dangerous action categories such as fast arm swinging, sudden acceleration, and large-angle tilting. Through the classification extraction and parameterized representation based on the risk label, the forbidden action set is established.

[0050] In step S130, a paradox verification scene is constructed by combining the optimal operation sequence and the forbidden action set, and the operation boundary parameters are generated by performing contradiction conflict test on the paradox verification scene.

[0051] In some embodiments, the combination of the optimal operation sequence and the forbidden action set to construct the paradox verification scene includes: performing intersection analysis on the optimal operation sequence and the forbidden action set to obtain an intersection analysis result; identifying a potential conflict node based on the intersection analysis result; performing contradiction amplification processing on the potential conflict node to generate a conflict amplification factor; and constructing a paradox verification scene using the conflict amplification factor.

[0052] The optimal maneuver sequence is intersected with the forbidden action set to identify overlapping and adjacent regions in the space-time domain. The intersection analysis employs set operation I = Q_opt∩F_forbid, where Q_opt is the extended domain of the optimal sequence and F_forbid is the parameter space of the forbidden set. The direct intersection is usually empty (a well-designed controller will not plan the optimal path in the forbidden region), so neighborhood expansion analysis is needed. The distance metric uses a weighted norm d = √(w_p||Δp||²+w_θ||Δθ||²+w_v||Δv||²), where w_p, w_θ, w_v are the weight coefficients of position, angle, and velocity, respectively. The neighborhood radius is determined according to the control accuracy and disturbance level, with a typical value of 1.5-2 times the optimal trajectory envelope. Time sequence correlation analysis checks the proximity of the optimal sequence to forbidden actions at different times, generating a distance-time curve d(t). Intersection metric indicators include minimum distance d_min, average distance d_avg, and risk exposure time T_risk. The boundary identification algorithm extracts the common boundary of the optimal sequence and the forbidden region, which is the most likely to cause conflicts. Intersection visualization maps the results to the configuration space, displaying the conflict intensity distribution in a heat map. For example, the analysis found that the elbow joint angle was only 8° away from the forbidden region boundary at t = 3.5s, constituting a potential risk point. Through multidimensional set operation and distance analysis, detailed intersection analysis results are obtained.

[0053] Based on the intersection analysis results, potential conflict nodes are identified, and the key positions and times that are most likely to cause conflicts are located. The intersection analysis results provide the distance distribution d(t) and boundary information of the optimal sequence and the forbidden region, and the conflict node identification is based on risk assessment. The node is defined as N_c = {(q_i, t_i) | d(q_i, F_forbid) < d_threshold}, where d_threshold is the risk distance threshold. The risk score considers both static proximity and dynamic trend: R_node = α / d + β|dd / dt|, when the trajectory rapidly approaches the forbidden region, the risk increases sharply. Key node screening retains a few points with the highest risk scores, usually selecting local maximum points. The influence range of the node is determined through sensitivity analysis, calculating the impact of the disturbance at this point on the overall trajectory. Time clustering combines adjacent conflict nodes to avoid over-subdivision. Node type classification includes: proximity type (small static distance), crossing type (dynamic trend pointing to the forbidden zone), and oscillation type (repeatedly near the boundary). Causal chain analysis traces the causes of each conflict node, whether it is caused by task demand or constraint limitation. For example, three main conflict nodes are identified: the speed near the upper limit in the rapid ascent segment, the angle near the singular configuration during turning, and the high force control accuracy requirement in the contact phase. Through risk scoring and multidimensional analysis, potential conflict nodes are accurately identified.

[0054] Contradiction intensification is applied to potential conflict nodes to artificially amplify conflict characteristics and fully expose the weaknesses of control schemes. Conflict nodes N_c contain the most problematic locations, and contradiction intensification transforms potential conflicts into explicit contradictions by adjusting parameters. Intensification strategies include constraint tightening and demand amplification: constraint tightening shrinks the forbidden boundary inward by Δd, and demand amplification increases the expected performance by Δp. Conflict amplification factors λ_i are designed for each node individually, reflecting the conflict sensitivity of the node. The calculation formula is λ_i = 1 + k_risk × R_node_i, where k_risk is the risk coefficient, and high-risk nodes obtain larger amplification factors. Parameter perturbation adopts a normal distribution N(μ, σ²), with mean μ = λ_i × p_nominal and standard deviation σ reflecting uncertainty. Time dimension intensification is achieved by compressing execution time, requiring the same action to be completed in a shorter time. Spatial dimension intensification reduces available space, simulating operation in a narrow environment. Combined intensification adjusts multiple parameters simultaneously, creating compound contradictions. Intensification boundaries ensure that physically impossible scenarios are not produced, maintaining the practical significance of testing. For example, for speed-critical nodes, the maximum speed limit is reduced by 20% (λ = 0.8), while maintaining the original execution time, creating a speed-time contradiction. Through systematic parameter adjustment and conflict amplification, effective conflict amplification factors are generated.

[0055] Contradiction intensification is applied to potential conflict nodes to artificially amplify conflict characteristics and fully expose the weaknesses of control schemes. Conflict nodes N_c contain the most problematic locations, and contradiction intensification transforms potential conflicts into explicit contradictions by adjusting parameters. Intensification strategies include constraint tightening and demand amplification: constraint tightening shrinks the forbidden boundary inward by Δd, and demand amplification increases the expected performance by Δp. Conflict amplification factors λ_i are designed for each node individually, reflecting the conflict sensitivity of the node. The calculation formula is λ_i = 1 + k_risk × R_node_i, where k_risk is the risk coefficient, and high-risk nodes obtain larger amplification factors. Parameter perturbation adopts a normal distribution N(μ, σ²), with mean μ = λ_i × p_nominal and standard deviation σ reflecting uncertainty. Time dimension intensification is achieved by compressing execution time, requiring the same action to be completed in a shorter time. Spatial dimension intensification reduces available space, simulating operation in a narrow environment. Combined intensification adjusts multiple parameters simultaneously, creating compound contradictions. Intensification boundaries ensure that physically impossible scenarios are not produced, maintaining the practical significance of testing. For example, for speed-critical nodes, the maximum speed limit is reduced by 20% (λ = 0.8), while maintaining the original execution time, creating a speed-time contradiction. Through systematic parameter adjustment and conflict amplification, effective conflict amplification factors are generated.

[0056] In some embodiments, the generating the manipulation boundary parameter by the paradox verification scene includes: simulating execution of the contradictory instruction in the paradox verification scene; performing process analysis on the contradictory instruction to obtain conflict response data; performing extreme value analysis based on the conflict response data to determine a bearing threshold; and obtaining the manipulation boundary parameter according to the bearing threshold.

[0057] In the paradox verification scene, the execution of the contradictory instruction is simulated to test the actual performance of the robot under the conflict condition. The paradox verification scene provides reinforced constraints and requirements, and the generation of the contradictory instruction needs to find a feasible solution under these conflicting conditions. The instruction generation adopts a multi-objective optimization method, and the objective function includes performance indicators and constraint violation degrees: J = w_perf × P + w_viol × V, where P is the performance score, and V is the constraint violation penalty. The optimization algorithm uses sequential quadratic programming (SQP) to handle nonlinear constraints. When it is impossible to find a solution that fully satisfies all conditions, the algorithm returns a suboptimal solution with the minimum constraint violation. The execution monitoring records the time series of key variables: position error e_p(t), force overrun f_over(t), velocity violation v_viol(t), etc. Conflict trigger detection identifies the exact time and position when the constraint is violated. Multiple executions use the Monte Carlo method to introduce random disturbances to test robustness. Execution strategies include: forced execution (ignore part of the constraints), adaptive execution (dynamically adjust parameters), segmented execution (use different strategies in different stages). For example, in a narrow space scenario, the simulation execution shows that the collision warning is triggered for the first time at t = 2.3s, and the velocity constraint is violated at t = 3.7s, forcing the robot to slow down, resulting in task timeout. Through the execution of the contradictory instruction and the whole process monitoring, the behavior data of the robot under the conflict condition is obtained.

[0058] The execution process of the conflicting instructions is analyzed to extract the response characteristics of the robot in the conflict state. The timing data generated by the execution process contains rich conflict response information, and key patterns can be identified through signal processing and feature extraction. The conflict recognition algorithm detects constraint violation events, and the frequency, amplitude, and duration of the violation are counted. The response delay analysis reflects the reaction speed of the controller by the time difference Δt_response from the conflict occurrence to the controller adjustment. Oscillation detection identifies high-frequency components in the control signal through spectral analysis, and excessive oscillation indicates that the control is close to instability. The performance degradation curve P(t) shows the downward trend of the execution performance as the conflict intensifies. The constraint priority is inferred by observing which constraints are relaxed first, and the inherent priority setting of the control strategy is inferred. The failure mode classification categorizes the observed problems: graceful degradation (performance degradation but controllable), oscillation divergence (control instability), and hard failure (complete inability to execute). Stress concentration analysis identifies the weakest link in the control scheme, which is usually the first variable to reach the limit. The recovery capability evaluates the robot's ability to return to normal from the conflict state. For example, analysis found that position control prioritizes accuracy during conflicts, and automatically reduces speed; force control exhibits 0.5 Hz oscillation, indicating that it is close to the stability boundary; the control strategy tends to sacrifice time performance to meet safety constraints. Through multi-dimensional signal analysis and pattern recognition, detailed conflict response data is obtained.

[0059] Based on the conflict response data, extreme value analysis is performed to determine the maximum conflict intensity that the control scheme can withstand. Conflict response data reveals the performance of the robot under different conflict levels, and extreme value analysis needs to find the acceptable performance boundary. The performance index is defined as the comprehensive score of key parameters: S=Σw_i×(1-|p_i-p_target| / p_range), and the performance is considered unacceptable when S is below the threshold. The extreme value search uses the bisection method to find the critical point λ_critical in the conflict intensity interval [λ_min,λ_max]. The bearing capacity curve plots the limit values under different types of conflicts, forming a multi-dimensional bearing capacity boundary. Statistical analysis considers the distribution characteristics of the response data to determine the conservative bearing threshold with a 95% confidence interval. The failure probability model P_fail(λ)=1 / (1+exp(-k(λ-λ_critical))) describes the relationship between conflict intensity and failure probability. Safety margin settings ensure that the control still has some buffer space even near the extreme value. Sensitivity analysis identifies the parameters that have the greatest impact on the extreme value, which need to be monitored. Time-varying extreme values consider the effects of actuator fatigue, and the bearing capacity will decrease after a long time of operation. For example, analysis determines that the bearing threshold for position conflict is ±12% trajectory deviation, and exceeding this value will cause a sharp increase in position error; the limit for speed conflict is 75% of the nominal value, and lower values will result in task failure; the oscillation amplitude of force control should not exceed 20% of the expected value. Through systematic extreme value search and statistical analysis, the bearing threshold of each dimension is determined.

[0060] The manipulation boundary parameters are obtained based on the bearing threshold, which defines a safe and feasible operation range. The bearing threshold provides the limit capability of the robot under various conflicts, and based on this, a reasonable safety boundary is set to obtain the manipulation boundary parameters. The boundary parameters are defined as B = {b_pos, b_vel, b_force, b_time}, which correspond to the limits of position, velocity, force, and time dimensions, respectively. The selection of the safety factor k_safe balances conservatism and availability, and the typical value is 0.7-0.8, i.e., the boundary is set to 70-80% of the bearing threshold. The coupling relationship between parameters is determined through correlation analysis, and some parameters need to be adjusted in conjunction. The hard and soft attributes of the boundary are distinguished: the hard boundary is absolutely not allowed to be exceeded, and the soft boundary allows short-term exceeding. The dynamic boundary is adjusted according to the current state, such as automatically tightening other boundaries when approaching a certain limit is detected. The boundary visualization draws a safety envelope in the operation space, which intuitively displays the feasible region. The verification test ensures that the set boundary can indeed avoid the problems observed in the conflict scenario. The structured representation of the boundary parameters supports fast query and real-time judgment. For example, the final determined manipulation boundary parameters: position boundary b_pos = ±8 cm (based on a 12% bearing threshold), velocity boundary b_vel = [0.1, 0.8] m / s, force boundary b_force = [10, 120] N, and time margin b_time = 1.2 × t_nominal. Through conservative design based on the bearing threshold and multi-dimensional parameter definition, the complete manipulation boundary parameters are finally generated.

[0061] In step S140, an interlocking verification condition is established based on the manipulation boundary parameters and the deviation tolerance threshold, and a safe unlocking sequence is generated based on the interlocking verification condition.

[0062] In some embodiments, the establishment of the interlocking verification condition based on the manipulation boundary parameters and the deviation tolerance threshold includes: performing correlation analysis on the manipulation boundary parameters and the deviation tolerance threshold to obtain a correlation analysis result; constructing a multi-dimensional verification matrix based on the correlation analysis result; performing constraint condition screening on the multi-dimensional verification matrix to obtain key constraint factors; and establishing an interlocking verification condition according to the key constraint factors.

[0063] The correlation analysis between the control boundary parameters and the deviation tolerance threshold is carried out to reveal the intrinsic connection between the two sets of parameter systems. Although the control boundary parameters B and the deviation tolerance threshold T have different sources, they have mutual influence in the control process. The correlation calculation adopts the mutual information method I(B,T)=ΣΣp(b,t)log(p(b,t) / (p(b)p(t))) to quantify the degree of information dependence between the parameters. Linear correlation analysis calculates the Pearson coefficient r_ij=cov(B_i,T_j) / (σ_Bi×σ_Tj) to identify direct linear relationships. Nonlinear correlation is mapped to a high-dimensional space through a kernel function and then analyzed to capture complex dependency patterns. Time-delay correlation considers the lag effect of parameter influence. Changes in certain boundary parameters will affect the effectiveness of the deviation threshold after a delay of τ. Cross-impact analysis identifies the mutual constraints between parameters. For example, tightening the position boundary will increase the sensitivity to angle deviation. The correlation strength grading divides the parameter pairs into strong correlation (r>0.7), medium correlation (0.3 <r≤0.7)、弱关联(r≤0.3)。动态关联追踪参数关系随执行进程的变化。关联图谱可视化展示参数网络,节点大小表示重要性,边粗细表示关联强度。例如分析发现:位置边界b_pos与位置偏差阈值T_pos呈强正相关(r=0.85),速度边界b_vel与时间偏差阈值T_time呈中等负相关(r=-0.52)。通过多种关联分析方法的综合应用,获取了详细的关联度分析结果。

[0064] Based on the correlation analysis results, a multi-dimensional verification matrix is constructed to form a structured verification framework. The correlation analysis results provide the relationship strength r_ij and influence mode between parameters, and the verification matrix organizes these relationships into an operable form. The matrix dimensions include boundary parameter dimensions (4 dimensions), bias threshold dimensions (4 dimensions), and verification type dimensions (3 categories: range check, trend check, and combination check). The matrix element V[i,j,k] defines the verification rule of the ith boundary parameter and the jth bias threshold under the kth verification type. The rule generation is determined according to the correlation strength, with strong correlation parameters using strict joint verification and weak correlation using independent verification. The verification function design considers the correlation type, with positive correlation using the same direction constraint and negative correlation using reverse compensation. The weight allocation is based on the correlation size, w_ij = |r_ij| / Σ|r_ij|, to ensure that important correlations receive more attention. Sparse processing removes verification items with a correlation degree below the threshold (|r| < 0.1) to improve computational efficiency. The block structure combines parameters with high correlation into a verification block, with joint verification within the block and independent processing between blocks. The time-varying characteristic is reflected by introducing the time dimension t, V[i,j,k,t], which allows the verification rule to evolve over time. For example, in the constructed verification matrix, the position-force coupling block contains 4 strong correlation verification rules, the position-velocity coupling block contains 3 strong correlation verification rules, and the velocity-time block contains 3 medium correlation rules. Through the structured organization of correlation relationships and rule generation, a multi-dimensional verification matrix is constructed.

[0065] The multi-dimensional verification matrix is subjected to constraint condition screening to identify the key factors that have the greatest impact on safety. The multi-dimensional verification matrix V contains a large number of verification rules, which need to be extracted by screening the most critical constraint conditions. The importance evaluation uses sensitivity analysis to calculate the change AS of the safety index after deleting a certain verification rule. Redundancy detection identifies verification rules with repeated functions, and retains the most stringent one. Constraint propagation analysis verifies the logical implication relationship between verification rules, which can be deleted if implied by other rules. Principal component analysis reduces the high-dimensional verification space, with the first 3-4 principal components usually containing more than 80% of the verification information. Key path identification identifies the verification nodes that must be passed from the initial state to the dangerous state, which correspond to the key constraints. Failure mode and effects analysis (FMEA) evaluates the severity of the consequences of each constraint failure. Statistical analysis of historical data on the triggering frequency and interception effect of each constraint. The screening criteria consider importance, uniqueness, and effectiveness, with a score S = a × Importance + β × Uniqueness + γ × Effectiveness. The threshold setting retains the top 70% of constraints with the highest scores as key constraint factors. For example, the screening results show that the position-velocity coupling constraint, the force-time gradual change constraint, and the angle-position coordination constraint are the three most critical constraint factors. Through multi-dimensional analysis and scientific screening methods, key constraint factors are obtained.

[0066] According to the key constraint factors, interlocking verification conditions are established to form an executable safety guarantee mechanism. Key constraint factors represent the most important safety requirements, and interlocking verification conditions need to convert these factors into specific judgment logic. Condition expression adopts Boolean logic form, and each key constraint factor corresponds to one or more logic expressions. Basic conditions include range check (x_min≤x≤x_max), rate limit (|dx / dt|≤v_max), and combination constraint (f(x,y)≤threshold). Compound conditions are connected by logical operators: C_total=C1AND(C2ORC3)ANDNOTC4. Timing logic handles the sequence of conditions, using temporal operators to represent time relationships such as "before", "after", and "during". Trigger mechanism defines the timing of condition checking, including periodic trigger, event trigger, and conditional trigger. Priority setting ensures that key conditions are checked first, and general conditions are processed later. Interlocking depth controls the strictness of conditions, which can be dynamically adjusted according to risk levels. Failure handling specifies the response actions when conditions are not met, such as limiting functions, running down, or emergency stop. The monitorability design of conditions ensures that all conditions can be obtained through sensors or calculations. For example, the core interlocking condition established is: IF(|p_actual-p_target|>0.8×T_pos)AND(v_actual>0.5×b_vel)THENlock_high_speed_motion. Through the design and execution mechanism definition of logic expressions, a complete interlocking verification condition is established.

[0067] Based on the established interlocking verification conditions, a safe unlocking sequence is generated, and a progressive permission release mechanism is designed. The interlocking verification conditions define various locking states and unlocking conditions, and the safe unlocking sequence needs to determine a reasonable unlocking order and timing. The unlocking sequence is represented by a state machine model, where each state corresponds to a set of active locking conditions, and state transition represents the unlocking operation. The sequence generation follows the principle of least privilege, and only unlocks the corresponding function when necessary. The unlocking priority is determined according to the task demand and safety level, and the basic movement function is unlocked first, and the high-risk operation is released last. The timing constraints ensure that the unlocking operation will not be too fast, and a minimum interval Δt_unlock is set between adjacent unlockings. The condition monitoring continuously checks the execution effect after unlocking, and immediately falls back to the safe state in case of abnormality. The unlocking token mechanism prevents unauthorized unlocking operations, and each unlocking step needs to verify a specific combination of conditions. The progressive test performs small amplitude actions after each unlocking to verify the normal response of the controller. The integrity check of the sequence ensures that all necessary functions can be unlocked eventually, while maintaining safety constraints. For example, a typical unlocking sequence: initially lock all degrees of freedom → unlock position monitoring after verifying that the sensor is normal → unlock slow movement after confirming that the environment is safe → unlock force control after establishing stable contact → unlock full speed movement after all conditions are met. By designing a state machine and a progressive unlocking strategy, a safe unlocking sequence is generated.

[0068] In step S150, the optimal control sequence is monitored in parallel based on the safe unlocking sequence to obtain execution verification data, and a target execution branch is determined according to the execution verification data, and a unified instruction stream is generated based on the target execution branch.

[0069] In some embodiments, the parallel monitoring of the optimal control sequence based on the safe unlocking sequence to obtain execution verification data includes: decomposing the safe unlocking sequence into multiple monitoring checkpoints; real-time state sampling of the optimal control sequence at the monitoring checkpoints to obtain sampling state data; analyzing the sampling state data to generate a state change trend; and generating execution verification data based on the state change trend.

[0070] The safety unlocking sequence is decomposed into multiple monitoring checkpoints, and state observation positions are set at key nodes of the unlocking process. The safety unlocking sequence contains the state transition sequence S_unlock = {s_0, s_1,..., s_n} from complete locking to full functional release, and each state transition point represents a change in control authority. The checkpoint setting adopts a risk-driven strategy, with dense arrangement before and after high-risk unlocking operations and appropriate sparseness in low-risk stages. Checkpoint types include mandatory checkpoints (which must be passed before each unlocking), conditional checkpoints (triggered by specific conditions), and periodic checkpoints (fixed time interval). Time stamp markers T_check = {t_0, t_1,..., t_m} record the expected trigger time of each checkpoint, where t_i+1-t_i ≥ Δt_min ensures that the check interval is not too dense. Check content definition C_i = {params_i, conditions_i, actions_i} includes the parameter set to be monitored, the judgment condition and the response action. Priority allocation ensures that critical checkpoints are executed first, P_check ∈ {HIGH, MEDIUM, LOW}. The spatial distribution of the checkpoints takes into account the trajectory characteristics of the optimal control sequence, with increased check density in locations with large trajectory curvature and rapid speed changes. Dynamic adjustment mechanism allows the addition or deletion of checkpoints according to real-time state. For example, in the unlocking sequence of the auxiliary standing task, 8 checkpoints are set: initial state confirmation point, sensor unlocking verification point, slow movement test point, contact establishment confirmation point, force control unlocking verification point, coordinated motion test point, full-speed unlocking verification point, and task completion confirmation point. Through scientific checkpoint arrangement and parameter design, the monitoring decomposition of the safety unlocking sequence is completed.

[0071] Real-time state sampling is performed at the monitoring checkpoints to capture the instantaneous characteristics of the execution process. The optimal manipulation sequence q_opt(t) defines the ideal execution trajectory, and the actual state q_actual(t_i) needs to be collected at each checkpoint t_i for comparison. The sampling content includes multi-dimensional information such as joint configuration vector q, Cartesian position p, velocity v, acceleration a, contact force f, etc. The sampling synchronization mechanism ensures the time consistency of all sensor data, using a combination of hardware triggering and timestamp correction. High-frequency sampling is extended by Δt_window time window before and after the checkpoint, with a sampling rate f_s=1kHz to obtain the detailed characteristics of the dynamic process. Data preprocessing includes outlier rejection, noise filtering and unit normalization to ensure data quality. State integrity check verifies the effectiveness of the sampling data, and missing rate exceeding 5% triggers resampling. Deviation calculation e_i=q_actual(t_i)-q_opt(t_i) quantifies the difference between actual and ideal. Multi-sensor fusion uses weighted least squares method to integrate multi-source information such as encoders, IMUs, force sensors, etc. The cache mechanism saves 100ms of data before and after the checkpoint to support post-analysis. The sampling trigger conditions include time trigger (arrival t_i), event trigger (unlock completion), and state trigger (deviation exceeds limit). For example, at the contact establishment checkpoint, the sampling data shows: position p_actual=[0.52,0.18,0.35]m, with a deviation of 2.3cm from the expected position; contact force f_actual=45N, within the expected range [40,60]N; joint speed is less than 0.3rad / s, meeting the safety requirements. Through high-precision real-time sampling and multi-dimensional state capture, the sampling state data is obtained.

[0072] In-depth analysis of the sampled state data to identify the state evolution law in the execution process. The discrete state points {q(t_0), q(t_1),..., q(t_m)} obtained by sampling need to be extracted through time series analysis to extract the trend. Trend fitting uses the least squares polynomial: q_trend(t) = a_0 + a_1t + a_2t², the coefficient a is obtained by regression of the last N sampling points. The rate of change dq / dt and d²q / dt² reflects the dynamic characteristics of the state, the first derivative represents the speed trend, and the second derivative represents the acceleration trend. Spectral analysis identifies periodic oscillations through FFT transformation, and the dominant frequency f_dominant>5Hz indicates that there may be control instability. Phase space reconstruction embeds time series data in high-dimensional space, revealing attractor structure and chaotic characteristics. Trend prediction uses the ARIMA model to predict the state of the next k steps based on historical data: q(t+k) = φ_1q(t) + φ_2q(t-1) +... + ε(t). Abnormal pattern recognition includes mutation detection (|dq / dt|>threshold), drift detection (cumulative deviation continues to grow), and oscillation detection (frequent positive and negative alternation). Trend classification classifies the identified patterns: convergent (deviation gradually decreases), divergent (deviation continues to increase), stable (deviation remains constant), and oscillatory (deviation changes periodically). Statistical feature extraction includes mean, variance, skewness, kurtosis, etc., to fully describe the state distribution characteristics. For example, analysis found that: the position deviation shows a convergent trend, the convergence time constant τ=2.5s; the speed has a small amplitude oscillation of 0.8Hz, with an amplitude of 0.05rad / s; the force control is stable, with a variance σ²=2.3N². Through the comprehensive application of various analysis methods, detailed state change trends are generated.

[0073] Execution verification data is generated based on state change trends. State change trends reveal the dynamic characteristics of the execution process, which need to be translated into verification data that can be used for decision-making. The verification data structure V_data = {V_safety, V_performance, V_stability, V_convergence} corresponds to safety, performance, stability, and convergence verification respectively. Safety verification V_safety checks whether all states are within the safety boundary: s_safe = min(1, d_boundary / d_actual), where d_boundary is the distance to the dangerous region. Performance verification V_performance evaluates the execution quality: p_score = 1 - ∑w_i|e_i| / e_max, integrating multi-dimensional deviations such as position, velocity, and force. Stability verification V_stability is based on Lyapunov theory: if V(x) = x^TPx is decreasing and dV / dt < 0, then it is locally stable. Convergence verification V_convergence judges whether it tends to the target: by fitting the deviation curve e(t) = e_0exp(-t / τ) to estimate the convergence time. Confidence calculation is based on the number of samples and data consistency: conf = 1 - σ_data / μ_data × √(n_min / n_actual). Timeliness markers ensure the freshness of verification data, and data that exceeds the validity period T_valid needs to be updated. Verification report generation includes numerical results, trend charts, risk prompts, and suggested measures. Abnormality markers mark indicators that exceed the normal range and analyze the reasons. For example, the generated verification data shows: safety 0.92 (good), performance 0.78 (to be improved), stability 0.95 (excellent), convergence 0.81 (expected to converge within 3.2 seconds), and overall confidence 0.87. Through systematic data organization and multi-dimensional verification, complete execution verification data is generated.

[0074] According to the execution verification data, the target execution branch is determined, and the control strategy most suitable for the current state is selected. The execution verification data V_data provides a comprehensive evaluation of the current execution state, and the target branch selection needs to find the optimal match in the predefined policy set. The execution branch definition B={B_nominal, B_conservative, B_aggressive, B_recovery} corresponds to the standard execution, conservative execution, aggressive execution, and recovery execution respectively. The branch selection function f_select:V_data→B is based on the comprehensive score of the verification data. The decision rule uses fuzzy logic: IF safety IS high AND performance IS low THEN B_aggressive, which allows performance to be improved under the premise of safety. Priority judgment ensures that safety always takes precedence: when V_safety<0.7, B_conservative is selected. The state machine model defines the transition conditions between branches to prevent instability caused by frequent switching. Each branch contains specific parameter adjustment strategies: B_conservative reduces speed by 50% and increases redundant checks, and B_aggressive increases speed by 30% and reduces intermediate verification. Historical effect learning dynamically adjusts the selection tendency by recording the execution results of each branch. Multi-objective trade-off finds a balance point among safety, performance, and energy consumption. Through scientific decision-making mechanism and strategy matching, the target execution branch is determined.

[0075] Based on the target execution branch, a unified instruction stream is generated, and the branch strategy is converted into an executable control command sequence. The target execution branch B_selected determines the control strategy, which needs to be combined with the optimal operation sequence to generate specific instructions. The instruction stream generation adopts the strategy superposition method: I_final=I_optimal+ΔI_branch, where I_optimal is the original optimal instruction, and ΔI_branch is the correction amount of the branch strategy. Parameter mapping converts the abstract strategy of the branch into specific numerical values: speed scaling factor k_v, position offset δ_p, force adjustment Δ_f, etc. Time alignment ensures the time continuity of the corrected instruction stream, and the possible gaps are filled by interpolation. The instruction format is unified as {timestamp, joint_angles, velocities, torques, flags}, which is convenient for the underlying controller to parse. Smooth processing avoids instruction mutation, and S-shaped curve transition is adopted: x(t)=x_0+(x_1-x_0)×(3t²-2t³). Integrity check verifies that the instruction stream covers the entire execution period without omission or repetition. Priority marking distinguishes between key instructions and auxiliary instructions to ensure that important commands are executed first. Buffer design allows a certain degree of execution flexibility, and non-critical instructions can be delayed or skipped. Real-time guarantee is achieved through timestamp and delay compensation mechanism. Through strategy mapping and instruction organization, a unified instruction stream is finally generated.

[0076] Step S160, the unified instruction stream is transmitted to the robot end through wireless communication, and the robot end generates a verification pass signal based on the security unlocking sequence. The verification pass signal is used to drive the robot to execute the actual execution state data.

[0077] Specifically, the unified instruction stream is transmitted to the robot end through wireless communication to ensure the reliable delivery of control instructions. The unified instruction stream I_unified contains structured information such as timestamps, control parameters, priority labels, etc., which need to be encoded and packaged to adapt to wireless transmission. The communication protocol uses the real-time control protocol RCP, uses UDP in the transport layer to ensure low latency, and adds sequence numbers and check mechanisms in the application layer to ensure reliability. The data frame format is [Header | Sequence | Timestamp | Priority | Payload | CRC], which ensures data integrity and timing accuracy. Transmission optimization uses differential encoding to reduce data volume, and only the changed part is transmitted for consecutive similar instructions. The priority queue mechanism ensures that critical instructions such as emergency stop are transmitted first. The wireless link selects the 5G millimeter wave frequency band, providing <5ms latency and >99.99% reliability. Channel adaptation dynamically adjusts the modulation and coding scheme according to RSSI and SNR. Redundant transmission uses spatial diversity to improve the reliability of high-priority instructions. Encryption uses AES-128 to protect data security, and the key is updated regularly. For example, transmitting a unified stream containing 500 instructions: the original 50KB data becomes 15KB after differential encoding and compression, and is divided into 15 1KB data frames. Critical instructions are marked with the highest priority and sent immediately, and normal instructions are transmitted in sequence. Through the 5G link, the transmission is completed in 1.5ms, and after the receiving end confirms that there is no error through CRC verification, it is sent to the verification module. Through the optimized protocol design and transmission strategy, efficient and reliable transmission of the unified instruction stream is realized.

[0078] The command verification is based on the safety unlocking sequence at the robot side, ensuring that the received command conforms to the current safety state. The safety unlocking sequence S_unlock defines a gradual release process of functions, containing states such as {s_lock, s_sense, s_slow, s_contact, s_force, s_full}, each corresponding to a specific permission set. The verification process first parses the command type, parameters, and execution requirements. The permission mapping table M_permission establishes the correspondence between the command and the permission, such as "full_speed" permission for high-speed motion and "force_control" permission for force control. The current permission P_current is obtained from the execution progress of the unlocking sequence. The verification algorithm checks each command: if Required_Permission(i) ⊆ P_current then Valid else Invalid. Timing verification ensures that unlocked functions are not executed prematurely. Parameter range checking verifies whether it is within the allowed range at the current level, such as v ≤ v_max(s_current). Integrity verification checks the dependency relationship between commands to avoid breaking the execution continuity due to permission restrictions. Conflict detection identifies command sequences that are safe individually but dangerous in combination. The verification result is divided into four categories: PASS (direct execution), MODIFY (adjusted execution), WAIT (wait for unlocking), and REJECT (reject). Intelligent correction automatically adjusts out-of-range parameters to a safe range. For example, in the s_contact state, a 150N force control command is received, and the current limit is 100N. The verification module automatically adjusts it to 100N and generates a warning, ensuring safety while maximizing user intent. Through multi-level permission verification and intelligent processing, a verification pass signal is generated.

[0079] The robot is driven based on the verification pass signal and the actual execution state data is obtained. The verification pass signal V_pass contains executable command queues and monitoring parameters, which are translated into low-level control commands by the actuator interface layer. The real-time controller runs at a frequency of 1kHz, and each cycle completes: sensor reading → state estimation → control calculation → command output → data recording. Multi-axis coordination is achieved through EtherCAT bus synchronization at the μs level. Execution monitoring collects multi-dimensional state data: joint position q is obtained through high-resolution encoders (accuracy 0.01°), velocity q -The contact force F is measured by a 6D force sensor (accuracy 0.1N) with differential plus filtering calculation, and the temperature T is monitored to prevent overheating. The state data format is [timestamp|joint_states|force|temperature|status], and high-efficiency binary encoding is used. The abnormality detection rules include position overrun, speed overrun, force overrun, temperature overload, etc., triggering a hierarchical protection response. The performance index real-time calculation includes trajectory tracking error, steady-state accuracy, response time, etc. The data buffer uses a ring buffer with a capacity of 10 seconds of historical data. For example, to perform a grabbing action: collect 1000 sets of position data in the approach phase for 1 second to display the smooth trajectory; detect a contact force of 28N at the contact moment to trigger mode switching; maintain a constant force of 30±2N during the grabbing phase with a position accuracy of ±0.5mm; the whole tracking error RMS is 1.8mm, the maximum error is 3.2mm, and there is no overrun alarm. Through comprehensive state monitoring and high-frequency data acquisition, detailed actual execution state data is finally obtained.

[0080] Step S170, based on the actual execution state data and the deviation tolerance threshold, a convergence analysis is performed to generate a convergence monitoring result, a causal feedback chain is obtained according to the convergence monitoring result, and the subsequent operation behavior is predictively corrected using the causal feedback chain to complete remote control.

[0081] Specifically, based on actual execution state data and deviation tolerance threshold, convergence analysis is performed to evaluate the stability and target approximation characteristics of the execution process. The actual execution state data D_actual contains multi-dimensional time series information such as position, velocity, force, etc., and the deviation tolerance threshold T defines the allowed deviation range in each dimension. The convergence analysis first calculates the real-time deviation e(t) = x_actual(t) - x_target, where x represents each state variable. The deviation evolution trend is obtained by exponential fitting e(t) = e_0 x exp(-t / τ), and the convergence time constant τ reflects the convergence speed. The convergence criterion is set as |e(t)| < T for a duration of more than t_stable, indicating that the system enters a stable convergence state. Multi-dimensional convergence evaluation analyzes the convergence characteristics of each degree of freedom respectively, and identifies the slowest bottleneck dimension. The convergence rate calculation r = -de / dt / e quantifies the instantaneous convergence speed. Oscillation detection determines whether there is limit cycle oscillation by zero-crossing frequency and energy distribution. Convergence domain estimation uses Lyapunov method, V(e) = e^TPe, if dV / dt < 0, then local convergence. Statistical property analysis calculates the convergence parameter distribution of different execution cycles. Abnormal pattern recognition includes divergence (τ < 0), oscillation (periodic), stagnation (r ≈ 0), etc. For example, analysis shows that the position convergence time constant τ_p = 2.5s, the force convergence τ_f = 1.8s, and the overall convergence to the threshold range within 4 seconds; but a 0.5Hz small amplitude oscillation is detected in the velocity dimension, which needs attention. Through multi-dimensional convergence analysis, detailed convergence monitoring results are generated.

[0082] In some embodiments, the causal feedback chain is obtained according to the convergence monitoring results, including: performing hierarchical causal relationship identification on the convergence monitoring results to obtain strong correlation factors and weak correlation factors; performing time series characteristic analysis based on the strong correlation factors and weak correlation factors, subdividing the strong correlation factors into fast strong correlation and slow strong correlation, and subdividing the weak correlation factors into fast weak correlation and slow weak correlation, to obtain four types of spatiotemporal correlation factors; constructing a multi-level feedback network with hierarchical speed based on the four types of spatiotemporal correlation factors, wherein the fast strong correlation constitutes a core control layer, the slow strong correlation constitutes a steady-state regulation layer, the fast weak correlation constitutes a disturbance compensation layer, and the slow weak correlation constitutes a trend prediction layer; performing path search with different priorities based on the multi-level feedback network to generate a causal feedback chain with hierarchical spatiotemporal characteristics.

[0083] The hierarchical causal relationship identification of the convergence monitoring results extracts different intensity causal correlation characteristics. The convergence monitoring results contain convergence parameters τ, oscillation characteristics, abnormal patterns and other information of each dimension. These manifestations hide complex causal driving mechanisms that need to be identified through systematic methods. Strongly associated factors represent factors that have a direct and significant impact on system performance, such as control parameters directly determining response speed and actuator characteristics directly affecting tracking accuracy. Changes in these factors will immediately reflect in system performance. Weakly associated factors have a smaller impact when acting alone, but can have a non-negligible effect under certain conditions or after long-term accumulation, such as environmental temperature slowly affecting device characteristics and measurement noise gradually accumulating to cause control deviation. The causal strength is determined by the path coefficient β = |ΔY / ΔX| × P(X→Y), where ΔY / ΔX represents the response sensitivity of Y to the change of X, and P(X→Y) represents the conditional probability of X leading to Y. β>0.7 is determined as strong correlation, and 0.3<β≤0.7 is determined as weak correlation. The core value of hierarchical identification is to distinguish between primary and secondary factors, allowing the control system to prioritize key factors while not ignoring secondary but persistent sources of influence. For example, in robot control, identifying that high control gain leading to speed oscillation is a strong correlation (β=0.85) requires immediate adjustment, and environmental temperature changes causing steady-state deviation are a weak correlation (β=0.42) that can be addressed through slow compensation. Through scientific hierarchical identification methods, strong and weak correlation factors are accurately obtained.

[0084] Temporal characteristic analysis based on strong and weak correlation factors refines the time scale of causal transmission. Strong correlation factors have different time scales. Some have almost instantaneous effects, such as control commands to motor responses, while others take time to manifest, such as load changes to steady-state error formation. The difference in time scale determines the control strategy: fast causal relationships require real-time response and fast compensation, while slow causal relationships can be handled through prediction and gradual adjustment. Time delay analysis is determined by the cross-correlation function R_xy(τ)=E[(X_t-μ_x)(Y_{t+τ}-μ_y)] / (σ_xσ_y), where τ is the time delay, μ_x and μ_y are the means of X and Y, and σ_x and σ_y are the standard deviations. The τ value corresponding to the function peak is the causal delay time. Fast response is defined as τ<0.5s, and slow response is defined as τ≥0.5s. This threshold is based on the sampling period and dynamic response requirements of the control system. Temporal subdivision of strong correlation factors helps identify which are transient problems that require high-frequency controllers and which are gradual problems that can be addressed through parameter adaptation. Temporal analysis of weak correlation factors is also important. Fast weak correlations such as high-frequency noise require filtering, while slow weak correlations such as performance degradation require trend prediction. Through temporal analysis, four types of spatiotemporal correlation factors are successfully obtained: fast strong correlation, slow strong correlation, fast weak correlation, and slow weak correlation.

[0085] A multi-level feedback network is constructed based on four types of spatiotemporal correlation factors, forming a functional control architecture. The core control layer is composed of fast strong correlation factors, which is responsible for the basic control function of the system and must have the fastest response speed and the highest reliability. Any delay or failure will directly affect the system performance. The steady-state regulation layer deals with slow strong correlation, although the response is slower, but the impact is far-reaching. This layer ensures the long-term stable operation of the system through continuous parameter optimization and compensation, avoiding performance drift. The disturbance compensation layer is specifically designed to deal with the interference caused by fast weak correlation. Although these disturbances have little impact individually, if not suppressed, they will be transmitted and amplified through the system, ultimately affecting control quality. The trend prediction layer focuses on slow weak correlation, predicting the evolution trend of the system through long-term data analysis, providing a basis for preventive maintenance and performance optimization. The inter-layer connection weight is defined as w_ij=β_ij×exp(-d_ij / d_0), where β_ij is the causal strength of node i to j, d_ij is the inter-layer distance (difference in layer number), and d_0 is the normalization constant (typical value is 1). This exponential decay form reflects the rapid weakening characteristics of cross-layer influence. Each layer has an independent update frequency: core layer 1 kHz, steady-state layer 10 Hz, disturbance layer 100 Hz, and prediction layer 1 Hz. This multi-rate design greatly improves the computational efficiency. Through hierarchical mapping, a multi-level feedback network is successfully constructed.

[0086] The path search is not a simple graph traversal, but a comprehensive optimization problem that needs to consider multiple factors such as causal strength, time delay, path length, etc. Priority setting ensures that the most influential dominant paths are found first, which usually pass through the core control layer and have strong causal relationships and fast transmission characteristics. The path importance score is S = G x exp(-T / T_0), where G = Πβ_i is the product of all causal strengths on the path (total gain of the path), T = Στ_i is the sum of all delays on the path, and T_0 is the time normalization constant (usually the system time constant). This score takes into account both causal strength and time impact. After identifying the dominant paths, the algorithm continues to search for secondary but still important auxiliary paths, which may pass through other layers and although the impact is smaller, they are indispensable for a complete understanding of the system. The preservation of timing information is crucial, as the delay time of each node is recorded, so that the final causal chain not only describes "what affects what", but also explains "how quickly the impact is produced". The integration of search results removes redundant paths, merges similar branches, and retains complete chains with unique contributions. For example, a typical dominant chain: control gain K_p→(τ=0.1s,β=0.9)→speed response→(τ=0.3s,β=0.8)→position tracking→(τ=0.5s,β=0.7)→convergence performance, total delay 0.9s, path gain 0.504. Through systematic prioritized search, complete causal feedback chains with spatiotemporal characteristics are generated.

[0087] A causal feedback chain is used to predictively correct subsequent control actions, achieving feedforward compensation and performance optimization. The causal feedback chain C = {c_1→c_2→...→c_n} reveals the transmission mechanism from control input to performance output, which can be used for prediction and improvement. The prediction model is constructed based on the chain: y(t+h) = Σα_i × u(t-d_i), where h is the prediction horizon, d_i is the causal delay, and α_i is the transfer coefficient. The correction strategy is designed based on model predictive control, making advance adjustments when undesirable trends are detected. Feedforward compensation applies a reverse correction u_ff = -β × e_predicted d times in advance based on the delay characteristics of the causal chain. Parameter optimization blocks undesirable causal transmission by adjusting the control parameters at the source of the chain. Online learning continuously updates the causal model parameters to adapt to system changes. The correction calculation considers the combined impact of multiple causal chains: Δu = Σw_i × Δu_i, with weights reflecting the importance of the chains. Stability constraints ensure that the correction does not introduce new problems, |Δu| <u_max×0.2。例如在辅助老人床椅转移任务中,因果链分析发现"髋关节驱动器响应延迟(0.2s)→身体倾斜角速度增大(0.5s)→重心偏移超限(1.0s)→可能跌倒风险"的传递路径,立即采取预测性修正:提前0.2s增大膝关节支撑力矩补偿髋关节延迟,同时降低整体移动速度20%防止重心快速偏移,修正后老人身体倾斜角度保持在±5°安全范围内,成功完成了从轮椅到床的平稳转移。通过基于因果反馈链条的预测分析和主动修正,能够提前识别潜在问题并采取针对性措施,实现了对后续操控行为的预测性修正,完成了人形机器人的高质量远程操控。

[0088] In order to implement the above method embodiment corresponding to the humanoid robot remote control method based on wireless communication, to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The following is a block diagram of a humanoid robot remote control device 200 based on wireless communication provided by an embodiment of the present application. For ease of explanation, only the parts related to this embodiment are shown. The humanoid robot remote control device 200 based on wireless communication provided by an embodiment of the present application includes:

[0089] The command parsing module 201 is configured to receive a remote control command from the operator, perform semantic analysis on the remote control command to generate control intention data, establish a two-way desired calibration channel based on the control intention data, and generate desired consistency parameters and deviation tolerance thresholds based on the two-way desired calibration channel;

[0090] The reverse reasoning module 202 is configured to obtain an expected execution result based on the expected consistency parameter, to reversely deduce an optimal operation sequence according to the expected execution result, and to perform negative space analysis on the optimal operation sequence to identify a set of prohibited execution actions;

[0091] The paradox verification module 203 is configured to construct a paradox verification scene by combining the optimal operation sequence and the set of prohibited execution actions, and to perform a contradiction conflict test through the paradox verification scene to generate an operation boundary parameter;

[0092] The safety control module 204 is configured to establish an interlocking verification condition based on the operation boundary parameter and the deviation tolerance threshold, and to generate a safety unlocking sequence based on the interlocking verification condition;

[0093] The execution monitoring module 205 is configured to perform parallel monitoring on the optimal operation sequence based on the safety unlocking sequence to obtain execution verification data, to determine a target execution branch according to the execution verification data, and to generate a unified instruction stream based on the target execution branch;

[0094] The communication execution module 206 is configured to transmit the unified instruction stream to a robot end through wireless communication, to perform instruction verification at the robot end based on the safety unlocking sequence to generate a verification pass signal, and to drive the robot to execute to obtain actual execution state data based on the verification pass signal.

[0095] The feedback optimization module 207 is configured to perform convergence analysis based on the actual execution state data and the deviation tolerance threshold to generate a convergence monitoring result, to obtain a causal feedback chain according to the convergence monitoring result, and to predictively correct a subsequent operation behavior by using the causal feedback chain to complete remote operation.

[0096] The above-described wireless communication-based remote operation device 200 for a humanoid robot can implement the wireless communication-based remote operation method for a humanoid robot of the above-described method embodiment. The optional items in the above-described method embodiment are also applicable to the present embodiment, and will not be described in detail herein. The remaining contents of the present embodiment can be referred to the contents of the above-described method embodiment, and will not be described in detail herein.

[0097] As shown in Figure 3 the third embodiment of the present application further provides a computer device, which comprises a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302, and characterized in that the processor 302 implements the steps of the wireless communication-based remote operation method for a humanoid robot according to the first embodiment of the present application when executing the program.

[0098] The above examples are intended to illustrate and deduce the technical solutions of the present application, and to completely describe the technical solutions, objects and effects of the present application. The purpose is to make the public more thoroughly and comprehensively understand the disclosed content of the present application, and does not limit the protection scope of the present application.

[0099] The above examples are not based on an exhaustive enumeration of the present application, and there can be many other unlisted embodiments. Any substitutions and improvements made without violating the concept of the present application are within the scope of protection of the present application.

Claims

1. A method for remotely controlling a humanoid robot based on wireless communication, characterized in that: include: receiving a remote control command from an operator, performing semantic analysis on the remote control command to generate control intention data, establishing a two-way expectation calibration channel based on the control intention data, and generating an expected consistency parameter and a deviation tolerance threshold based on the two-way expectation calibration channel; Obtaining an expected execution result based on the expected consistency parameter, inferring an optimal control sequence based on the expected execution result, and performing negative space analysis on the optimal control sequence to identify a prohibited action set; Combining the optimal control sequence and the prohibited action set to construct a paradox verification scenario, and performing a conflict test on the paradox verification scenario to generate control boundary parameters; establishing an interlock verification condition based on the manipulation boundary parameter and the deviation tolerance threshold, and generating a safe unlocking sequence based on the interlock verification condition; Parallel monitoring of the optimal control sequence based on the security unlocking sequence to obtain execution verification data, determining a target execution branch according to the execution verification data, and generating a unified instruction stream based on the target execution branch; The unified instruction stream is transmitted to the robot side via wireless communication, the robot side performs instruction verification based on the safety unlocking sequence to generate a verification pass signal, and the robot is driven to execute and obtain actual execution status data based on the verification pass signal; A convergence analysis is performed based on the actual execution state data and the deviation tolerance threshold to generate a convergence monitoring result, a causal feedback chain is obtained according to the convergence monitoring result, and the causal feedback chain is used to predictively correct subsequent control behaviors to complete remote control.

2. The method according to claim 1, characterized in that The reverse deducing of the optimal control sequence according to the expected execution result includes: Constructing a result status mapping table based on the expected execution result; Reverse-analyze the result state mapping table to obtain a causal association path; The optimal control sequence is generated by reverse deduction based on the causal relationship path.

3. The method according to claim 1, characterized in that The performing negative space analysis on the optimal control sequence to identify a prohibited action set includes: Constructing a full action space matrix corresponding to the optimal control sequence; Excluding the optimal manipulation sequence from the full action space matrix to obtain a negative space region; Analyzing the negative space area to generate a dangerous action mark; A prohibited action set is acquired based on the dangerous action mark.

4. The method according to claim 1, wherein The step of constructing a paradox verification scenario by combining the optimal manipulation sequence and the prohibited action set includes: Performing intersection analysis on the optimal control sequence and the prohibited action set to obtain an intersection analysis result; Identifying potential conflicting nodes based on the intersection analysis results; Performing conflict reinforcement processing on the potential conflict node to generate a conflict amplification factor; The conflict amplification factor is used to construct a paradox verification scenario.

5. The method according to claim 1, wherein The generating of the control boundary parameters by performing the conflict test through the paradox verification scenario includes: simulating the execution of contradictory instructions in the paradox verification scenario; Analyze the execution process of conflicting instructions and obtain conflict response data; Performing extreme value analysis based on the conflict response data to determine a tolerance threshold; A control boundary parameter is obtained according to the tolerance threshold.

6. The method according to claim 1, characterized in that The establishing of the interlock verification condition based on the manipulation boundary parameter and the deviation tolerance threshold comprises: Performing a correlation analysis on the control boundary parameter and the deviation tolerance threshold to obtain a correlation analysis result; Constructing a multidimensional verification matrix based on the association analysis results; Performing constraint screening on the multidimensional verification matrix to obtain key constraint factors; An interlock verification condition is established according to the key constraint factors.

7. The method according to claim 1, characterized in that The step of performing parallel monitoring on the optimal control sequence based on the safety unlocking sequence to obtain execution verification data includes: Decomposing the security unlocking sequence into a plurality of monitoring checkpoints; Performing real-time state sampling on the optimal control sequence at the monitoring checkpoint to obtain sampling state data; Analyzing the sampled state data to generate a state change trend; Execution verification data is generated based on the state change trend.

8. The method according to claim 1, characterized in that The obtaining of a causal feedback chain according to the convergence monitoring result includes: Performing hierarchical causal relationship identification on the convergence monitoring results to obtain strong correlation factors and weak correlation factors; Based on the strong correlation factor and the weak correlation factor, a time series characteristic analysis is performed to subdivide the strong correlation factor into fast strong correlation and slow strong correlation, and the weak correlation factor into fast weak correlation and slow weak correlation, thereby obtaining four types of spatiotemporal correlation factors; Based on the four types of spatiotemporal correlation factors, a multi-level feedback network with different layers and speeds is constructed, in which the fast strong correlation constitutes the core control layer, the slow strong correlation constitutes the steady-state regulation layer, the fast weak correlation constitutes the disturbance compensation layer, and the slow weak correlation constitutes the trend prediction layer; A prioritized path search is performed based on the multi-level feedback network to generate a causal feedback chain with hierarchical spatiotemporal characteristics.

9. A remote control device for a humanoid robot based on wireless communication, characterized in that: include: a command parsing module, configured to receive a remote control command from an operator, perform semantic parsing on the remote control command to generate control intention data, establish a two-way desired calibration channel based on the control intention data, and generate desired consistency parameters and a deviation tolerance threshold based on the two-way desired calibration channel; a reverse reasoning module, configured to obtain an expected execution result based on the expected consistency parameter, reversely infer an optimal control sequence based on the expected execution result, and perform negative space analysis on the optimal control sequence to identify a set of prohibited actions; A paradox verification module, configured to construct a paradox verification scenario by combining the optimal control sequence and the prohibited action set, and to generate control boundary parameters by performing a conflict test on the paradox verification scenario; a safety control module, configured to establish an interlock verification condition based on the manipulation boundary parameter and the deviation tolerance threshold, and generate a safety unlocking sequence based on the interlock verification condition; an execution monitoring module, configured to monitor the optimal control sequence in parallel based on the security unlocking sequence to obtain execution verification data, determine a target execution branch based on the execution verification data, and generate a unified instruction stream based on the target execution branch; a communication execution module, configured to transmit the unified instruction stream to the robot side via wireless communication, perform instruction verification on the robot side based on the safety unlock sequence to generate a verification pass signal, and drive the robot to execute and obtain actual execution status data based on the verification pass signal; A feedback optimization module is used to perform convergence analysis based on the actual execution state data and the deviation tolerance threshold to generate a convergence monitoring result, obtain a causal feedback chain based on the convergence monitoring result, and use the causal feedback chain to predictively correct subsequent control behaviors to complete remote control.

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

Images