Intelligent inspection operation robot

Abstract

The invention provides an intelligent inspection operation robot, which comprises a self-adaptive interaction assembly for collecting unstructured features in an inspection environment of the inspection operation robot in real time, converting continuous signals of the unstructured features into discrete interaction decision nodes, and generating an interaction parameter set capable of being iteratively optimized through a cross validation mechanism; the autonomous intention analysis component constructs a probabilistic decision-making program based on implicit features of an operator behavior track of the inspection operation robot, extracts a potential mode of an operation intention through nonlinear correlation analysis in combination with an interaction parameter set, and recognizes an interaction demand which is not clearly expressed; and the collaborative response component constructs a response decision mechanism with self-organizing characteristics, and jointly inputs an interaction parameter set and an interaction demand into the response decision mechanism by establishing a dynamic association rule between different interaction channels of the inspection operation robot and an operator to generate a composite interaction scheme. According to the method, the composite interaction scheme comprehensively considering multiple factors can be generated.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to an intelligent inspection operation robot. Background Art

[0002] The demand for automated inspection of various industrial facilities is increasing day by day. Traditional manual inspection methods have problems such as low efficiency, high cost, and many safety hazards. Especially in high-risk environments or harsh conditions, manual inspection not only poses great risks but also is difficult to ensure the inspection quality. Currently, industries such as power, petrochemical, and rail transit have strict requirements for the real-time monitoring and regular inspection of equipment status; the equipment in these industries often has a wide distribution range, complex structure, and harsh environment. Manual inspection not only consumes time and effort but also is prone to missed or false inspections due to fatigue or negligence. Therefore, the development of intelligent inspection robots that can replace humans has become an urgent need in the industry.

[0003] Prior Art One, Application No.: CN202411843614.6 discloses an intelligent obstacle avoidance method for an inspection operation robot in a converter station, including the following steps: Step 1, select an inspection operation robot that can automatically cruise and can be switched to manual control; Step 2, add an extendable electric push rod in the forward direction of the inspection operation robot, and install a triangular push frame on the output rod of the electric push rod. When the camera monitors an obstacle below a certain height in front, extend the electric push rod by a certain length, and use the triangular push frame to push the obstacle to move and push it to both sides; Step 3, add human body temperature sensors at eight angles on the inspection operation robot, and the human body temperature sensors are incorporated into the reference basis of the automatic cruise route. At the same time, the signals of the human body temperature sensors can be transmitted back to the monitoring room terminal. Although it relates to the field of robot inspection in a converter station, specifically an intelligent obstacle avoidance method for an inspection operation robot in a converter station; it only solves the physical obstacle avoidance problem of low-height obstacles and lacks the intelligent analysis ability for the unstructured features of the environment; it relies on preset mechanical obstacle avoidance actions and cannot dynamically generate an optimized interaction plan according to the environmental features.

[0004] Prior Art 2, Application No.: CN202410846517.6 discloses a PID shift system and its control method for a high-voltage chamber inspection operation robot, belonging to the field of high-voltage chamber inspection operation robots. The control method includes obtaining the expected angular velocity and the actual angular velocity; obtaining the tracking error, the error integral, and the error derivative; using a second-order non-affine nonlinear calculation method to obtain a mapping equation; using a self-coupled PID control method to calculate to obtain a first output signal; using an ultra-short feedback method to calculate to obtain a second output signal; and a brushless DC motor obtaining the first output signal and the second output signal to complete the adjustment. Although the angular velocity of the brushless DC motor is accurately adjusted and controlled by using the ultra-short feedback and self-coupled PID control means to ensure that the inspection operation robot can move at a constant speed when operating the trolley, to ensure that it does not tip over, thereby improving work efficiency and protecting the safety of personnel and equipment; however, it focuses on the motor control accuracy and does not involve the recognition of the operator's intention and the optimization of interaction; the motion control strategy is completely separated from the environmental perception, lacking a collaborative decision-making mechanism; and it is unable to dynamically adjust the control parameters according to the operator's behavior characteristics.

[0005] Prior Art 3, Application No.: CN202411717775.0 discloses a control method, device, equipment, and storage medium for a power plant inspection operation robot. The method includes: when detecting an abnormal gathering point in the target plant area, selecting a target robot for inspection; controlling the target robot to collect on-site audio and video information feedback by the on-site personnel during the interview; and sorting out the inspection abnormal information through the target language large model. Although it realizes that when the staff discovers an abnormality but fails to report it, it can automatically determine the location of the abnormal gathering of personnel, then select a robot for inspection, and finally analyze according to the on-site audio and video, so that the inspection abnormal information can be obtained, enabling timely discovery of abnormalities and controlling the inspection robot for inspection when not reported, improving the timeliness and flexibility of the inspection; however, the abnormal detection depends on the phenomenon of personnel gathering and lacks the ability to actively identify implicit requirements; the audio and video analysis adopts a fixed mode and does not construct an iteratively optimized interaction parameter system; the response mechanism is single and cannot generate a composite interaction plan.

[0006] Currently, Prior Art 1, Prior Art 2, and Prior Art 3 have the problem of separation between the environmental perception and decision-making of the inspection robot, and the limitation that the recognition of the operation intention only depends on explicit instructions. Therefore, the present invention provides an intelligent inspection operation robot. Summary of the Invention

[0007] To solve the above technical problems, the present invention provides an intelligent inspection operation robot, including:

[0008] A collaborative response component is used to build a response decision-making mechanism with self-organization characteristics. By establishing dynamic association rules between different interaction channels of the inspection operation robot and the operator, the interaction parameter set and the interaction requirements are jointly input into the response decision-making mechanism to generate a composite interaction plan for the inspection operation robot and the operator.

[0009] Optionally, an adaptive interaction component of the intelligent inspection operation robot is used to collect unstructured features in the inspection environment of the inspection operation robot in real time, convert the continuous signal of the unstructured features into discrete interaction decision nodes, and generate an iteratively optimizable interaction parameter set through a cross-validation mechanism.

[0010] Optionally, the adaptive interaction component includes:

[0011] A signal feature extraction module is used to capture the original continuous signal of the unstructured features including spatial form fluctuations and energy distribution gradients through the environmental perception unit, and cut the original continuous signal stream on the time axis into time series segments with overlapping features; each segment passes through a multi-dimensional feature extraction layer to separate the signal intensity, frequency domain distribution, and spatial coupling degree to form a weighted feature vector cluster;

[0012] A time map construction module is used to input the feature vector cluster into a discretization processing layer, and map the original continuous signal segment to discrete event nodes with clear semantics through a dynamic clustering algorithm in the feature space; the topological connection relationship between discrete event nodes is automatically constructed based on the spatio-temporal correlation of the signal segments to form a dynamically evolving event map;

[0013] An iterative optimization module is used to compare the feature vectors captured by different environmental perception units within the same time window to eliminate abnormal nodes caused by environmental noise; match the currently generated interaction decision nodes with the historical event map to correct the deviation of the node attribute parameters; the interaction decision nodes that pass the verification enter the parameter optimizer, and the weight distribution strategy of the interaction parameter set is dynamically adjusted according to the actual interaction effect feedback after the interaction decision nodes are triggered.

[0014] Optionally, the time map construction module includes:

[0015] A node activation condition matching sub-module is used to compare the matching degree between the feature vector extracted from the current environmental perception signal and the environmental state label of the node in real time based on the environmental state label of the discrete event node in the event map; when the multi-dimensional indicators such as the signal intensity and frequency domain distribution of the feature vector reach the preset spatial coupling threshold with the physical feature pattern of the environmental state label of the node, the candidate qualification of the node is triggered;

[0016] The interaction trigger threshold determination sub-module is used to evaluate the threshold of nodes with candidate qualifications. The built-in interaction trigger threshold is generated by training historical interaction effect data. When the composite evaluation value of the real-time environmental state fluctuation parameter and the operation requirement index exceeds the interaction trigger threshold, the node is activated as an effective interaction decision node.

[0017] The node association network collaborative decision sub-module is used to establish a logical operation relationship between the activated interaction decision nodes and adjacent interaction decision nodes through an association weight matrix to form a local decision network. The activated interaction decision nodes within the same time window exchange environmental state parameters through the weight matrix to generate a comprehensive environmental evaluation value. Trace back the response effect data of historical nodes along the time evolution path of the event graph to optimize the reliability of the current decision, and screen out a set of interaction decision nodes that meet the spatio-temporal consistency requirements through a multi-dimensional voting mechanism.

[0018] Optionally, the autonomous intention parsing component of the intelligent inspection operation robot is used to construct a probability decision program based on the implicit features of the operator's behavior trajectory of the inspection operation robot, extract the potential patterns of operation intentions through non-linear correlation analysis in combination with the interaction parameter set, and identify the unexpressed interaction requirements.

[0019] Optionally, the autonomous intention parsing component includes:

[0020] The implicit feature extraction module is used to extract the implicit features of time series dependence, operation preference, and abnormal operation patterns from the operator's behavior trajectory data recorded during the operation of the inspection operation robot to form a feature vector.

[0021] The probability decision program construction module is used to establish a probability model based on the data of the historical operator's behavior trajectory, calculate the occurrence probability of different operation modes and their conditional dependencies; combine the current interaction parameter set to dynamically adjust the probability weights to form a decision logic adapted to different scenarios.

[0022] The potential pattern extraction module is used to jointly analyze the interaction parameter set and the implicit features, and use non-linear mapping to mine the potential correlation between the interaction parameter set and the implicit features; identify the potential intentions of the operator under different environmental conditions, including unexpressed interaction requirements, through pattern matching and clustering analysis.

[0023] Optionally, the operator's behavior trajectory data of the implicit feature extraction module includes operation instruction sequences, operation frequencies, operation intervals, path selections, and equipment adjustment modes.

[0024] Optionally, the implicit feature extraction module includes:

[0025] The data structuring and processing sub-module is used to receive the original data stream of the operator's behavior trajectory data, including discrete operation instruction sequences and continuous timestamp records; divide the continuous original data stream into analyzable segments through time window segmentation, and each segment retains the complete operation context relationship;

[0026] The feature content conversion sub-module is used to perform time series dependency modeling, operation preference quantification, and abnormal pattern detection;

[0027] The feature vector synthesis sub-module is used to expand the three-dimensional time series tensor into a sparse vector, perform eigenvalue decomposition on the edge weight matrix of the behavior graph, and convert the mask of the abnormal label into a position encoding; the sparse vector, eigenvalue decomposition, and position encoding are orthogonally spliced to form a composite feature vector with a fixed dimension.

[0028] Optionally, in time series dependency modeling, the transition probability matrix between operation instructions of adjacent timestamps is analyzed to capture the mandatory order relationship and optional branches between operation steps; at the same time, the periodic pattern repeatability across time windows is calculated to identify the operator's habitual work rhythm, and a three-dimensional time series tensor containing the time series constraint intensity is output, where the third dimension encodes the dependency intensity at different time scales;

[0029] Operation preference quantification converts discrete behaviors such as the path selection frequency of operation preferences and the adjustment amplitude of device parameters in the original data stream into probability distributions to determine the degree of decision certainty of the operator; co-occurrence analysis is performed on high-frequency operation combinations to establish a behavior association graph, and the edge weights of the graph reflect the stability of the operation combination. The time series tensor and the behavior graph together constitute the intermediate feature representation;

[0030] Abnormal pattern detection defines a dynamic density threshold, and an abnormal label is triggered when the local density of the newly input operator behavior trajectory data is lower than the historical distribution threshold; the abnormal label includes instruction jumps that violate time series constraints and isolated operation nodes that deviate from the behavior graph, and is stored in parallel with the normal features in the form of a binary mask.

[0031] Optionally, the collaborative response component includes:

[0032] The multi-modal input fusion module is used to perform spatio-temporal alignment on the interaction parameter set and the interaction requirements, establish a parameter-requirement mapping table; and standardize the environmental quantification indicators in the interaction parameters;

[0033] The association rule activation module is used to parse the interaction channel configuration preset in the collaborative response component, extract the physical constraint parameters and state variables of each channel, and construct the basic topological structure of the channel feature space; retrieve the eligible subset of association rules, apply the weight vector in the interaction requirements to perform rule priority sorting, and filter out the feasible rule combinations that meet the current environmental constraints;

[0034] The solution space exploration module is used to generate a decision tree with the highest priority rule as the root node, prune the conflicting branches according to the verification flag in the interaction parameters, and retain the decision paths that conform to the characteristics of potential patterns; enumerate solutions within the pruned decision tree space, calculate the matching degree scores of each solution with the interaction requirements, and screen out the composite interaction solutions that meet the requirements.

[0035] The present invention realizes enhanced environmental adaptability: by converting continuous environmental signals into discretized decision nodes that can be quantitatively processed through adaptive interaction components, the inspection operation robot can systematically process the complex and variable unstructured inspection environmental characteristics. The decision-making process is optimized, and the set of interaction parameters generated by the cross-validation mechanism provides an iteratively optimizable decision-making basis for the system, enabling the interaction behavior of the inspection operation robot to continuously improve during the use process. The intention recognition ability is improved: the autonomous intention parsing component realizes the quantitative extraction of the potential needs of the operator through a probability decision-making program and non-linear correlation analysis, solving the problem of inaccurate intention recognition in traditional systems. The response mechanism is intelligent: the self-organizing decision-making mechanism constructed by the collaborative response component integrates environmental parameters and operation requirements through dynamic association rules, enabling the system to generate composite interaction solutions that comprehensively consider multiple factors. The system coordination is strengthened: the combined operation of the three components establishes a complete technical chain for the inspection robot from environmental perception to intention recognition and then to response decision-making, forming a closed-loop optimized intelligent interaction system. It realizes the conversion process from unstructured environmental characteristics to structured decision-making solutions, improving the systematicness and reliability of information processing in complex inspection scenarios.

[0036] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written specification and the drawings.

[0037] The technical solutions of the present invention will be further described in detail below with reference to the drawings and embodiments. Description of the Drawings

[0038] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0039] Figure 1 It is a block diagram of the intelligent inspection operation robot in Embodiment 1 of the present invention;

[0040] Figure 2 It is a block diagram of the adaptive interaction component in Embodiment 2 of the present invention;

[0041] Figure 3 It is a block diagram of the autonomous intention parsing component in Embodiment 4 of the present invention;

[0042] Figure 4 This is the block diagram of the collaborative response component in Embodiment 8 of the present invention. Detailed implementation manners

[0043] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0044] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. In the embodiments of the present application, the singular forms "a", "the", and "said" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0045] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0046] Embodiment 1: As Figure 1 shown, the embodiment of the present invention provides an intelligent inspection operation robot, comprising:

[0047] An adaptive interaction component, configured to collect unstructured features in the inspection environment of the inspection operation robot in real time, convert the continuous signal of the unstructured features into discrete interaction decision nodes, and generate an iteratively optimizable interaction parameter set through a cross-validation mechanism;

[0048] An autonomous intention parsing component, configured to construct a probabilistic decision program based on the implicit features of the operator's behavior trajectory of the inspection operation robot, extract potential patterns of operation intentions through non-linear correlation analysis in combination with the interaction parameter set, and identify unexpressed interaction requirements;

[0049] A collaborative response component, configured to construct a response decision mechanism with self-organizing features, generate a composite interaction plan between the inspection operation robot and the operator by establishing dynamic association rules between different interaction channels of the inspection operation robot and the operator, and input the interaction parameter set and the interaction requirements into the response decision mechanism.

[0050] The working principle and beneficial effects of the above technical solution are as follows: The adaptive interaction component in this embodiment is used to collect unstructured features in the inspection environment of the inspection operation robot in real time, convert the continuous signal of the unstructured features into discrete interaction decision nodes, and generate an iteratively optimizable set of interaction parameters through a cross-validation mechanism; the autonomous intention parsing component is used to construct a probabilistic decision-making program based on the implicit features of the operator's behavior trajectory of the inspection operation robot, extract potential patterns of operation intentions through non-linear correlation analysis in combination with the set of interaction parameters, and identify unexpressed interaction requirements; the collaborative response component is used to construct a response decision-making mechanism with self-organization features, generate a composite interaction plan for the inspection operation robot and the operator by establishing dynamic association rules between different interaction channels of the inspection operation robot and the operator, and input the set of interaction parameters and interaction requirements into the response decision-making mechanism together. The above solution realizes enhanced environmental adaptability: by converting the continuous environmental signal into discrete decision nodes that can be quantitatively processed through the adaptive interaction component, the inspection operation robot can systematically process the complex and changeable unstructured inspection environment features. The decision-making process is optimized, and the set of interaction parameters generated by the cross-validation mechanism provides an iteratively optimizable decision-making basis for the system, enabling the interaction behavior of the inspection operation robot to be continuously improved during the use process. The intention recognition ability is improved: the autonomous intention parsing component realizes the quantitative extraction of the operator's potential needs through the probabilistic decision-making program and non-linear correlation analysis, and solves the problem of inaccurate intention recognition in traditional systems. The response mechanism is intelligent: the self-organization decision-making mechanism constructed by the collaborative response component integrates environmental parameters and operation requirements through dynamic association rules, enabling the system to generate a composite interaction plan that comprehensively considers multiple factors. The system coordination is strengthened: the combined operation of the three components establishes a complete technical chain for the inspection robot from environmental perception to intention recognition and then to response decision-making, forming a closed-loop optimized intelligent interaction system. It realizes the transformation process from unstructured environmental features to structured decision-making schemes, and improves the systematicness and reliability of information processing in complex inspection scenarios.

[0051] Embodiment 2: As Figure 2 shown, on the basis of Embodiment 1, the adaptive interaction component provided by the embodiment of the present invention includes:

[0052] A signal feature extraction module, configured to capture the original continuous signal including spatial form fluctuations and energy distribution gradients of unstructured features through an environmental perception unit, and cut the original continuous signal stream on the time axis into time series segments with overlapping features; each segment passes through a multi-dimensional feature extraction layer to separate core indicators such as signal intensity, frequency domain distribution, and spatial coupling degree, and form a weighted feature vector cluster;

[0053] The time graph construction module is used to input the feature vector clusters into the discretization processing layer. Through the dynamic clustering algorithm in the feature space, the original continuous signal segments are mapped into discrete event nodes with clear semantics; each discrete event node contains three types of attributes; the topological connection relationship between discrete event nodes is automatically constructed based on the spatio-temporal correlation of signal segments to form a dynamically evolving event graph;

[0054] Among them, the three types of attributes include: environmental state label (reflecting the physical feature pattern of the current signal segment); interaction trigger threshold (response condition parameter obtained by training with historical data); association weight matrix (defining the logical relationship between this discrete event node and other nodes);

[0055] Among them, the mapping process of the dynamic clustering algorithm:

[0056] Feature decoupling of the spatio-temporal signal field, multi-scale signal cutting, and dynamic segmentation of the original continuous signal by a non-uniform time window generator: the window length is adaptively adjusted according to the energy distribution gradient mutation rate in the signal stream to ensure that each time series segment contains the minimum period of a complete physical event; an overlap factor is set for adjacent segments, and its width is dynamically calculated by the change slope of the spatial coupling degree in the previous window to prevent feature breakage;

[0057] Composite feature extraction, each signal segment is input into a five-dimensional decoupling layer for separation: intensity fluctuation spectrum (quantifying the non-steady state features of the signal amplitude), frequency domain distortion field (capturing the local abnormal patterns of energy distribution), spatial coherence tensor (describing the phase correlation between multiple sensing units), gradient evolution trajectory (recording the differential change path of feature parameters), historical coupling residual (inheriting the unconverged features of the previous segment); when generating the super-dimensional feature vector, the weights of each dimension are jointly determined by the energy entropy value within the segment and the historical response error;

[0058] Node generation in the dynamic semantic space, construction of a deformable clustering kernel, whose characteristics include: the kernel radius decays logarithmically with the local feature density, and the kernel shape is distorted by the energy field of neighboring nodes to form a non-Euclidean geometric clustering boundary; semantic projection and node hatching, performing bidirectional semantic mapping on each clustering kernel: forward projection performs tensor convolution on the feature vector and the historical event template library to generate candidate semantic labels; backward tracing verifies the evolutionary continuity between the candidate label and the historical graph through spatio-temporal causal chain analysis; only when the forward matching degree > threshold and the backward verification passes, an effective discrete node is hatched;

[0059] The iterative optimization module is used to compare the feature vectors captured by different environmental perception units within the same time window, and eliminate abnormal nodes caused by environmental noise; match the currently generated interaction decision nodes with the historical event graph, and correct the deviation of the node attribute parameters; the verified interaction decision nodes enter the parameter optimizer, and according to the actual interaction effect feedback after the interaction decision nodes are triggered, dynamically adjust the weight allocation strategy of the interaction parameter set.

[0060] The working principle and beneficial effects of the above technical solution are as follows: The signal feature extraction module of this embodiment captures the original continuous signal containing spatial morphological fluctuations and energy distribution gradients of unstructured features through the environmental perception unit, and cuts the original continuous signal stream on the time axis into time series segments with overlapping features; each segment passes through the multi-dimensional feature extraction layer to separate core indicators such as signal strength, frequency domain distribution, and spatial coupling degree, forming a weighted feature vector cluster; the time graph construction module inputs the feature vector cluster into the discretization processing layer, and through the dynamic clustering algorithm in the feature space, maps the original continuous signal segment into discrete event nodes with clear semantics; each discrete event node contains three types of attributes; the topological connection relationship between discrete event nodes is automatically constructed based on the spatio-temporal correlation of the signal segments, forming a dynamically evolving event graph; among them, the three types of attributes include: environmental state label (reflecting the physical feature pattern of the current signal segment); interaction trigger threshold (response condition parameter obtained by training with historical data); association weight matrix (defining the logical relationship between this discrete event node and other nodes); the iterative optimization module compares the feature vectors captured by different environmental perception units within the same time window, and eliminates abnormal nodes caused by environmental noise; matches the currently generated interaction decision nodes with the historical event graph, and corrects the deviation of the node attribute parameters; the verified interaction decision nodes enter the parameter optimizer, and according to the actual interaction effect feedback after the interaction decision nodes are triggered, dynamically adjust the weight allocation strategy of the interaction parameter set. The above solution realizes the intelligent analysis of the original continuous signal through the signal feature extraction module, and converts the unstructured continuous environmental signal into a quantifiable feature vector cluster; the separation of signal strength, frequency domain distribution, and spatial coupling degree by the multi-dimensional feature extraction layer establishes a mapping bridge between the physical environment and the digital model. The time graph construction module combines discretization processing with the dynamic clustering algorithm to realize the transformation from the underlying signal features to the high-level semantic events; by defining three types of node attributes including environmental state label, interaction trigger threshold, and association weight matrix, a dynamically evolving event graph with spatio-temporal correlation is constructed, providing an interpretable decision basis for the system. The iterative optimization module ensures the reliability of the system output through the multi-source data verification and pattern matching mechanism; the parameter optimizer introduces a closed-loop feedback mechanism, enabling the interaction parameter set to be dynamically adjusted according to the actual effect, forming a continuously evolving decision-making ability; the abnormal node elimination and attribute parameter correction mechanisms jointly ensure the anti-interference ability of the system.

[0061] In summary, this embodiment realizes a complete closed-loop from raw signal acquisition to intelligent decision-making: the continuous signals captured by the environmental perception unit are transformed into discrete event nodes with clear semantics after feature extraction and event mapping; these nodes form an evolvable knowledge graph through dynamically constructed topological relationships; finally, an interaction strategy that conforms to the environmental characteristics is output through a continuously optimized decision-making mechanism.

[0062] Embodiment 3: On the basis of Embodiment 2, the time graph construction module provided by the embodiment of the present invention includes:

[0063] The node activation condition matching sub-module is used to compare the matching degree between the feature vector extracted from the current environmental perception signal and the environmental state label of the node in real time based on the environmental state label of the discrete event node in the event graph; when the multi-dimensional indicators such as the signal strength and frequency domain distribution of the feature vector reach the preset spatial coupling threshold with the physical feature pattern of the environmental state label of the node, the candidate qualification of the node is triggered.

[0064] The interaction trigger threshold determination sub-module is used to evaluate the threshold of the node with candidate qualification. Its built-in interaction trigger threshold is generated by training with historical interaction effect data; when the composite evaluation value of the real-time environmental state fluctuation parameter and the operation requirement index exceeds the interaction trigger threshold, the node is activated as an effective interaction decision node.

[0065] The node association network collaborative decision-making sub-module is used to establish a logical operation relationship between the activated interaction decision nodes through an association weight matrix and adjacent interaction decision nodes to form a local decision network; the activated interaction decision nodes within the same time window exchange environmental state parameters through the weight matrix to generate a comprehensive environmental evaluation value; trace back the response effect data of historical nodes along the time evolution path of the event graph to optimize the reliability of the current decision, and screen out a set of interaction decision nodes that meet the spatio-temporal consistency requirements through a multi-dimensional voting mechanism.

[0066] The working principle and beneficial effects of the above technical solution are as follows: The node activation condition matching sub-module of this embodiment is based on the environmental state labels of discrete event nodes in the event graph, and compares in real time the matching degree between the feature vector extracted from the current environmental perception signal and the environmental state labels of the nodes; when the multi-dimensional indicators such as the signal strength and frequency domain distribution of the feature vector reach the preset spatial coupling threshold with the physical feature pattern of the environmental state labels of the nodes, the candidate qualification of the nodes is triggered; the interaction trigger threshold determination sub-module evaluates the nodes with candidate qualifications for the threshold, and its built-in interaction trigger threshold is generated by training with historical interaction effect data; when the composite evaluation value of the real-time environmental state fluctuation parameters and the operation requirement indicators exceeds the interaction trigger threshold, the node is activated as an effective interaction decision node; the node association network collaborative decision sub-module establishes a logical operation relationship between the activated interaction decision nodes and adjacent interaction decision nodes through the association weight matrix to form a local decision network; the activated interaction decision nodes within the same time window exchange environmental state parameters through the weight matrix to generate a comprehensive environmental evaluation value; trace back the response effect data of historical nodes along the time evolution path of the event graph to optimize the reliability of the current decision, and screen out a set of interaction decision nodes that meet the spatio-temporal consistency requirements through a multi-dimensional voting mechanism. The time graph construction module of the above solution realizes a closed-loop processing flow from environmental state perception to interaction decision generation; through the multi-dimensional matching mechanism of feature vectors and state labels, continuous environmental signals are discretized into candidate event nodes, completing the computable conversion from the physical environment to graph nodes, and solving the modal alignment problem between environmental signals and knowledge graphs. Based on the dynamic threshold mechanism driven by historical data, effective nodes with operational significance are screened out among candidate nodes, realizing the dual-parameter coupling decision of environmental fluctuations and operation requirements, and ensuring the necessity of interaction triggering. By constructing a non-linear relationship network between nodes through the weight matrix, combining the time dimension and spatial association dimension of historical response data, a decision set that meets the spatio-temporal consistency constraints is output, solving the locality problem of single-point decision-making and forming a globally optimized decision-making scheme.

[0067] In summary, this embodiment realizes the quantitative mapping from environmental state to operation decision, establishes a dynamic trigger mechanism based on historical experience, ensures the reliability of the decision through spatio-temporal dimension verification, and forms a closed-loop decision-making system with environmental adaptability. The output result is a set of interaction decision nodes verified through multiple dimensions, which contains both the optimal response strategy for the current environment and inherits the effective experience mode of historical decisions.

[0068] Embodiment 4: As Figure 3 shown, on the basis of Embodiment 1, the autonomous intention parsing component provided by the embodiment of the present invention includes:

[0069] An implicit feature extraction module is used to extract implicit features such as temporal dependence, operation preferences, and abnormal operation patterns from the operator behavior trajectory data recorded by the inspection operation robot during operation, and form feature vectors; the operator behavior trajectory data includes operation instruction sequences, operation frequencies, operation intervals, path selections, and equipment adjustment modes, etc.

[0070] A probability decision program construction module is used to establish a probability model based on the historical operator behavior trajectory data, calculate the occurrence probabilities of different operation modes and their conditional dependencies; combine the current interaction parameter set to dynamically adjust the probability weights and form a decision-making logic adapted to different scenarios.

[0071] A potential pattern extraction module is used to jointly analyze the interaction parameter set and the implicit features, and use non-linear mapping to mine the potential correlation between the interaction parameter set and the implicit features; through pattern matching and clustering analysis, identify the potential intentions of the operator under different environmental conditions, including unexpressed interaction requirements.

[0072] The working principles and beneficial effects of the above technical solutions are as follows: The implicit feature extraction module of this embodiment extracts the operator behavior trajectory data recorded by the inspection operation robot during operation, extracts implicit features such as temporal dependence, operation preferences, and abnormal operation patterns from the behavior trajectory data, and forms feature vectors; the operator behavior trajectory data includes operation instruction sequences, operation frequencies, operation intervals, path selections, and equipment adjustment modes, etc.; the probability decision program construction module is used to establish a probability model based on the historical operator behavior trajectory data, calculate the occurrence probabilities of different operation modes and their conditional dependencies; combine the current interaction parameter set to dynamically adjust the probability weights and form a decision-making logic adapted to different scenarios; the potential pattern extraction module is used to jointly analyze the interaction parameter set and the implicit features, use non-linear mapping to mine the potential correlation between the interaction parameter set and the implicit features; through pattern matching and clustering analysis, identify the potential intentions of the operator under different environmental conditions, including unexpressed interaction requirements. The implicit feature extraction module of the above solution extracts implicit features such as temporal dependence, operation preferences, and abnormal operation patterns from the operator behavior trajectory data and converts them into computable feature vectors. The probability decision program construction module establishes a probability model based on the historical behavior data, calculates the occurrence probabilities of different operation modes and their conditional dependencies, and dynamically adjusts the decision weights in combination with the current interaction parameter set to make the decision-making logic adapt to different inspection scenarios. The potential pattern extraction module jointly analyzes the interaction parameter set (from environmental perception) and the operator's implicit features, and through non-linear mapping and pattern matching, mines the potential correlation between the two, and identifies the potential intentions of the operator under different environmental conditions, including unexpressed interaction requirements.

[0073] In summary, this embodiment converts the operator's behavior data into computable decision-making basis, enhancing the robot's autonomous decision-making ability; dynamically adjusts the decision-making logic in combination with environmental parameters, enabling the robot to adapt to different inspection scenarios; through non-linear correlation analysis, identifies the needs not explicitly expressed by the operator, enhancing the accuracy and adaptability of human-robot collaboration. It forms a data-driven intelligent intention parsing mechanism, enabling the inspection operation robot to more efficiently understand the operator's intention and make reasonable responses.

[0074] Embodiment 5: On the basis of Embodiment 4, the implicit feature extraction module provided by the embodiment of the present invention includes:

[0075] A data structuring and processing sub-module for receiving the original data stream of the operator's behavior trajectory, including discrete operation instruction sequences and continuous timestamp records; dividing the continuous original data stream into analyzable segments through time window segmentation, and each segment retains the complete operation context relationship;

[0076] A feature content conversion sub-module for performing temporal dependence relationship modeling, operation preference quantification, and abnormal pattern detection;

[0077] Among them, the temporal dependence relationship modeling analyzes the transition probability matrix between operation instructions of adjacent timestamps, captures the forced sequential relationship and optional branches between operation steps; at the same time, calculates the periodic pattern repeatability across time windows, identifies the operator's habitual work rhythm, and outputs a three-dimensional temporal tensor containing the temporal constraint intensity, where the third dimension encodes the dependence intensity at different time scales;

[0078] The operation preference quantification converts discrete behaviors such as the path selection frequency of the operation preference of the original data stream and the adjustment amplitude of device parameters into probability distributions, determining the degree of decision certainty of the operator; performs co-occurrence analysis on high-frequency operation combinations, establishes a behavior association graph, and the edge weight of the graph reflects the stability degree of the operation combination. The temporal tensor and the behavior graph together constitute the intermediate feature representation;

[0079] The abnormal pattern detection defines a dynamic density threshold, and triggers an abnormal mark when the local density of the newly input operator's behavior trajectory data is lower than the historical distribution threshold; the abnormal mark includes instruction jumps that violate temporal constraints and isolated operation nodes that deviate from the behavior graph, and is stored in parallel with the normal features in the form of a binary mask;

[0080] A feature vector synthesis sub-module for expanding the three-dimensional temporal tensor into a sparse vector, performing eigenvalue decomposition on the edge weight matrix of the behavior graph, and converting the mask of the abnormal mark into a position encoding; the sparse vector, eigenvalue decomposition, and position encoding are orthogonally spliced to form a composite feature vector with a fixed dimension.

[0081] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the data structured processing sub-module receives the original data stream of the operator's behavior trajectory, including discrete operation instruction sequences and continuous timestamp records; the continuous original data stream is divided into analyzable segments through time window segmentation, and each segment retains the complete operation context relationship; the feature content conversion sub-module performs temporal dependency relationship modeling, operation preference quantification, and abnormal pattern detection; among them, the temporal dependency relationship modeling analyzes the transition probability matrix between operation instructions of adjacent timestamps to capture the mandatory sequential relationship and optional branches between operation steps; at the same time, the periodic pattern repeatability across time windows is calculated to identify the operator's habitual work rhythm, and a three-dimensional temporal tensor containing the temporal constraint strength is output, where the third dimension encodes the dependency strength at different time scales; the operation preference quantification converts discrete behaviors such as the path selection frequency of the operation preference of the original data stream and the adjustment amplitude of device parameters into a probability distribution to determine the degree of decision certainty of the operator; co-occurrence analysis is performed on high-frequency operation combinations to establish a behavior association graph, and the edge weight of the graph reflects the stability of the operation combination. The temporal tensor and the behavior graph together constitute the intermediate feature representation; the abnormal pattern detection defines a dynamic density threshold, and an abnormal mark is triggered when the local density of the newly input operator behavior trajectory data is lower than the historical distribution threshold; the abnormal mark includes instruction jumps that violate temporal constraints and isolated operation nodes that deviate from the behavior graph, and is stored in parallel with the normal features in the form of a binary mask; the feature vector synthesis sub-module unfolds the three-dimensional temporal tensor into a sparse vector, performs eigenvalue decomposition on the edge weight matrix of the behavior graph, and converts the mask of the abnormal mark into a position encoding; the sparse vector, eigenvalue decomposition, and position encoding are orthogonally spliced to form a composite feature vector with a fixed dimension. The above solution converts the original operation behavior stream into discrete segments with complete semantics through time window segmentation, ensuring that the analysis is always based on meaningful operation contexts and avoiding information fragmentation. The temporal dependency modeling quantifies the logical constraints (mandatory sequence / optional branch) and long-term work rules between operation steps through the transition probability matrix and periodic pattern analysis, and outputs a three-dimensional temporal tensor to retain multi-scale temporal relationships. The behavior preference quantification converts discrete operations into probability distributions and association graphs, reflecting both the operator's decision-making mode (degree of certainty) and the stability of operation combinations, and supplementing discrete behavior characteristics not covered by the temporal model. The abnormal detection robustness dynamic density threshold mechanism distinguishes normal and abnormal behaviors: it detects both instruction jumps that violate temporal constraints (such as skipping necessary steps) and isolated operations that deviate from the behavior graph (such as unconventional parameter adjustments), and realizes the parallel storage of abnormal marks and normal features through a binary mask. The feature space unified encoding orthogonally splices heterogeneous temporal tensors (sparse vectors), behavior graphs (eigenvalue decomposition), and abnormal marks (position encoding) into a fixed-dimension vector to solve the problem of multi-modal feature fusion.Implement the end - to - end conversion of operation behavior data from the original time series stream to machine - parsable feature vectors, while preserving three key features: time series logic, behavior preferences, and anomaly information.

[0082] Embodiment 6: Based on Embodiment 4, the probability - decision - program construction module provided by the embodiments of the present invention includes:

[0083] A probability - model initialization sub - module, which is used to construct an initial probability model based on historical operator behavior trajectory data and the feature vectors output by the implicit - feature extraction module, quantify the occurrence probabilities of different operation modes, and analyze their conditional dependencies; including:

[0084] Operation - mode probability distribution: Statistically analyze the occurrence frequencies of high - frequency operation combinations to form a basic probability distribution;

[0085] Conditional - dependency network: By analyzing the transition - probability matrix in the time - series tensor and the edge weights in the behavior graph, establish the conditional dependencies between operation steps (e.g., a certain operation is likely to be followed by a specific subsequent operation);

[0086] A dynamic - weight - adjustment sub - module, which is used to dynamically adjust the weight allocation of the decision - making logic based on the initial probability model and in combination with the current interaction parameter set;

[0087] Scene - adaptability evaluation: Compare the current interaction parameters with the environmental conditions in historical data, screen out operation modes in similar scenarios, and adjust their probability weights (e.g., some operations are more likely to occur under specific environments);

[0088] Anomaly - mode suppression: If the anomaly marker (binary mask) in the implicit features detects that the current operation deviates from the norm, then reduce the decision weight of the abnormal path to avoid incorrect decisions;

[0089] A potential - intention feedback - correction sub - module, which is used to input the dynamically adjusted probability model into the potential - mode extraction module and conduct joint analysis with the implicit features; form decision - making logics suitable for different scenarios;

[0090] Non - linear - mapping mining: Explore the potential correlations between interaction parameters (such as environmental light, device status) and operation preferences (such as path selection, adjustment amplitude), and identify unexpressed requirements (e.g., the operator tends to a certain optimization strategy under specific conditions);

[0091] Mode clustering and intention matching: Through clustering analysis, match the current operation features with historical potential - intention categories, and further correct the output of the probability model to make it more in line with the operator's true intention.

[0092] The working principle and beneficial effects of the above technical solution are as follows: The probability model initialization sub-module of this embodiment constructs an initial probability model based on historical operator behavior trajectory data and the feature vectors output by the implicit feature extraction module, quantifies the occurrence probabilities of different operation modes, and analyzes their conditional dependence relationships; the dynamic weight adjustment sub-module dynamically adjusts the weight distribution of the decision-making logic based on the initial probability model and in combination with the current interaction parameter set; the potential intention feedback and correction sub-module is used to input the dynamically adjusted probability model into the potential mode extraction module for joint analysis with the implicit features; and a decision-making logic adapted to different scenarios is formed. The above solution, through the probability model initialization sub-module, converts the structured data (such as time series tensors, behavior graphs, etc.) output by the implicit feature extraction module into computable decision-making bases; establishes a probabilistic expression of the operation mode, quantifies the normal distribution of various operation combinations; constructs a conditional dependence network, and clarifies the logical constraint relationships between operation steps. The real-time decision-making optimization is achieved by means of the dynamic weight adjustment sub-module. The scenario adaptation mechanism ensures that the decision-making weights are adjusted according to changes in environmental parameters, and the anomaly suppression function maintains the robustness of the decision-making system and filters out unconventional interferences. The decision-making closed-loop is completed through the potential intention feedback and correction sub-module. The non-linear correlation analysis reveals the hidden relationships between environmental parameters and operation preferences, and the pattern matching enables the decision-making model to respond to potential demands that are not explicitly expressed. The evolution from a static probability model to dynamic scenario adaptation is realized, and finally a decision-making logic that conforms to operation routines and can adapt to special scenarios is output, providing the autonomous system with the ability to continuously optimize the generation of behavior strategies.

[0093] Embodiment 7: On the basis of Embodiment 4, the potential mode extraction module provided by the embodiment of the present invention includes:

[0094] The feature space fusion sub-module is used to perform multi-dimensional space fusion on the feature vectors generated by the implicit feature extraction module and the current interaction parameter set; establish a unified representation space, convert the discrete operation instruction sequence into a continuous behavior trajectory vector, and generate an enhanced feature tensor containing the environment-behavior coupling relationship through feature crossing;

[0095] The association pattern discovery sub-module is used to perform hierarchical association analysis in the fused feature space;

[0096] Primary association layer: Use the conditional dependence relationship provided by the probability decision module as a prior constraint; identify the linear correspondence between the explicit operation mode and environmental parameters; screen out abnormal combinations that deviate significantly from the historical probability distribution;

[0097] Deep association layer: Construct a feature interaction network to capture cross-dimensional non-linear associations; discover implicit conditional dependence chains through iterative optimization; verify the logical consistency between the potential mode and the operation preferences in the implicit features;

[0098] The intention graph construction sub-module is used for dynamic clustering execution, obtaining the similarity of behavior patterns based on the enhanced feature tensor, combining the scene classification results in the probability model to divide the clustering boundary, and generating a set of intention prototypes with environmental condition annotations;

[0099] Among them, the calculation process of the similarity of behavior patterns based on the enhanced feature tensor: perform dimensionality compression on the enhanced feature tensor, retain the key environment-behavior coupling dimensions, and filter out the redundant dimensions caused by noise; define a distance function in the reduced tensor space, incorporate the scene weight parameters in the probability model, and establish a composite metric standard considering the influence of environmental conditions; obtain the pairwise distances between behavior trajectory vectors, apply a non-linear transformation to convert them into similarity scores, and use a symmetric similarity relationship matrix; adjust the parameters of the distance function according to real-time interaction parameters, and correct the marginal similarity values in combination with anomaly markers, and output a time-varying evaluation of the similarity of behavior patterns;

[0100] The real-time intention mapping sub-module is used to project the current behavior trajectory into the intention graph space, calculate the multi-dimensional distance metric with each intention prototype, and perform confidence calibration in combination with the dynamic weight adjustment result; generate a set of potential intentions with probability weights, and attach an environmental condition matching degree index.

[0101] The working principle and beneficial effects of the above technical solution are as follows: The feature space fusion sub-module of this embodiment performs multi-dimensional space fusion based on the feature vectors generated by the implicit feature extraction module and the current interaction parameter set; establishes a unified representation space, converts the discrete operation instruction sequence into a continuous behavior trajectory vector, and generates an enhanced feature tensor containing the environment-behavior coupling relationship through feature crossing; the association pattern discovery sub-module is used to perform hierarchical association analysis in the fused feature space; the intention graph construction sub-module performs dynamic clustering execution, calculates the similarity of behavior patterns based on the enhanced feature tensor, combines the scene classification results in the probability model to divide the clustering boundary, and generates a set of intention prototypes with environmental condition annotations; the real-time intention mapping sub-module projects the current behavior trajectory into the intention graph space, calculates the multi-dimensional distance metric with each intention prototype, and performs confidence calibration in combination with the dynamic weight adjustment result; generates a set of potential intentions with probability weights, and attaches an environmental condition matching degree index. The above solution is implemented through the feature space fusion sub-module, which converts discrete operation instructions into continuous computable behavior vectors and generates an enhanced feature tensor that retains the characteristics of the original data. The association pattern discovery sub-module provides a dual analysis mechanism: the primary association layer ensures that pattern recognition conforms to historical operation rules, and the deep association layer expands the discovery ability of non-linear relationships, forming a complete pattern discovery chain from explicit to implicit. The intention graph construction and real-time mapping sub-modules jointly achieve: converting abstract behavior patterns into quantifiable intention prototypes, establishing a three-dimensional mapping relationship of behavior-environment-intention, and outputting a standardized intention description with probability evaluation.

[0102] Example 8: AsFigure 4 As shown, based on Embodiment 1, the collaborative response component provided by the embodiment of the present invention includes:

[0103] A multimodal input fusion module, configured to perform spatio-temporal alignment between the interaction parameter set and the interaction requirements, establish a parameter-requirement mapping table; and perform standardization processing on the environmental quantification indicators in the interaction parameters;

[0104] An association rule activation module, configured to parse the interaction channel configuration preset in the collaborative response component, extract the physical constraint parameters and state variables of each channel, and construct the basic topological structure of the channel feature space; retrieve the eligible subset of association rules, apply the weight vector in the interaction requirements to perform rule priority ranking, and filter out the feasible rule combinations that meet the current environmental constraints;

[0105] A solution space exploration module, configured to generate a decision tree with the highest-priority rule as the root node, prune the conflicting branches according to the verification flag in the interaction parameters, and retain the decision paths that conform to the potential pattern features; perform solution enumeration in the pruned decision tree space, calculate the matching degree scores between each solution and the interaction requirements, and filter out the compliant composite interaction solutions.

[0106] The working principle and beneficial effects of the above technical solution are as follows: The multimodal input fusion module in this embodiment is used to perform spatio-temporal alignment between the interaction parameter set and the interaction requirements, and establish a parameter-requirement mapping table; perform standardization processing on the environmental quantification indicators in the interaction parameters; the association rule activation module is used to parse the interaction channel configuration preset by the collaborative response component, extract the physical constraint parameters and state variables of each channel, and construct the basic topological structure of the channel feature space; retrieve the eligible subset of association rules, perform rule priority sorting using the weight vector in the interaction requirements, and screen out the feasible rule combinations that meet the current environmental constraints; the solution space exploration module is used to generate a decision tree with the highest-priority rule as the root node, prune the conflicting branches according to the verification flag in the interaction parameters, and retain the decision paths that conform to the potential pattern features; perform solution enumeration within the pruned decision tree space, calculate the matching degree scores between each solution and the interaction requirements, and screen out the compliant composite interaction solutions. The above solution is implemented through the multimodal input fusion module, which unifies the spatio-temporal benchmarks of heterogeneous interaction parameters and requirements, performs unit consistency conversion on environmental quantification indicators, and establishes a computable parameter-requirement mapping relationship. The association rule activation module provides channel feature modeling based on physical constraints, dynamically adjusts rule priorities driven by weights, and an environmental perception-based feasible rule screening mechanism. The solution space exploration module realizes decision tree construction based on rule priorities, decision path screening guided by verification flags, and solution optimization with quantified demand matching degrees. It realizes the complete conversion process from the original interaction parameters to executable solutions, and each module maintains processing continuity through a standardized data interface, and finally outputs an interaction solution that meets the triple constraints of spatio-temporal consistency, environmental adaptability, and demand matching.

[0107] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the equivalent technology of the present invention, the present invention also intends to include these changes and modifications.

Claims

1. An intelligent inspection operation robot, characterized in that, Comprising: A collaborative response component, which is used to construct a response decision-making mechanism with self-organization characteristics. By establishing dynamic association rules between different interaction channels of the inspection operation robot and the operator, the interaction parameter set and the interaction requirements are jointly input into the response decision-making mechanism to generate a composite interaction plan for the inspection operation robot and the operator.

2. The intelligent inspection operation robot according to claim 1, wherein, An adaptive interaction component of the intelligent inspection operation robot, which is used to collect the unstructured features in the inspection environment of the inspection operation robot in real time, convert the continuous signal of the unstructured features into discrete interaction decision nodes, and generate an iteratively optimizable interaction parameter set through a cross-validation mechanism.

3. The intelligent inspection operation robot according to claim 2, wherein, The adaptive interaction component comprises: A signal feature extraction module, which is used to capture the original continuous signal containing the spatial form fluctuation and energy distribution gradient of the unstructured features through the environment perception unit, and cut the original continuous signal flow on the time axis into time series segments with overlapping features; each segment passes through a multi-dimensional feature extraction layer to separate the signal intensity, frequency domain distribution and spatial coupling degree, and form a weighted feature vector cluster. A time graph construction module, which is used to input the feature vector cluster into the discretization processing layer, and map the original continuous signal segment into discrete event nodes with clear semantics through the dynamic clustering algorithm in the feature space; the topological connection relationship between the discrete event nodes is automatically constructed based on the spatio-temporal correlation of the signal segments to form a dynamically evolving event graph. An iterative optimization module, which is used to compare the feature vectors captured by different environment perception units within the same time window to eliminate abnormal nodes caused by environmental noise; match the currently generated interaction decision nodes with the historical event graph to correct the deviation of the node attribute parameters; the interaction decision nodes that pass the verification enter the parameter optimizer, and the weight allocation strategy of the interaction parameter set is dynamically adjusted according to the actual interaction effect feedback after the interaction decision nodes are triggered.

4. The intelligent inspection operation robot according to claim 3, characterized in that, The time graph construction module comprises: A node activation condition matching sub-module, which is used to compare the matching degree between the feature vector extracted from the current environment perception signal and the environmental state label of the node in real time based on the environmental state label of the discrete event node in the event graph; when the multi-dimensional indexes of the signal intensity and frequency domain distribution of the feature vector reach the preset spatial coupling threshold with the physical feature pattern of the environmental state label of the node, the candidate qualification of the node is triggered. An interaction trigger threshold determination sub-module, which is used to evaluate the threshold of the nodes with candidate qualifications, and its built-in interaction trigger threshold is generated by training with historical interaction effect data. When the composite evaluation value of the real-time environmental state fluctuation parameter and the operation requirement index exceeds the interaction trigger threshold, the node is activated as an effective interaction decision node. A node association network collaborative decision-making sub-module, which is used to establish a logical operation relationship between the activated interaction decision nodes and adjacent interaction decision nodes through an association weight matrix to form a local decision-making network; the activated interaction decision nodes within the same time window exchange environmental state parameters through the weight matrix to generate a comprehensive environmental evaluation value; trace back the response effect data of historical nodes along the time evolution path of the event graph to optimize the reliability of the current decision, and screen out a set of interaction decision nodes that meet the spatio-temporal consistency requirements through a multi-dimensional voting mechanism.

5. The intelligent inspection operation robot according to claim 1, characterized in that, The autonomous intention parsing component of the intelligent inspection operation robot is used to construct a probability decision program based on the implicit features of the operator's behavior trajectory of the inspection operation robot, extract the potential patterns of operation intentions through non-linear correlation analysis in combination with the interaction parameter set, and identify the unexpressed interaction requirements.

6. The intelligent inspection operation robot according to claim 5, wherein, The autonomous intention parsing component includes: The implicit feature extraction module is used for the operator behavior trajectory data recorded by the inspection operation robot during operation, extracts the implicit features of temporal dependence, operation preference and abnormal operation mode from the behavior trajectory data, and forms a feature vector; The probability decision program construction module is used to establish a probability model based on the data of the historical operator behavior trajectory, calculate the occurrence probability of different operation modes and their conditional dependence relationships; in combination with the current interaction parameter set, dynamically adjust the probability weights to form decision-making logics adapted to different scenarios; The potential pattern extraction module is used to jointly analyze the interaction parameter set and the implicit features, and uses non-linear mapping to mine the potential correlation between the interaction parameter set and the implicit features; through pattern matching and clustering analysis, identify the potential intentions of the operator under different environmental conditions, including unexpressed interaction requirements.

7. The intelligent inspection operation robot according to claim 6, characterized in that, The operator behavior trajectory data of the implicit feature extraction module includes operation instruction sequences, operation frequencies, operation intervals, path selections and equipment adjustment modes.

8. The intelligent inspection operation robot according to claim 6, wherein, The implicit feature extraction module includes: The data structured processing sub-module is used to receive the original data stream of the operator behavior trajectory, including discrete operation instruction sequences and continuous timestamp records; divides the continuous original data stream into analyzable segments through time window segmentation, and each segment retains the complete operation context relationship; The feature content conversion sub-module is used to perform temporal dependence relationship modeling, operation preference quantification and abnormal mode detection; The feature vector synthesis sub-module is used to expand the three-dimensional temporal tensor into a sparse vector, perform eigenvalue decomposition on the edge weight matrix of the behavior graph, and convert the mask of the abnormal mark into a position encoding; the sparse vector, eigenvalue decomposition and position encoding are orthogonally spliced to form a composite feature vector with a fixed dimension.

9. The intelligent inspection operation robot according to claim 8, wherein, Among them, The temporal dependence relationship modeling analyzes the transition probability matrix between operation instructions of adjacent timestamps, captures the mandatory order relationship and optional branches between operation steps; at the same time, calculates the periodic pattern repeatability across time windows, identifies the habitual work rhythm of the operator, and outputs a three-dimensional temporal tensor containing the temporal constraint intensity, where the third dimension encodes the dependence intensity at different time scales; The operation preference quantification converts the discrete behaviors of the path selection frequency and the equipment parameter adjustment amplitude of the operation preference of the original data stream into a probability distribution, and determines the degree of decision certainty of the operator; performs co-occurrence analysis on high-frequency operation combinations, establishes a behavior association graph, and the edge weight of the graph reflects the stability degree of the operation combination. The temporal tensor and the behavior graph jointly constitute the intermediate feature representation; The abnormal mode detection defines a dynamic density threshold, which triggers an abnormal flag when the local density of the newly input operator behavior trajectory data is lower than the historical distribution threshold; the abnormal flag includes instruction jumps that violate the timing constraint and isolated operation nodes that deviate from the behavior pattern map, and is stored in parallel with the normal features in the form of a binary mask.

10. The intelligent inspection operation robot according to claim 1, characterized in that, The collaborative response component includes: The multimodal input fusion module is used to perform spatio-temporal alignment between the interaction parameter set and the interaction requirements, establish a parameter-requirement mapping table; and standardize the environmental quantization indicators in the interaction parameters. The association rule activation module is used to parse the interaction channel configuration preset in the collaborative response component, extract the physical constraint parameters and state variables of each channel, and construct the basic topological structure of the channel feature space; retrieve the eligible subset of association rules, apply the weight vector in the interaction requirements to perform rule priority sorting, and filter out the feasible rule combinations that meet the current environmental constraints. The solution space exploration module is used to generate a decision tree with the highest priority rule as the root node, prune the conflict branches according to the verification flag in the interaction parameters, and retain the decision paths that conform to the potential pattern features; perform solution enumeration in the pruned decision tree space, calculate the matching degree scores of each solution and the interaction requirements, and filter out the composite interaction solutions that meet the requirements.

Images