A voice interaction processing system and method for a digitalized building block robot

Abstract

The application relates to the technical fields of robot control and man-machine interaction, and discloses a voice interaction processing system and method for a digitalized building block robot, which comprises a topology construction module, a feature evaluation module, a retrieval module and a degradation control module. The topology construction module constructs a topology directed acyclic graph according to bus detection data or dynamics characteristic data of an inertial measurement module; the feature evaluation module calculates spatial feature parameters of a building block module and converts the spatial feature parameters into semantic retrieval alias reconstruction inverted index; the retrieval module analyzes voice to extract an intention text, generates an action script candidate set and a target topology subgraph; and the degradation control module performs a one-to-one mapping comparison on a graph structure, calculates a maximum common subgraph when the comparison fails, extracts a compensation script in combination with a key degree penalty factor, and replaces a missing node control code with an isochronous long empty operation instruction frame. The application solves the problems of topology identification under communication restriction, semantic matching after shape change and safe degradation when components are damaged.

Description

Technical Field

[0001] This invention relates to the field of robot control and human-computer interaction technology, specifically to a voice interaction processing system and method for a digital building block robot. Background Technology

[0002] Digital modular robots are assembled from a main control module and various actuator and sensor modules with different functions. Users can issue interactive commands via voice to drive the robot to perform specific actions. In practical applications, the main control unit needs to accurately obtain the physical assembly status of each module to call the corresponding drive script. However, existing recognition schemes heavily rely on the communication protocol stack of the underlying data bus. When the robot uses bus hardware with a simple structure and limited communication capabilities, the main control unit can often only detect the set of devices connected to the bus and cannot directly read the cascading dependency relationships between various nodes. This communication limitation prevents the system from effectively reconstructing the true assembly form of the entire robot and reduces the system's compatibility with different underlying hardware.

[0003] Meanwhile, a core application characteristic of modular robots lies in the variability of their assembly form. When a user changes the physical assembly form of the robot, the topological spatial structure of the hardware changes accordingly. Existing voice interaction systems typically statically bind control scripts to fixed hardware descriptions or instruction vocabulary, lacking an internal data mapping mechanism to convert physical topological parameters into natural language text. This fixed matching method results in the original voice action commands being detached from the preset hardware structure in the newly assembled robot form. The system cannot accurately retrieve and match the corresponding action script in the current physical entity, rendering the voice interaction function ineffective.

[0004] Furthermore, during the actual assembly process of the robot, it is inevitable that components will be missing or physical ports will be incorrectly connected. Existing control systems, upon receiving voice control commands, typically send control code directly to the underlying driver stack, lacking a graph-based physical entity constraint verification step before command execution. Forcing the original multi-axis linkage motions onto a flawed or erroneous mechanical structure can easily lead to interference between mechanical components, and even cause center of gravity shifts and equipment tipping over. Even when error interception is implemented, some existing systems often simply cut off the execution link, lacking a safety compensation assessment strategy based on existing hardware conditions. Moreover, hard blocking disrupts the timing integrity of the underlying motor multi-axis interpolation motion. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a voice interaction processing system and method for digital building block robots. It solves the problems of existing digital building block robots being unable to recognize physical topology when communication at the underlying level is limited, unable to perform semantic matching of voice action commands after assembly reconstruction, and lacking action degradation compensation and timing protection mechanisms when body components are damaged.

[0006] To achieve the above objectives, the first aspect of the present invention provides a voice interaction processing system for a digital building block robot, comprising:

[0007] The system includes a main control unit, a data bus, and multiple modular modules electrically connected to the main control unit via the data bus; the system also includes a map storage module and a script storage module; the main control unit internally includes a topology construction module, a feature evaluation module, a retrieval module, and a degradation control module.

[0008] The topology building module is used to periodically poll the data bus to obtain connection relationship data and generate a real-time topology directed acyclic graph in memory;

[0009] The feature evaluation module is used to calculate the spatial feature parameters of the building block module in the topology and convert them into semantic retrieval aliases to reconstruct the inverted index;

[0010] The retrieval module is used to parse external speech audio streams to extract text strings, and generate a candidate set of action scripts and a target topology subgraph based on the reconstructed index;

[0011] The degradation control module is used to output compensatory action control instructions based on the graph structure similarity calculation results when the action script cannot accurately match the physical shape.

[0012] Preferably, the main control unit integrates an inertial measurement module; when the underlying communication link is limited, the topology building module sends a micro-motion detection pulse signal to the building block module, and the inertial measurement module collects three-axis motion data to generate a transient kinematic signature feature vector for matching the standard topology map.

[0013] Preferably, the retrieval module includes an isomorphic matching submodule and a subgraph calculation submodule; the isomorphic matching submodule is used to perform injective function constraint comparison; the subgraph calculation submodule is used to calculate the maximum common subgraph and the difference set of physically missing nodes between the target topological subgraph and the real-time topological directed acyclic graph.

[0014] A second aspect of the present invention provides a voice interaction processing method for a digital building block robot, applied to any of the systems described in the first aspect of the present invention, the method comprising the following steps:

[0015] Step S10: Periodically poll the data bus and construct a real-time topological directed acyclic graph representing the current physical hardware state based on the detection data of the physical link or the dynamic characteristic data of the inertial measurement module.

[0016] Step S20: After the real-time topology directed acyclic graph is updated, the spatial feature parameters of the building block module in the topology are calculated, the spatial feature parameters are converted into semantic retrieval aliases, and the semantic retrieval aliases are written into the inverted index of the script storage module to complete the index reconstruction.

[0017] Step S30: Parse the external speech audio stream to obtain the core intent text string, use the core intent text string as the query key to query in the reconstructed script storage module, generate an action script candidate set, and simultaneously extract the target topology subgraph bound to each script in the action script candidate set.

[0018] Step S40: Perform injective mapping comparison between the target topology subgraph and the real-time topology directed acyclic graph. If the target topology subgraph is completely contained in the real-time topology directed acyclic graph, output the corresponding action script to the underlying execution control link.

[0019] Step S50: If all target topology subgraphs in the action script candidate set fail the comparison verification, calculate the maximum common subgraph between each target topology subgraph and the real-time topology directed acyclic graph, score and rank the calculation results using preset similarity logic, extract the action script with the highest score and rank, and execute it. At the same time, extract the difference node attributes in the graph data structure to generate an abnormal feedback signal.

[0020] Preferably, in step S10, if the data bus supports a cascading communication protocol, an addressing request is sent to each access port through the data bus, the handshake data packet returned by the bus is parsed to extract the hardware identifier to generate vertex set data, and the link cascading level field and port identifier in the handshake data packet are read to generate edge set data; the vertex set data and edge set data are combined to generate the real-time topology directed acyclic graph.

[0021] Preferably, in step S10, if the data bus only supports a single parallel communication protocol, then the following sub-steps are performed:

[0022] Based on the bus polling results, hardware identifiers are obtained to generate an unordered set of devices. This set is then input into the graph storage module to perform a set inclusion relationship comparison. Standard topology graphs that completely contain the set of vertices are selected to generate candidate topology graph sets.

[0023] A micro-motion detection pulse signal is sent to the actuator-type building block module in the disordered device set, and the inertial measurement module is called within the synchronization time window to obtain the triaxial acceleration and triaxial angular velocity values;

[0024] The triaxial acceleration and triaxial angular velocity values ​​are filtered and denoised, and then concatenated to generate a transient kinematic signature feature vector characterizing the current dynamic response of the organism.

[0025] Calculate the feature matching score between the transient kinematic signature feature vector and the expected dynamic feature vector pre-bound to each candidate topology graph in the candidate topology graph set. Select the candidate topology graph with the largest feature matching score that is greater than the preset matching threshold as the matching benchmark, and extract its prior edge set data to construct the real-time topology directed acyclic graph.

[0026] Preferably, the feature matching score is calculated using cosine similarity logic, specifically by calculating the dot product of the transient kinematic signature feature vector and the expected dynamic feature vector, and dividing by the product of their L2 norms to obtain the corresponding feature matching score value.

[0027] Preferably, in step S20, calculating the spatial characteristic parameters of the building block module includes:

[0028] With the main control unit as the root node, a breadth-first search algorithm is run to traverse the topology links, and the number of shortest path edges from each executor node to the root node is defined as the topology depth data of the node; at the same time, the connection direction relationship of the nodes is read, and the in-degree ratio data of each node is obtained and calculated.

[0029] The in-degree ratio data is calculated as follows: the total number of physical child nodes of each node is obtained as the out-degree data, and the number of parent nodes of each node is obtained as the in-degree data; the out-degree data of each node is divided by the sum of the in-degree data and the preset anti-zero division constant to calculate the corresponding in-degree ratio data.

[0030] Preferably, the step of converting the spatial feature parameters into semantic retrieval aliases includes:

[0031] The topological depth data and in-degree ratio parameters of each node are input into a preset feature matrix mapping rule table for logical comparison; the feature matrix mapping rule table constructs a two-dimensional parameter space grid and pre-allocates and stores a specific set of semantic aliases.

[0032] When the topological depth data and in-degree ratio parameter of a node fall within a specific threshold range, a pre-configured text search term is generated as the semantic search alias.

[0033] Preferably, the step of writing the semantic retrieval alias into the inverted index of the script storage module to complete index reconstruction includes:

[0034] Compare and remove semantic alias data left in the historical index table for the action script corresponding to the evaluated node;

[0035] The newly generated semantic retrieval aliases are dynamically injected into the corresponding inverted index structure table and assigned a query weight identifier that is lower than the weight of the main word of the system's native instruction action word;

[0036] Send a memory refresh command to lock the updated index mapping region.

[0037] Preferably, in step S30, parsing the external speech audio stream to obtain the core intent text string includes: performing speech activity detection and background noise suppression preprocessing on the incoming speech audio stream; calling an automatic speech recognition engine to convert it into an original text sequence; performing word segmentation and part-of-speech tagging, comparing with a stop word dictionary to filter out words without actual control significance, and extracting verbs and specific nouns to generate the core intent text string.

[0038] Preferably, an action script candidate set is generated based on text matching scores, wherein the calculation logic of the relevance score value is as follows: for each lexical feature item contained in the core intent text string, the inverse document frequency weight parameter in the global inverted dictionary is multiplied by the word frequency data appearing in the corresponding action script semantic description, and the product results of all lexical feature items are summed to obtain the relevance score value of the action script.

[0039] Preferably, in step S40, the injective mapping comparison specifically includes: extracting the total number of nodes in the target topology subgraph and comparing it with the total number of nodes in the real-time topology directed acyclic graph for scale pre-detection; if the scale constraint condition is met, then constructing an injective function to verify whether the physical component module type code of the mapped node is consistent with the underlying communication protocol version.

[0040] Preferably, step S40 further includes performing topological verification of the graph structure connection relationship: for any directed edge in the edge set of the target topological subgraph, determine whether the corresponding node pair mapped in the real-time topological directed acyclic graph constitutes a directed edge, and whether the preset communication port number parameter is consistent with the actual physical plug-in port number parameter.

[0041] The above-mentioned injective mapping and connection relationship verification, from the two dimensions of node type and physical port, verifies the actual carrying capacity of the existing physical structure for action scripts, and prevents structural interference caused by module position swapping or incorrect port connection.

[0042] Preferably, in step S50, if the number of nodes included in the calculated maximum common subgraph is less than the set minimum connected node threshold, the target topology subgraph is directly removed to terminate the compensation evaluation; when calculating the maximum common subgraph, the node data of the target topology subgraph that is not included in the maximum common subgraph is extracted to generate a physical missing node difference set.

[0043] Preferably, the calculation logic for the graph structure similarity comprehensive score is as follows: calculate the ratio of the number of nodes and the ratio of the number of directed edges between the optimal maximum common subgraph and the target topological subgraph, respectively, and obtain a weighted basic score by combining the preset node coverage weight coefficient and edge coverage weight coefficient; obtain the preset criticality penalty factor parameters of each missing node in the corresponding physical missing node difference set, and subtract the sum of the criticality penalty factor parameters of all missing nodes from the weighted basic score to obtain the graph structure similarity comprehensive score.

[0044] Preferably, the step of extracting the action script with the highest score ranking result and executing it includes:

[0045] Extract the action script associated with the target topology subgraph that ranks first and has a similarity score greater than the safe execution threshold.

[0046] Intercept the control flow code segment in the action script that depends on the physical missing node difference set, and replace it with a no-operation instruction frame sequence with the same instruction cycle duration as the original instruction and output it to the lower layer.

[0047] The underlying data retrieval dictionary is called to obtain semantic noun entries, and the difference set of physical missing nodes is combined to generate and output feedback prompts related to the degraded action caused by the loss of specific physical parts.

[0048] This invention provides a voice interaction processing system and method for a digital building block robot. It has the following beneficial effects:

[0049] 1. This invention sends micro-motion detection pulse signals to the actuator, uses the inertial measurement module to collect three-axis motion data to generate transient kinematic signature feature vectors, and performs similarity matching calculations with preset maps. Therefore, when the communication capability of the underlying data bus is limited and the cascade relationship cannot be directly read, the actual topological connection form can be deduced by extracting the physical dynamic features of the mechanical structure, thereby improving the system's compatibility with hardware of different underlying communication protocols.

[0050] 2. This invention calculates the topological depth and in-degree ratio parameters of the building block modules by traversing the topological links, uses a threshold grid to transform the above spatial features into specific semantic retrieval aliases, and dynamically reconstructs the inverted index of the control script. Therefore, it can establish a data mapping relationship between abstract graph theory physical parameters and natural language text, and solve the technical problem that after the modular hardware changes its splicing form, the original voice action commands cannot be effectively retrieved and matched because they are no longer in the preset hardware description.

[0051] 3. This invention performs a graph structure injective mapping verification before outputting control commands. When the physical constraint verification fails, it calculates the maximum common subgraph and introduces a criticality penalty factor for missing components to evaluate the similarity score of the compensatory action. Therefore, it can extract alternative action scripts based on the calculation results and replace code segments that depend on missing physical nodes with empty operation instruction frames of equal duration. This maintains the temporal integrity of the multi-axis interpolation motion of the underlying control while avoiding overturning or interference damage caused by forced operation of the machine due to defects. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0053] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0054] Figure 3 This is a schematic diagram of the topology construction process of the present invention;

[0055] Figure 4 This is a schematic diagram of the feature evaluation and index reconstruction process of the present invention;

[0056] Figure 5 This is a schematic diagram of the retrieval and action subgraph extraction process of the present invention;

[0057] Figure 6 This is a schematic diagram of the isomorphic matching verification process of the present invention;

[0058] Figure 7 This is a schematic diagram of the degradation control compensation process of the present invention;

[0059] Figure 8 This is a comparison curve of system performance under different node damage rates according to the present invention;

[0060] Figure 9 This is a scatter plot of the underlying timing state of the degradation control frame in this invention.

[0061] Among them, 10 is the main control unit; 101 is the topology construction module; 102 is the feature evaluation module; 103 is the retrieval module; 1031 isomorphic matching module; 1032 is the subgraph calculation module; 104 is the degradation control module; 11 is the data bus; 12 is the building block module; 13 is the inertial measurement module; 201 is the graph storage module; and 202 is the script storage module. Detailed Implementation

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] See attached document Figure 1 , Figure 1 This is a schematic diagram of the system architecture of a voice interaction processing system for a digital building block robot according to an embodiment of the present invention. The present invention provides a voice interaction processing system for a digital building block robot, including: a main control unit 10, a data bus 11, and multiple building block modules 12.

[0064] The main control unit 10 establishes an electrical connection with multiple modular modules 12 via a data bus 11. The data bus 11 supports data communication and underlying link status detection services. The main control unit 10 integrates an inertial measurement module 13. The modular modules 12 specifically include sensor modules and actuator modules. The main control unit 10 collects the physical connection status of the modular modules 12 and executes control logic.

[0065] The main control unit 10 has multiple logic processing components inside, including:

[0066] Topology building module 101 acquires node identifiers and connection relationship data on data bus 11 and generates a real-time directed acyclic graph in memory.

[0067] The feature evaluation module 102 receives a real-time topological directed acyclic graph, performs graph traversal operations to obtain the topological depth data and in-degree ratio data of each node, and modifies the semantic mapping relationship of the action script accordingly.

[0068] The retrieval module 103 receives and parses the speech audio stream to extract the text string. Internally, the retrieval module 103 includes an isomorphic matching module 1031 and a subgraph calculation module 1032. The isomorphic matching module 1031 performs Boolean conditional filtering operations. The subgraph calculation module 1032 performs graph structure similarity calculations.

[0069] The degradation control module 104 outputs action control commands based on the calculation results of the subgraph calculation module 1032.

[0070] A data storage area is also provided, which includes a graph storage module 201 and a script storage module 202. The graph storage module 201 stores graph data corresponding to preset known hardware assembly configurations. The script storage module 202 stores the control scripts that drive the block module 12 and their bound target topology subgraphs.

[0071] See attached document Figure 2 , Figure 2 This is a schematic flowchart illustrating a voice interaction processing method for a digital building block robot according to an embodiment of the present invention. The present invention provides a voice interaction processing method for a digital building block robot, comprising the following steps:

[0072] S10, the topology construction module 101 periodically polls the data bus 11 and constructs a real-time topology directed acyclic graph representing the current physical hardware state based on the detection data of the physical link or the dynamic characteristic data of the inertial measurement module 13.

[0073] S20, after the real-time topological directed acyclic graph is updated, the feature evaluation module 102 calculates the spatial feature parameters of the building block module 12 in the topological structure, converts the spatial feature parameters into semantic retrieval aliases, and writes the semantic retrieval aliases into the inverted index of the script storage module 202 to complete the index reconstruction.

[0074] S30, the retrieval module 103 parses the external voice audio stream to obtain the core intent text string, uses the core intent text string as the query key to query in the reconstructed script storage module 202, generates an action script candidate set, and simultaneously extracts the target topology subgraph bound to each script in the action script candidate set.

[0075] S40, the isomorphic matching module 1031 performs a single-injection mapping comparison between the target topology subgraph and the real-time topology directed acyclic graph. If the target topology subgraph is completely contained in the real-time topology directed acyclic graph, the isomorphic matching module 1031 outputs the corresponding action script to the underlying execution control link.

[0076] S50, if all target topology subgraphs in the action script candidate set fail the comparison verification by the isomorphic matching module 1031, the subgraph calculation module 1032 calculates the maximum common subgraph between each target topology subgraph and the real-time topology directed acyclic graph, scores and sorts the calculation results using a preset similarity formula, and the downgrade control module 104 extracts the action script with the highest score and sorts it for execution, while extracting the difference node attributes in the graph data structure to generate an abnormal feedback signal.

[0077] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0078] See attached document Figure 3 In this embodiment, the topology construction module 101 executes corresponding data acquisition and graph generation logic based on the differences in the underlying communication capabilities of the data bus 11, so as to output a structured real-time topology directed acyclic graph. This step S10 may include the following sub-steps:

[0079] S101, the topology building module 101 sends addressing requests to each access port through the data bus 11 to read the physical node connection data of the building block module 12. If the data bus 11 supports the cascading communication protocol, the system will parse the handshake data packet returned by the bus, extract the hardware identifier of each building block module 12 to generate vertex set data. At the same time, it reads the link cascading level field and port identifier in the handshake data packet, and extracts the parent-child node connection relationship and the actual physical plug-in port number between the building block modules 12 to generate edge set data.

[0080] Subsequently, the vertex set data and edge set data are combined to generate the real-time topological directed acyclic graph in memory. For the process of hardware handshaking and hierarchical resolution on a data bus supporting a cascading communication protocol, those skilled in the art can use existing automatic addressing technology for communication buses. The communication frame structure and response mechanism are well-known technologies in the field and will not be elaborated upon here.

[0081] S102, as a preferred approach, when faced with limited underlying hardware capabilities, such as when the data bus 11 uses a single parallel communication protocol, the topology building module 101 can only obtain the hardware identifiers of each building block 12 based on the bus polling results, generating an unordered device set. Since the cascading dependency relationships between nodes cannot be directly obtained under this communication link, the topology building module 101 inputs this unordered device set as a retrieval query parameter into the graph storage module 201. Furthermore, the graph storage module 201 can pre-store multiple sets of standard topology graph data structures corresponding to known hardware assembly forms. The further data comparison operation process is as follows: the topology building module 101 performs a set inclusion relationship comparison operation, filtering out all standard topology graphs in the graph storage module 201 whose vertex sets completely contain the unordered device set, thereby generating a candidate topology graph set. To ensure the security of system operation, if the candidate topology set is empty after traversal and comparison, it indicates that the current hardware connection belongs to an unknown non-standard assembly form. The system will stop topology reconstruction and trigger a hardware reassembly alarm. If there is a valid candidate matching result, it will proceed to the subsequent verification steps.

[0082] S103, in order to pinpoint the unique true physical form among multiple potential candidate structures, the topology construction module 101 calls the main control unit 10 to send a micro-motion detection pulse signal to the actuator-type building block module in the disordered device set. This micro-motion detection pulse signal is defined as a motor drive level signal with a fixed duration and duty cycle, used to excite the mechanical transient response of the physical body. From a physical principle perspective, due to the significant differences in the mass distribution and moment of inertia corresponding to different topologies, the same drive pulse will induce drastically different micro-vibration characteristics of the system. Based on this principle, within the synchronization time window of the micro-motion detection pulse signal, the inertial measurement module 13 acquires the three-axis acceleration and three-axis angular velocity values ​​in the physical space of the main control unit 10. The topology construction module 101 records the acquired numerical sequence and performs filtering and noise reduction operations, splicing and assembling the processed numerical sequence to generate a transient kinematic signature feature vector characterizing the current dynamic response of the body.

[0083] The topology construction module 101 reads the pre-bound expected dynamic feature vectors of each candidate topology graph in the candidate topology graph set. Then, the system uses a cosine similarity algorithm to calculate the feature matching score between the transient kinematic signature feature vector and each expected dynamic feature vector. The specific similarity formula is as follows:

[0084] ;

[0085] In the formula, Indicates the first The feature matching scores of the candidate topological graphs range from 1 to 10. between; Represents the first in the candidate topology set One graph element object; This represents the transient kinematic signature feature vector generated by actual data acquisition. The graph storage module 201 indicates that the graph is for the first... A candidate topological graph with pre-defined expected dynamic eigenvectors; This represents the result of the dot product operation between the transient kinematic signature eigenvector and the expected dynamic eigenvector; The L2 norm represents the eigenvector of the transient kinematic signature. The L2 norm represents the expected dynamic eigenvector.

[0086] The topology construction module 101 compares the feature matching scores of each candidate topology map and selects the candidate topology map with the largest calculated score that is greater than a preset matching threshold as the matching benchmark for the current physical hardware state. Here, the preset matching threshold can be defined by those skilled in the art in combination with the sensor environmental noise baseline of the inertial measurement module 13 and a large amount of prior calibration experimental data, aiming to filter out forced mismatches caused by environmental interference.

[0087] The topology construction module 101 extracts the prior edge set data corresponding to the matching benchmark, uses the prior edge set data as supplementary features of node connection relationships, integrates it with the aforementioned unordered device set, and finally uses the derived edge connection relationships to construct a complete real-time topology directed acyclic graph in memory. If the comparison finds that the feature matching scores of all candidate topology graphs do not reach the above-mentioned preset matching threshold, the system also considers the recognition failure and outputs a topology parsing anomaly signal to avoid the drive control layer falling into an unpredictable execution state.

[0088] See attached document Figure 4 In this embodiment, the feature evaluation module 102 converts the physical layer hardware connection location features into natural language layer text search terms, establishing a data mapping relationship between the physical space architecture and the semantic recognition dictionary to solve the problem of instruction mismatch after the modular hardware changes its splicing form. This step S20 may include the following sub-steps:

[0089] S201, the feature evaluation module 102 acquires the real-time topological directed acyclic graph and performs graph traversal operations on the actuator-type building block modules in the graph structure. To quantify the relative position attributes of each node in the physical space of the machine, the feature evaluation module 102 uses the main control unit 10 as the root node, runs a breadth-first search algorithm to traverse the topological links, calculates the number of shortest path edges from each actuator node to the root node, and defines this number of edges as the topological depth data of the node. This topological depth data is used to characterize the topological extension distance of the actuator physical entity from the power and control core.

[0090] To ensure the stability of the traversal algorithm, the system synchronously records the set of visited nodes during breadth-first search. After traversal, if any nodes remain unvisited, the system determines these nodes to be isolated nodes with poor electrical connections or detached buses, and removes them from the feature evaluation queue to prevent null value anomalies. For normally connected nodes, the feature evaluation module 102 reads the node connection direction relationships in the directed acyclic graph, obtains and calculates the in-degree ratio data for each node. The specific formula for the in-degree ratio is:

[0091] ;

[0092] In the formula, Represents a node The in-degree / out-degree ratio parameter; Represents the i-th element in the real-time topological directed acyclic graph. One executor node object; Represents a node The out-degree data refers to the total number of child nodes that are physically directly cascaded and attached to the downstream of this node. Represents a node The in-degree data, that is, the number of parent nodes located upstream of this node; To prevent division by a minimum constant, as a preferred method, The value range can be set to 10. -5 Up to 10 -3 The in-degree ratio is used to prevent arithmetic overflow exceptions caused by a node having an in-degree of zero. The in-degree ratio parameter reflects the relative hub function of the node entity in the local topology connectivity.

[0093] S202, the feature evaluation module 102 performs a cross-domain transformation from graph theory physical parameters to semantic retrieval text dimensions based on the spatial feature parameters calculated above. In specific engineering implementations, a feature matrix mapping rule table is pre-installed in the system memory. This table defines the correlation between the value range of feature parameters and the corresponding spatial natural language attributes. The feature evaluation module 102 inputs the topological depth data and in-degree ratio parameters of each node into this rule table for logical comparison.

[0094] When the spatial parameters of a node fall within a specific threshold range, a pre-configured text search term is generated. For example, if the system identifies a certain node... A value of 0 and a topology depth greater than a preset depth threshold parameter indicate that the component does not bear the subsequent load structure and is located at the physical extension end. The system then generates a semantic retrieval alias combination representing the spatial location of the end, such as the action pointing to the end of the word, the front end, or the grasping part. Similarly, if a node's... If the value exceeds the preset hub distribution threshold, the system will generate a semantic retrieval alias representing the chassis or main base. The preset depth threshold parameter mentioned here can be dynamically configured to two-thirds of the current maximum topology depth of the entire machine, while the preset hub distribution threshold can be set to a constant of 3. This value is based on engineering statistics of the number of conventional interfaces of the building block modules, and can cover most building block assembly forms while ensuring recognition and differentiation.

[0095] In one embodiment, the pre-set feature matrix mapping rule table essentially constructs a two-dimensional parameter space grid in memory. Its row indices correspond to the discretized hierarchical intervals of the topology depth data, and its column indices correspond to the numerical range of the in-degree ratio parameter. Each grid intersection pre-allocates and stores a specific set of semantic aliases. To establish accurate data mapping, the feature evaluation module 102 addresses specific grid cells in the rule table based on the input parameter coordinates, and then extracts the associated text data. For example, when the feature matrix mapping rule table receives a node whose in-degree ratio parameter falls within the minimum range of 0 to 0.1, and the calculated topology depth data is between 70% and 100% of the maximum depth of the entire machine, the system parsing logic determines that the physical node belongs to a topology endpoint that does not support downstream branches. Based on this determination, the feature evaluation module 102 extracts semantic retrieval aliases such as endpoint, front end, and operation unit from the corresponding grid cells.

[0096] As a preferred approach, if the input parameters show that the topology depth data of a node is a constant 1 (i.e., directly mounted on the main control unit 10 on the physical line), and the in-degree ratio parameter is greater than or equal to 3.0, it indicates that the node undertakes the function of a multi-path cascaded communication distribution hub. The feature matrix mapping rule table will then match and output location terms such as base, main trunk base, or central load-bearing part. Furthermore, for nodes whose parameter coordinates are in the intermediate transition range, such as those with an in-degree ratio parameter between 0.8 and 1.2 and a topology depth in the middle range, the system determines them as linear transmission components and extracts and outputs relay feature terms such as joints or extension arms. Through the clear definition of the above multi-dimensional numerical ranges, the rule table can materialize abstract graph theory mathematical parameters into natural language spatial attributes that conform to real physical forms, providing substantial data conversion support between the underlying hardware status and the upper-level speech retrieval dictionary.

[0097] S203, to maintain the accuracy of the instruction retrieval system and avoid semantic conflicts introduced by hardware reconfiguration, the feature evaluation module 102 further extracts the configuration files of each action script bound to the evaluated node in the system and accesses the underlying data retrieval dictionary of the script storage module 202. The system first performs an erasure operation, comparing and clearing the spatial semantic alias data left in the historical index table for the action script corresponding to the current node, thereby releasing the old form of memory mapping.

[0098] After completing the historical cleanup, the feature evaluation module 102 dynamically injects the newly generated semantic retrieval aliases as supplementary terms into the inverted index structure table corresponding to the action script. During the process of writing new terms into the inverted index table, the feature evaluation module 102 simultaneously assigns corresponding query weight identifiers to these semantic retrieval aliases. To prevent aliases from triggering misidentification, this query weight identifier is configured to be lower than the main term weight of the system's native instruction action word. After the data writing is complete, the system sends a memory refresh instruction to the storage interface controller, submits and locks the updated index mapping area, ensuring that subsequent audio intent retrieval services can read the latest semantic state. For the underlying storage addressing control of the inverted index table and the read / write security lock mechanism under concurrent conditions, those skilled in the art can use existing key-value database memory management technologies. The data block refresh strategy is a well-known technology in the field and will not be elaborated here.

[0099] See attached document Figure 5 In this embodiment, the retrieval module 103 converts unstructured speech signals into internal graph structure data. During this process, the system needs to parse externally input natural language instructions and define executable control scripts and their associated physical graph theory constraints within the reconstructed storage space. This step S30 may include the following sub-steps:

[0100] S301, the retrieval module 103 acquires an external speech audio stream through the audio receiving interface, performs speech activity detection and background noise suppression preprocessing on the incoming speech audio stream, and then extracts audio segments containing effective speech features. Subsequently, the system internally uses an automatic speech recognition engine to convert the preprocessed audio segments into the original text sequence.

[0101] For the original text sequence, the retrieval module 103 calls the built-in lexical parser to perform word segmentation and part-of-speech tagging. In this step, by comparing with a preset stop word dictionary, modal particles and modifying words without actual control meaning in the original text sequence are filtered out. After the above filtering operation, lexical units labeled as verbs and specific nouns are extracted, and these lexical units are combined to generate the core action command text string. For the acoustic feature extraction of the speech audio stream and the construction of the automatic speech recognition engine, those skilled in the art can use existing hidden Markov models, and its audio transcription logic is a well-known technology in the field, which will not be described in detail here.

[0102] S302, after obtaining the core action command text string, the retrieval module 103 uses it as the query key to access the updated inverted index table in the script storage module 202. To ensure that the command intent effectively reaches the relevant control action, the retrieval module 103 calculates the text matching score between the core action command text string and each action script entry in the inverted index. From the perspective of text retrieval principles, this calculation process aims to evaluate the degree of overlap between the entity words converted from the user's speech and the preset semantic description of the action script, while using inverse document frequency features to weaken the interference of high-frequency common words on the matching results. The specific text matching score calculation formula is as follows:

[0103] ;

[0104] In the formula, Indicates the first The relevance score of each action script to the core action instruction text string; This represents the set of words contained in the core action instruction text string extracted after lexical analysis; This indicates the first in script storage module 202 One ActionScript object; This represents a single lexical feature item extracted from the core action instruction text string through traversal and segmentation. Representing lexical features Inverse document frequency weight parameter in the system's global inverted index; Representing lexical features In the Frequency data of words appearing in the semantic description of an action script.

[0105] Considering that topological semantic alias entries were dynamically injected into the index in step S20, the system no longer performs static corpus word frequency statistics on these newly injected alias entries, but instead directly uses them... The value is configured as a preset constant weight, while the native instruction vocabulary... The data remains unchanged, continuing to use the offline action corpus data loaded during system initialization.

[0106] The retrieval module 103 iterates through all action script objects to obtain their corresponding relevance scores. Based on this, the system filters and removes action script objects whose relevance scores are lower than a preset text matching threshold parameter, and then aggregates the remaining action script objects that meet the criteria to generate an action script candidate set.

[0107] As a preferred approach, the aforementioned text matching threshold parameter can be dynamically and adaptively adjusted based on the total number of feature words contained in the core action command text string. It is typically set to 60% of the theoretical maximum total matching score of all words in the current command text string to accommodate the possibility of missing words in spoken language. If, after traversal calculation, it is found that the relevance scores of all action script objects do not reach the preset text matching threshold parameter, indicating that there is no matching semantic in the current dictionary, the system will output an error feedback signal indicating that the voice command cannot be recognized and terminate the current round of interactive data processing.

[0108] S303, for each action script object included in the action script candidate set, the retrieval module 103 further triggers the internal data extraction linkage mechanism. Since the script storage module 202 uses relational mapping in its underlying data structure, each action script is independently configured with a target topology subgraph. The retrieval module 103 can then synchronously initiate a read request to the script storage module 202 based on the unique index code of each script in the candidate set, instantiating the corresponding target topology subgraph set in the system's running memory. This target topology subgraph defines the minimum hardware component topology structure indispensable for the successful execution of the bound action script, and its data structure contains the necessary component type attributes and hardware edge connection relationships. Finally, these extracted target topology subgraphs and the corresponding action scripts are jointly stored in the system's running queue as benchmark reference data for subsequent implementation of physical Boolean constraint filtering.

[0109] See attached document Figure 6 In this embodiment, the isomorphic matching module 1031 is responsible for establishing the constraint relationship between physical hardware and execution instructions at the graph theory structure level, so as to determine whether the current block assembly form fully meets the minimum hardware configuration required to execute the candidate action script. This determination process is mathematically transformed into seeking the injective mapping relationship between the target topological subgraph and the real-time topological directed acyclic graph. This step S40 may include the following sub-steps:

[0110] S401, the isomorphic matching module 1031 sequentially extracts each target topology subgraph from the system's running queue and compares it with the real-time topology directed acyclic graph representing the current physical hardware state, performing node-level attribute comparison. To improve algorithm execution efficiency and avoid unnecessary computational consumption, the isomorphic matching module 1031 performs a scale pre-check before performing a depth comparison, extracting the total number of nodes in the target topology subgraph and comparing it with the total number of nodes in the real-time topology directed acyclic graph.

[0111] If the target number of nodes is determined to be greater than the actual number of nodes, a fast pruning mechanism is directly triggered to determine that the current subgraph does not match; otherwise, if the size constraint condition is met, the isomorphic matching module 1031 attempts to construct an injective function. ,in The set of nodes representing the target topological subgraph. This represents the set of nodes in a directed acyclic graph (DAG) of the real-time topology. As a preferred approach, for any node in the target topology subgraph... Mapping nodes found in a real-time topological directed acyclic graph The consistency condition of node attributes must be met. Specifically, this consistency condition means that the module type code of the physical component and the version of the underlying communication protocol must be identical. If this condition cannot be met, it indicates that there are differences in the hardware of the current physical components, and the mapping attempt will fail. For the conventional backtracking search operation in graph node matching, those skilled in the art can use the existing VF2 algorithm. Its state space tree pruning strategy is a well-known technique in the field and will not be elaborated upon here.

[0112] S402, on the basis of ensuring the consistency of node attributes, the isomorphic matching module 1031 further performs topological verification of the graph structure connection relationship.

[0113] From an engineering physics perspective, even if two building block modules have perfectly matched hardware types, errors in their installation location, cascading sequence, or physical connection ports on the overall system can still lead to interference or loss of control of the mechanical structure if drive commands are directly issued. Therefore, the system needs to verify whether the hardware edge connection relationships in the target topology subgraph are accurately reproduced on the physical machine.

[0114] For the edge set of the target topological subgraph any directed edge in The isomorphic matching module 1031 determines the mapping on the edge set of the real-time topological directed acyclic graph. Corresponding node pairs in Does it also constitute a directed edge? Simultaneously, the system extracts the physical interface attribute features bound to the directed edge for secondary verification. The logical judgment condition set by the system here is: if... Then it should satisfy And the side The corresponding preset communication port number parameter needs to be consistent with the edge. The actual physical plug-in port number parameter returned remains consistent. The isomorphic matching module 1031 determines that the target topology subgraph is completely contained in the real-time topology directed acyclic graph if and only if the aforementioned node attribute conditions and the edge connection conditions here are both satisfied, and the isomorphic condition verification corresponding to the current action script is passed.

[0115] In step S403, for multiple action script objects extracted from the action script candidate set, the isomorphic matching module 1031 determines the final control link direction through a state arbitration mechanism. When at least one target topology subgraph passes the above isomorphic verification, the system changes the Boolean matching flag of its internal running state from the default suspended interception state to the matching and allowing state. Considering that hardware redundancy may lead to multiple target topology subgraphs passing verification simultaneously, to avoid concurrent execution conflicts, the isomorphic matching module 1031 sorts the text matching scores calculated in step S30 in reverse order, and prioritizes the unique item with the highest relevance score as the arbitration result. Subsequently, the isomorphic matching module 1031 extracts the action script corresponding to the verified target topology subgraph from the stack and outputs it to the underlying execution control link, whereby the main control unit 10 drives each block module 12 to operate according to the instruction timing.

[0116] Conversely, if, after traversal and comparison, all target topological subgraphs in the action script candidate set fail to establish a valid injective mapping function, it indicates that the current physical entity has experienced component incompleteness or hardware loss, failing to meet the minimum safe hardware constraints for intended execution. In this case, the isomorphic matching module 1031 will intercept the direct release of the action script and transfer the abnormal control flow state of the failed match to subsequent modules to trigger the system's degradation compensation processing branch based on graph structure approximation.

[0117] See attached document Figure 7 In this embodiment, to address the failure of precise matching verification due to missing physical components or abnormal connection patterns, the subgraph calculation module 1032 performs approximate dimensionality reduction operations at the graph structure level to find feasible alternative execution strategies on the current incomplete machine, thus maintaining the robustness of the human-computer interaction system. This step S50 may include the following sub-steps:

[0118] S501, after receiving the verification failure status signal from the isomorphic matching module, the subgraph calculation module 1032 will traverse each target topology subgraph in the action script candidate set. For each target topology subgraph, the system will cross-compare it with the real-time topology directed acyclic graph representing the current hardware state and calculate the maximum common subgraph between the two. The maximum common subgraph is defined as the graph model entity that exists in both graph structures and contains the largest number of graph nodes.

[0119] During this operation, the subgraph calculation module 1032 simultaneously performs a topology subtraction operation, extracting node data that exists in the target topology subgraph but is not included in the maximum common subgraph, generating a physical missing node difference set. This physical missing node difference set identifies the specific hardware location that prevents the current instruction from being fully executed. Considering the actual executability of the compensation action, if the number of nodes included in the calculated maximum common subgraph is less than the minimum connected node threshold set by the system (e.g., only a single root node remains), it indicates that the candidate action has lost its basic hardware execution carrier, and the system will directly eliminate the target topology subgraph and terminate its subsequent compensation evaluation. For the enumeration and backtracking search solution of the maximum common subgraph, those skilled in the art can use the existing McGregor algorithm, whose graph theory state space tree search mechanism is a well-known technology in the field and will not be described in detail here.

[0120] S502, to quantitatively evaluate the compensatory execution capability of the incomplete machine for each candidate action script, the subgraph calculation module 1032 introduces a multi-dimensional graph structure similarity scoring algorithm. Since a single comparison of node counts cannot truly reflect the constraints of mechanical control, this scoring algorithm comprehensively considers structural coverage and the criticality penalty effect of missing hardware. The specific formula for calculating the comprehensive graph structure similarity score is as follows:

[0121] ;

[0122] In the formula, This represents the overall similarity score for a specific target topological subgraph. This represents the total number of nodes contained in the generated optimal maximum common subgraph; This represents the total number of nodes preset in the target topology subgraph; This represents the total number of directed edges that maintain valid connections in the optimal maximum common subgraph; This indicates the total number of directed edges preset in the target topological subgraph; Minimal normal quantities to prevent arithmetic anomalies caused by an empty edge set in the target structure; and These represent the node coverage weight coefficient and the edge coverage weight coefficient, respectively.

[0123] As a preferred approach, the sum of the two values ​​is set to 1. Given the difficulty in establishing physical connections and the fact that the constraint on the mechanism's motion is usually greater than that of a single node, it is advisable to... The value is configured to be higher than constants, for example Take 0.3, Take 0.7. This represents the difference set of the previously generated physical missing nodes; Represents a single missing node object in the difference set; This indicates that the system pre-configuration is assigned to the node. The criticality penalty factor parameter. This criticality penalty factor parameter is assigned based on the functional level of the node in the body's dynamic transmission chain, and its value is usually set between 0.05 and 0.5. Chassis-type nodes that bear core torque or play a major supporting role will be matched with a higher penalty factor value, while decorative nodes at the end of the topology will have a relatively lower value.

[0124] The subgraph calculation module 1032 performs the above calculations for each action script in the candidate set and sorts them in descending order according to the calculated comprehensive similarity score, generating a script ranking queue representing the priority of alternative execution. Subsequently, the subgraph calculation module 1032 synchronously submits the script ranking queue and the associated physical missing node difference set data to the degradation control module 104 to trigger the subsequent underlying compensation control flow.

[0125] S503, based on the script ranking queue generated above, the degradation control module 104 receives data and executes degradation action control and feedback arbitration. To prevent the robot from overturning due to forced movement in the extreme state of loss of core support structure, the system sets a safety execution threshold. This safety execution threshold can be pre-assessed and set by those skilled in the art based on the overall center of gravity distribution characteristics and fault tolerance safety margin, and its value is not lower than 0.4 under normal operating conditions.

[0126] The degradation control module 104 extracts the target topology subgraph that is at the top of the ranking queue and has a similarity score greater than the lower limit of the safe execution threshold, and determines the action script bound to it as the alternative action script. Before the alternative action is issued, the degradation control module 104 automatically intercepts control flow code segments in the action script that directly call or depend on the difference set of physically missing nodes. In order to maintain the timing integrity of the underlying hardware multi-axis interpolation motion and avoid timer errors, the degradation control module 104 replaces these intercepted code segments with no-operation instruction frames with a cycle length that is exactly the same as the original instruction. The processed safe execution frame sequence is then output to the underlying master control driver for execution.

[0127] Based on the actions sent in the control frame sequence, the degradation control module 104 parses the hardware device type tags attached to the physical missing node difference set, and calls the semantic noun entries obtained from the underlying data retrieval dictionary of the script storage module 202 to generate a human-computer interaction abnormality alarm signal with a clear direction. The degradation control module 104 then outputs feedback prompts to the external display interface regarding the specific physical part missing that caused the action degradation. If, after traversing the ranking queue, the degradation control module 104 finds that the item with the highest similarity score is still below the aforementioned safe execution threshold, it indicates that the current hardware damage level has exceeded the physical boundary that the algorithm can compensate for. At this time, the degradation control module 104 will block all driver sending links, directly triggering a global error alarm signal for a complete system failure, thereby ensuring the absolute safety of hardware operation while attempting fault-tolerant recovery.

[0128] To enable those skilled in the art to more clearly understand the purpose, technical solution, and advantages of this invention, the present invention will be further described in detail below with reference to specific application embodiments, real experimental test data, and corresponding drawings. It should be noted that the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0129] As a specific application embodiment of the present invention, suppose a user assembles an engineering exploration vehicle containing a main control unit 10, with six modular modules 12 mounted on its data bus 11, specifically including four drive wheels for propulsion motors, one gimbal base, and one robotic arm gripper. After physical assembly and power-on, the topology construction module 101 automatically scans the underlying physical links and successfully constructs a real-time directed acyclic graph with tree-like branching characteristics in memory. Based on this, the feature evaluation module 102 calculates through breadth-first search that the node containing the robotic arm gripper has an out-degree of zero and the maximum topology depth. According to the built-in feature matrix mapping rules, the system determines that the component is at the physical extension end, and then dynamically generates semantic retrieval aliases such as front end and robotic arm, and uses them as constants. The weighted terms are injected into the inverted index corresponding to the action script.

[0130] In normal operation, when the user issues a voice command to control the front end to grasp an obstacle in front, the retrieval module 103 extracts the core intent text string, such as "front end" and "grab". Since physical topological features have been dynamically injected into the inverted index, the system can accurately match it with the action script for gripper closure. Subsequently, the isomorphic matching module 1031 verifies that the current physical map is completely consistent with the target topological subgraph, and successfully issues the drive command, allowing the robot to execute a precise grasping action.

[0131] When performing a complex terrain exploration task, suppose the right front wheel of the exploration vehicle is physically detached and disconnected due to an accidental collision. At this time, the user issues a voice command to move forward at full speed, and the retrieval module 103 normally matches a candidate set of forward action scripts that require coordinated operation of all four wheels. However, when the isomorphic matching module 1031 attempts to construct an injective mapping function, it finds that the right front wheel node required by the target topological subgraph has been lost in the real-time topological directed acyclic graph. The absolute consistency condition of node attributes and edge connections is broken, and the direct issuance of the action is intercepted by the system.

[0132] Next, the subgraph calculation module 1032 takes over the abnormal control flow, traverses and calculates the maximum common subgraph between the current incomplete robot and the target topology subgraph, and accurately extracts the physical missing node difference set for the calibrated right front wheel. After substituting into the multi-dimensional graph structure similarity scoring formula, the system calculates the current robot's overall similarity score to be 0.76. Since this value is significantly higher than the system's preset safe execution threshold of 0.4, the degradation control module 104 determines that the current robot has the conditions for compensatory operation. To prevent the underlying timer from malfunctioning, the degradation control module 104 precisely intercepts the PWM control flow code originally intended to be sent to the right front wheel and replaces it with an idle operation instruction frame of equal duration. The action sequences of the other three drive wheels are synchronously sent while maintaining the original interpolation timing. At the same time, the system parses the missing node difference set and outputs an abnormal alarm signal to the external display interface indicating that the right front wheel connection is lost and the robot has switched to a three-wheeled degradation driving mode. Through the above closed-loop processing, the robot avoids control flow collapse caused by missing components and safely completes the degradation compensation of the intended action in the incomplete state.

[0133] To further verify the technical effectiveness of the embodiments of the present invention in dealing with topological changes and hardware failures, a hardware-in-the-loop simulation test platform was built using MATLAB-R2025a software, and related experimental comparative analysis was carried out. The experiments conducted multi-dimensional performance tests on the dynamic topology-aware fault-tolerant scheme proposed in this invention (hereinafter referred to as the "scheme") and the traditional static keyword matching system (hereinafter referred to as the "control group").

[0134] The test results are as follows:

[0135] according to Figure 8It can be seen that, in the dual Y-axis comparison of system recognition accuracy and compensation success rate under different node damage rates, when the block robot is in a node-complete state, the voice command intent recognition accuracy of both the proposed solution and the control group can maintain above 98%. As the proportion of missing nodes gradually increases, the control group suffers from a large number of deadlocks in the underlying control flow due to the lack of dynamic feedback with physical constraints. Its action compensation execution success rate drops sharply to below 15% when the node loss reaches 20%. In contrast, the proposed solution, thanks to the fault-tolerant degradation mapping mechanism of the maximum common subgraph, can still maintain a recognition accuracy of 82.3% and a compensation success rate of 76.5% even under the severe condition of a node loss rate of 40%, through dynamic feature injection and no-operation placeholders. The gentle decay trend of the curve demonstrates the advantage of the proposed scoring algorithm in maintaining system robustness.

[0136] according to Figure 9 As can be seen from the scatter plot of the underlying timing state of the degradation control frame, the multi-axis interpolation timing of the main control unit precisely intervenes in the underlying execution logic. For channel 2, where the physically detached right front wheel is located, its original pulse width modulation drive code has been completely replaced with a stable low-level no-operation instruction sequence. Meanwhile, the drive frames of the undamaged channels 1, 3, and 4 strictly maintain their original phase difference and period synchronization law. These timing waveform characteristics confirm that while eliminating abnormal defects, the system avoids concurrent uncontrollable issues caused by timing misalignment in other normal drive channels, ensuring the safety of the degradation compensation mechanism and the completeness of the algorithm logic from an electrical control perspective.

[0137] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A voice interaction processing system for a digital building block robot, characterized in that, include: The system includes a main control unit, a data bus, multiple modular components, and a data storage area. The main control unit establishes an electrical connection with the multiple modular components through the data bus. The main control unit includes: The topology construction module is used to obtain the physical node connection data on the data bus and construct a real-time topology directed acyclic graph that represents the current physical hardware state. The feature evaluation module is used to calculate the spatial feature parameters of each building block based on the real-time topological directed acyclic graph, and convert the spatial feature parameters into semantic retrieval aliases to reconstruct the inverted index in the data storage area; The retrieval module is used to parse the externally input speech audio stream to extract the core intent text string, and combine it with the reconstructed inverted index to match and generate an action script candidate set and a bound target topology subgraph; The isomorphic matching module is used to verify the injective mapping relationship between the target topology subgraph and the real-time topology directed acyclic graph, and output the corresponding action script to the execution link when the verification is successful; The subgraph calculation module is used to calculate the maximum common subgraph between the target topological subgraph and the real-time topological directed acyclic graph and perform similarity scoring when the injective mapping relationship verification fails. The degradation control module is used to extract alternative action scripts based on the ranking results of the similarity scores for degradation execution, and to extract the difference nodes in the graph data structure to generate anomaly feedback signals.

2. The voice interaction processing system for the digital building block robot according to claim 1, characterized in that, The data storage area includes a graph storage module; the main control unit also integrates an inertial measurement module; when the data bus adopts a single parallel communication protocol, the topology construction module obtains the hardware identifiers of each building block module through the data bus to generate an unordered device set, and filters out standard topology graphs in the graph storage module whose vertex set contains the unordered device set to generate a candidate topology graph set; The topology construction module calls the main control unit to send micro-motion detection pulse signals to the actuator-type building block modules in the disordered device set, and uses the inertial measurement module to obtain triaxial acceleration and triaxial angular velocity values ​​to generate transient kinematic signature feature vectors; calculates the feature matching score between the transient kinematic signature feature vector and the expected dynamic feature vectors pre-bound to each candidate topology graph, selects the candidate topology graph with the largest feature matching score that is greater than the preset matching threshold as the matching benchmark, and extracts the corresponding prior edge set data to construct the real-time topology directed acyclic graph in memory.

3. The voice interaction processing system for the digital building block robot according to claim 1, characterized in that, The spatial feature parameters include topology depth data and in-degree ratio data. The feature evaluation module runs a breadth-first search algorithm to traverse the topology links with the main control unit as the root node, and calculates the number of shortest path edges from each actuator node to the root node, which is defined as the topology depth data. At the same time, the feature evaluation module reads the node connection direction relationship in the directed acyclic graph and calculates the ratio of the out-degree data to the in-degree data of each node as the in-degree ratio data.

4. The voice interaction processing system for the digital building block robot according to claim 3, characterized in that, The feature evaluation module inputs the topology depth data and the in-degree ratio data into a preset feature matrix mapping rule table for logical comparison. When the spatial parameters of a node fall into a specific threshold range, the corresponding text search terms are extracted to generate the semantic search alias. After the feature evaluation module clears the data left in the historical index table for the action script corresponding to the current node, it configures the semantic search alias as a supplementary term and assigns a query weight identifier, and dynamically injects it into the inverted index.

5. The voice interaction processing system for the digital building block robot according to claim 1, characterized in that, The data storage area includes a script storage module for storing action scripts and bound target topology subgraphs. The retrieval module performs speech activity detection and background noise suppression preprocessing on the input external speech audio stream, converts it into an original text sequence using an automatic speech recognition engine, and performs word segmentation and part-of-speech tagging to filter out stop words, thereby extracting the core intent text string. The retrieval module uses the core intent text string as a query key, calculates the text matching score with each action script by combining the inverse document frequency weight parameter and word frequency data, and eliminates action scripts with text matching scores lower than a preset text matching threshold parameter to generate the action script candidate set in the script storage module.

6. The voice interaction processing system for the digital building block robot according to claim 1, characterized in that, The isomorphic matching module extracts the total number of nodes in the target topology subgraph and compares it with the total number of nodes in the real-time topology directed acyclic graph. After meeting the scale constraint, it constructs an injective function from the target topology subgraph node set to the real-time topology directed acyclic graph node set, sequentially verifying the consistency of module type code and underlying communication protocol version, and checking whether the preset communication port number parameter corresponding to the edge connection relationship is consistent with the actual physical plug-in port number parameter. When all conditions are met, it is determined that the injective mapping relationship has been verified.

7. The voice interaction processing system for the digital building block robot according to claim 1, characterized in that, The subgraph calculation module extracts node data that exists in the target topology subgraph but is not included in the maximum common subgraph, and generates a physical missing node difference set; During the similarity scoring process, the subgraph calculation module integrates the node coverage weight of the number of nodes in the maximum common subgraph relative to the preset total number of nodes in the target topology subgraph, the edge coverage weight of the number of directed edges in the maximum common subgraph relative to the preset total number of directed edges in the target topology subgraph, and deducts the criticality penalty factor parameter pre-configured for each node in the physical missing node difference set, calculates the comprehensive similarity score, and generates a script ranking queue by sorting them in descending order.

8. The voice interaction processing system for the digital building block robot according to claim 7, characterized in that, The degradation control module extracts the first target topology subgraph of the script ranking queue and whose comprehensive similarity score is greater than the bottom line of the safe execution threshold. The bound action script is determined as the alternative action script. The degradation control module intercepts the control flow code segment in the alternative action script that depends on the difference set of the physical missing nodes, replaces it with a no-operation instruction frame with the same duration as the original instruction cycle, and outputs it to the underlying master control driver for execution.

9. The voice interaction processing system for the digital building block robot according to claim 8, characterized in that, The degradation control module parses the hardware device type tags attached to the physical missing node difference set, and calls the semantic noun entries obtained from the data retrieval dictionary to generate directional prompt information, which is output to the external display interface as the abnormal feedback signal.

10. A voice interaction processing method for a digital building block robot, applied to a voice interaction processing system for a digital building block robot as described in any one of claims 1-9, characterized in that, Includes the following steps: The topology building module obtains physical node connection data on the data bus and constructs a real-time directed acyclic graph representing the current physical hardware state. The feature evaluation module calculates the spatial feature parameters of each building block based on the real-time topological directed acyclic graph, and converts the spatial feature parameters into semantic retrieval aliases to reconstruct the inverted index in the data storage area; The retrieval module parses the externally input speech audio stream to extract the core intent text string, and combines it with the reconstructed inverted index to generate a candidate set of action scripts and a bound target topology subgraph; The isomorphic matching module verifies the injective mapping relationship between the target topology subgraph and the real-time topology directed acyclic graph, and outputs the corresponding action script to the execution link when the verification is successful; If the injective mapping relationship verification fails, the subgraph calculation module calculates the maximum common subgraph between the target topology subgraph and the real-time topology directed acyclic graph and performs similarity scoring. The degradation control module extracts alternative action scripts based on the ranking results of the similarity scores for degradation execution, and extracts difference nodes in the graph data structure to generate abnormal feedback signals.

Images