Digital twin control method for remotely operated robots integrating biomedical cleaning standards

Abstract

This invention relates to the field of intelligent manufacturing equipment technology, specifically to a digital twin control method for teleoperated robots that integrates biopharmaceutical cleaning standards. The invention determines spatial constraints based on the distance between the robot's current end-effector position and a restricted boundary, sequential constraints based on the current area and cleaning standards, and temporal constraints based on the time interval between historical cleaning operations and the current time, as well as the distance between the operation position and the current end-effector position. This comprehensive approach determines the intensity of regulatory conflicts and the dominant direction of constraints at the robot's end-effector. Then, by combining this with the robot's desired end-effector speed, the probability of operational conflicts is determined to intervene and correct the robot's operational commands. This invention constructs a digital twin dynamic field integrating three-dimensional constraints of space, sequence, and time, and improves the compliance and effectiveness of digital twin control for teleoperated robots in biopharmaceutical workshops by quantifying the probability of conflicts between operational intentions and cleaning standards.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing equipment technology, specifically to a digital twin control method for remotely operated robots that integrates biomedical cleaning standards. Background Technology

[0002] As a typical application of intelligent manufacturing equipment, remote-controlled robots in biopharmaceutical workshops mainly use sensors to achieve geometric and state synchronization between virtual and real scenes. Operators input the desired pose and contact force through a remote control console such as a control handle. The digital twin system maps the position, posture, and environmental point cloud of the physical robot to the virtual end in real time, realizing remote control mapping and geometric boundary obstacle avoidance.

[0003] However, existing digital twin systems only synchronize physical quantities and cannot embed biomedical cleaning standards (such as wiping sequence, disinfectant residence time, and prohibition of cross-area operation) into the control loop in real time. It is difficult to quantify and predict the probability of conflict between the operator's movement tendency and the standard requirements. Moreover, when the operator's teleoperation intention conflicts with the standard requirements, the system lacks a constraint mechanism to identify and correct the violation. Traditional forced locking can easily lead to overload of the underlying force control or cause cross-contamination. It cannot take into account both the sense of presence of teleoperation and the compliance of operation, thus resulting in poor control effect of digital twin of teleoperated robot. Summary of the Invention

[0004] To address the technical problem of unsatisfactory digital twin control performance of existing teleoperated robots in biopharmaceutical workshops, the present invention aims to provide a digital twin control method for teleoperated robots that integrates biopharmaceutical cleaning standards. The specific technical solution adopted is as follows: A digital twin control method for remotely operated robots that integrates biomedical cleaning standards includes: Constructing a digital twin scenario includes: a semantic grid map containing restricted boundaries within the workshop; a knowledge graph recording the cleaning standards for the workshop area; and real-time status records of the robot's historical cleaning operations; and collecting the robot's real-time pose and operation commands. At the current moment, the current end-effector position of the robot in the real-time pose is extracted. Spatial constraints are determined based on the spatial distance between the current end-effector position and the no-go boundary in the semantic mesh map. Sequential constraints are determined based on the region where the current end-effector position is located and the knowledge graph. Temporal constraints are determined based on the time interval between the historical cleaning operation time and the current time in the real-time status record and the distance between the historical cleaning operation position and the current end-effector position. By combining spatial constraints, sequential constraints and temporal constraints, the intensity of the specification conflict and the dominant direction of the constraints of the robot end-effector are determined. At the current moment, the expected end-effector velocity of the robot is extracted from the operation instructions. Based on the expected end-effector velocity and the dominant constraint direction, the probability of operation conflict is determined in combination with the intensity of specification conflict. Based on the probability of operation conflict, the robot's operation instructions are intervened and corrected.

[0005] Furthermore, the method for obtaining the semantic grid map includes: The environmental geometry model of the workshop is acquired and rasterized to generate grid cells. Each grid cell is configured with area identifiers and cleanliness level attributes. Pre-built no-entry boundaries are marked in the rasterized environmental geometry model to obtain a semantic grid map.

[0006] Furthermore, the method for acquiring the knowledge graph includes: The cleaning specifications of the workshop area are analyzed, and cleaning area nodes, cleaning action nodes, and cleaning attribute nodes are extracted. Logical relationship edges connecting different nodes are constructed. The logical relationship edges include at least: temporal relationship edges representing the cleaning order between different cleaning area nodes and spatial mutual exclusion edges representing isolation rules, and conditional dependency edges representing triggering restrictions between cleaning action nodes and cleaning attribute nodes. A directed attribute graph is generated based on all nodes and all logical relationship edges to obtain a knowledge graph.

[0007] Furthermore, the method for obtaining the real-time status record includes: Based on the robot's pose flow and command flow, historical cleaning operation events containing job type, job time, and job location are extracted. The status identifiers at each moment are dynamically maintained and determined according to the robot's cleaning operation process. The status identifiers include at least: a completion indicator for recording the cleaning progress of the area, a spatial avoidance indicator for recording the cleaning impact range, and a wear indicator for recording the wear level of cleaning consumables. The wear level indicator is cleared to zero when a consumable replacement instruction is received. Each historical cleaning operation event and its status identifiers at each moment in the process are encapsulated in real time to form a real-time status record.

[0008] Furthermore, the method for obtaining the spatial constraints includes: Calculate the Euclidean distance between the current end position and each restricted boundary, and select the shortest Euclidean distance. Perform clamping and negative correlation normalization on the shortest Euclidean distance to obtain the spatial constraint strength. Take the direction of the restricted boundary corresponding to the shortest Euclidean distance pointing to the current end position as the spatial constraint direction. Determine the spatial constraint based on the spatial constraint strength and spatial constraint direction.

[0009] Furthermore, the method for obtaining the order constraint includes: Based on the knowledge graph, determine whether the current end position is located in a subsequent clean region. If it is located in any subsequent clean region and the corresponding preceding clean region has not yet been cleaned, apply a sequence constraint: The order constraint strength is determined by using the Euclidean distance between the current end position and the preceding cleaned area, and the direction from the current end position to the preceding cleaned area is taken as the order constraint direction; the order constraint is determined based on the order constraint strength and the order constraint direction.

[0010] Furthermore, the method for obtaining the time constraint includes: At the current moment, extract the most recent operation time and location from the historical cleaning operation data; based on the completion status indicator and spatial avoidance indicator, determine whether it is within the constraint period. If it is not within the constraint period, set the time constraint strength to 0; otherwise: The time interval between the most recent operation time and the current time is standardized and negatively correlated to obtain the schedule constraint contribution; the Euclidean distance between the most recent operation position and the current end position is standardized and negatively correlated to obtain the avoidance constraint contribution; the schedule constraint contribution and the avoidance constraint contribution are fused to obtain the time constraint strength. The direction from the most recent operation position to the current end position is used as the time constraint direction; the time constraint is determined based on the time constraint strength and the time constraint direction.

[0011] Furthermore, the method for obtaining the intensity of the normative conflict and the dominant direction of the constraint includes: The spatial constraints, sequence constraints, and time constraints are vector summed. The specification conflict strength is obtained based on the magnitude of the vector sum and the constraint strength of each constraint. The direction of the vector sum is taken as the dominant constraint direction.

[0012] Furthermore, the method for obtaining the operation conflict probability includes: Obtain the canonical conflict intensity of each grid cell within a preset neighborhood of the current end position in the semantic grid map, and determine the gradient of the canonical conflict intensity within the preset neighborhood. The violation factor is determined based on the consistency between the dominant constraint direction of the robot end effector and the direction of the desired end effector velocity. The hazard factor is determined based on the projection of the desired end effector velocity onto the specification conflict intensity gradient. The operational conflict probability is determined by combining the violation factor, the hazard factor, and the specification conflict intensity of the robot end effector.

[0013] Furthermore, methods for intervening in and correcting robot operation command execution based on the probability of operational conflict include: If the probability of the operation conflict is within the first preset range, it is determined to be a safe state, and the original operation instruction is executed. If the probability of the operation conflict is within the second preset range, it is determined to be a warning state, and the remote control terminal is prompted to modify the operation command. If the probability of the operation conflict is within the third preset range, it is determined to be an intervention state. The damping coefficient is determined according to the mapping ratio of the probability of the operation conflict within the third preset range. The damping coefficient is then used to reduce the motion component in the expected end velocity that deviates from the dominant direction of the constraint, thereby obtaining the corrected operation command. The first, second, and third preset intervals do not overlap, and the upper limits of the intervals increase sequentially.

[0014] The present invention has the following beneficial effects: This invention first constructs a digital twin scenario including a semantic mesh map recording prohibited boundaries within the workshop, a knowledge graph recording workshop area cleaning regulations, and real-time status records of robot historical cleaning operations. This deeply integrates the physical environment, operational logic, and operational status, providing comprehensive data support for subsequent intent arbitration and conflict prediction. Next, it extracts the robot's current end-effector position. Based on the spatial distance between the current end-effector position and prohibited boundaries in the semantic mesh map, it determines spatial constraints to block any attempt to cross prohibited boundaries. Based on the area where the current end-effector position is located and the knowledge graph, it determines virtual, guiding sequence constraints. Based on the time interval between historical cleaning operation moments and the current moment, and the distance between historical cleaning operation locations and the current end-effector position in the real-time status records, it determines temporal constraints through spatiotemporal coupling. Then, it comprehensively determines the intensity of regulatory conflict and the dominant direction of compliance constraints, which quantifies the cumulative effect of violation risks at the robot's end-effector. Furthermore, it extracts the robot's expected end-effector speed, identifies violation conflict intentions based on the expected end-effector speed and the dominant direction of constraints, and then determines the probability of operational conflict based on the intensity of regulatory conflict. Finally, it intervenes and corrects the robot's operational commands based on the probability of operational conflict. This invention constructs a digital twin dynamic field that integrates spatial, sequential, and temporal three-dimensional constraints. By quantifying the probability of conflict between operational intentions and cleaning specifications, an adaptive hierarchical damping intervention mechanism is introduced to improve the compliance and effectiveness of digital twin control for remotely operated robots in biomedical workshops. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The flowchart illustrates a digital twin control method for a teleoperated robot that integrates biomedical cleaning standards, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a digital twin control method for a teleoperated robot integrating biomedical cleaning standards proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for a digital twin control method for a teleoperated robot that integrates biomedical cleaning standards, provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a digital twin control method for a teleoperated robot integrating biomedical cleaning standards, provided by an embodiment of the present invention. The method specifically includes: Step S1: Construct a digital twin scenario. The digital twin scenario includes: a semantic grid map containing restricted boundaries in the workshop, a knowledge graph recording the cleaning standards of the workshop area, and real-time status records of the robot's historical cleaning operation data; and collect the robot's real-time pose and operation commands.

[0021] Cleaning in biopharmaceutical production is a task that heavily relies on standard operating procedures (SOPs). Operators must not only avoid physical collisions with robots, but also strictly adhere to cleanliness level classifications, work sequences, and cleaning waiting times. Traditional geometric synchronization control cannot comprehend these underlying logical norms.

[0022] Therefore, the embodiments of the present invention first construct a digital twin scenario for the teleoperated robot to achieve deep integration of physical environment, operation logic and operation status, and provide complete data support for subsequent intention arbitration and conflict prediction.

[0023] In one embodiment of the present invention, the digital twin scenario mainly consists of three components: a semantic grid map containing restricted boundaries within the workshop, a knowledge graph recording the cleaning specifications of the workshop area, and a real-time status record recording the robot's historical cleaning operation data.

[0024] Semantic grid maps transform physical space into quantifiable spatial semantic information by assigning cleanliness levels and restricted access attributes to the grid, ensuring that control commands do not violate area isolation specifications. The knowledge graph of cleaning specifications transforms discrete textual specifications into machine-readable logical graphs, which serve as a logic engine to determine the order of operations between areas and the constraints between actions. Real-time status records maintain dynamic operation progress, operation timestamps, and consumable consumption, providing real-time status criteria for commands.

[0025] Through the collaboration of the three components mentioned above, the digital twin scenario constructs a dynamic constraint field in virtual space that integrates three-dimensional attributes of space, sequence, and time. This allows the robot to quantify and analyze the probability of conflict between the operator's expected movement and the biological cleaning specifications based on the geometric coupling relationship between the constraint field and the operator's expected movement before executing instructions. This ensures that the remote operation process strictly complies with the compliance requirements of biomedical cleaning while retaining the operator's control.

[0026] Since the end effector of the robot is a continuous coordinate point, the environmental geometric model (such as point cloud or CAD model) of the biopharmaceutical production workshop is transformed from a complex physical space into discrete grid cells through rasterization. Then, each grid cell is labeled with independent semantic attributes, and no-go boundaries are planned in the spatial dimension in combination with normative constraints, thereby establishing a mapping relationship between spatial geometry and cleaning specifications, so as to quickly retrieve and evaluate spatial constraints to perform avoidance. Based on this, in a preferred embodiment of the present invention, the method for obtaining a semantic grid map includes: The environmental geometry model of the workshop is acquired and rasterized to generate grid cells. Each grid cell is configured with area identifiers and cleanliness level attributes. Pre-built no-entry boundaries are marked in the rasterized environmental geometry model to obtain a semantic grid map.

[0027] As an example, firstly, a 3D LiDAR (scanning radius 30m, angular resolution 0.1°) deployed in the workshop is used to collect 3D point cloud data of the biomedical workshop and construct a (3D) environmental geometric model; then, the 3D point cloud is processed into a 3D rasterization, dividing the 3D bounding box of the workshop into uniform grid units to generate a continuous 3D spatial grid. The above methods are conventional techniques known in the field and will not be elaborated further. In this example, the side length of the mesh cell is 2cm, but the implementer can adjust it as needed. In scenarios where computing resources are limited or all workshop operations are on the same horizontal plane, a (two-dimensional) environmental geometric model can also be collected for mesh generation. After generating the grid cells, the pre-established Building Information Model (BIM) and Standard Operating Procedures (SOP) are retrieved to configure the grid attributes. Specifically, each grid cell is assigned two attributes: area identification and cleanliness level. These attributes need to be divided and labeled according to the actual situation of the workshop. The area identification includes the identification of different operating areas such as filling area, capping area, pass-through window, and buffer room. The cleanliness level includes A, B, C, D, etc., corresponding to different levels of biological cleanliness requirements. Finally, the pre-marked restricted boundaries in the biopharmaceutical clean workshop are mapped onto the three-dimensional spatial grid to obtain a semantic grid map. In this example, the principles for setting no-entry boundaries include at least the provisions in the cleaning specifications: the principle of unidirectional flow between cleanliness levels, the principle of spatial area isolation, etc. For example, it is strictly forbidden for the robot end effector to directly cross different cleanliness levels, and it is forbidden for the robot end effector to cross areas with different area markings, etc. Implementers may also mark all no-entry boundaries according to actual application conditions. Based on the above principles, the common interface of adjacent grid units with different cleanliness levels and different area markings is used as the no-entry boundary.

[0028] Considering that cleaning specifications are unstructured text, which is difficult for robots to directly parse and execute, we extract three types of cleaning nodes: regions, actions, and attributes. We atomize the entities in the cleaning specifications to transform the text into a structured form, and construct temporal relation edges, spatial mutual exclusion edges, and conditional dependency edges to explicitly transform the logical relationships between entities into directed connections, thereby forming a structured and computable knowledge graph. Based on this, in a preferred embodiment of the present invention, the method for acquiring a knowledge graph includes: The cleaning specifications of the workshop area are analyzed, and cleaning area nodes, cleaning action nodes, and cleaning attribute nodes are extracted. Logical relationship edges connecting different nodes are constructed. The logical relationship edges include at least: temporal relationship edges representing the cleaning order between different cleaning area nodes and spatial mutual exclusion edges representing isolation rules, and conditional dependency edges representing triggering restrictions between cleaning action nodes and cleaning attribute nodes. A directed attribute graph is generated based on all nodes and all logical relationship edges to obtain a knowledge graph.

[0029] As an example, the cleaning specifications (written standard operating procedures (SOP) documents) for the workshop area are first parsed, such as the "Cleaning Operation Instructions" issued by a vaccine production workshop. Using Named Entity Recognition (NER) technology in Natural Language Processing (NLP) or a preset regular expression template, the "Cleaning Operation Instructions" are scanned to extract three types of entities and instantiated as graph nodes: (1) Cleaning area nodes: generated by recognizing spatial terms, such as “filling area”, “capping area”, “pass-through window”, and “buffer room”; (2) Cleaning action nodes: generated by recognizing operation verbs, such as "wiping", "spraying disinfectant", "changing lint-free cloth"; (3) Cleaning attribute nodes: generated by recognizing the combination of numerical values ​​and dimensions, such as extracting "30 seconds" (representing the residence time of disinfectant evaporation) and "6 times" (representing the maximum number of wipes per piece of lint-free cloth).

[0030] Then, using dependency parsing or rule-matching templates, logical relationships between entities are extracted from long sentences in the text, and different types of logical relation edges are constructed: (1) Temporal relationship edge (sequential edge): For standard text describing "from a certain area to another area", construct a one-way edge between two clean area nodes; for example, construct a directed edge from "filling area (A grade)" to "capping area (B grade)", which means that the source node must be cleaned before entering the target node; (2) Spatial mutually exclusive edges (no passage edges): In response to the cross-contamination isolation requirements in the specifications, spatial mutually exclusive edges are constructed between nodes in the clean area; for example, a directed blocking edge is constructed between the non-compliant physical boundary of the "filling area" and the "buffer room" to represent that it is prohibited to directly move through the wall or other boundary crossing movements without following the specified route. (3) Conditional Dependency Edges: For the triggering constraints of the action, construct directed edges from the cleaning action node to the cleaning attribute node; for example, construct a directed edge between the "spray disinfectant" node and the "30 seconds" node to represent that after the spraying action is performed, the associated area must be forced to enter the 30-second time forbidden zone constraint countdown; construct a dependency edge between the "wipe" node and the "6 times" node to represent the cumulative count limit of the wiping action; All extracted nodes and logical relation edges are transformed into Resource Description Framework (RDF) triples in the form of "subject-verb-object". The generated triple data is then imported in batches into a graph database (such as Neo4j) to generate a directed attribute graph that supports standard graph query languages ​​(such as Cypher or SPARQL), thus obtaining a clean and standardized knowledge graph.

[0031] It should be noted that the operations described above in the knowledge graph construction process are already well-known and conventional technical methods in this field, and will not be elaborated further.

[0032] Because the cleaning specifications also contain a large number of time-varying constraints based on operational history, such as which areas have been cleaned, which locations need to be avoided after disinfectant spraying, and the remaining lifespan of the cleaning cloth, these states cannot be obtained from static knowledge graphs or semantic mesh maps. It is necessary to dynamically extract historical cleaning operation events by monitoring the robot's pose flow and command flow in real time, and maintain state labels to reduce the complex historical operation trajectory to lightweight state labels. Each event and its corresponding state label at that time are encapsulated in real time to form a continuous state record for subsequent constraint calculation and query. Based on this, in a preferred embodiment of the present invention, the method for obtaining real-time status records includes: Based on the robot's pose flow and command flow, historical cleaning operation events containing job type, job time, and job location are extracted. The status identifiers at each moment are dynamically maintained and determined according to the robot's cleaning operation process. The status identifiers include at least: a completion indicator for recording the cleaning progress of the area, a spatial avoidance indicator for recording the cleaning impact range, and a wear indicator for recording the wear level of cleaning consumables. The wear level indicator is cleared to zero when a consumable replacement instruction is received. Each historical cleaning operation event and its status identifiers at each moment in the process are encapsulated in real time to form a real-time status record.

[0033] As an example, the command stream issued by the operator (operator handle) and the robot's end-effector pose are parsed frame by frame. When the command type changes or a specific command is continuously executed, the historical cleaning operation events of the command type (i.e., job type, 0 move, 1 wipe, 2 spray, 3 change cloth), action timestamp (i.e. job time) and the corresponding end-effector position point set (i.e. job position) are recorded. A cleaning status table is built in memory for each region, with the initial value being "not cleaned". For example, when the job type of a historical cleaning job event is identified as "wiping", the job location of the action is mapped to the semantic grid map. If a region's grid is covered by the action trajectory, the completion status of that region in the cleaning status table is updated to "cleaned and completed". Otherwise, the completion status is updated to "not cleaned and completed", which is used to support the condition determination of sequential constraints. When a historical cleaning operation event is identified as "spraying", all work locations during the spraying (disinfectant) operation are designated as the center of a restricted area, generating a spatial avoidance marker with an influence radius of a preset distance (e.g., 100mm). The current operation time plus the required duration of disinfectant stay (i.e., the constraint period, e.g., 30s) is set as the expiration timestamp for this marker. When the time exceeds this expiration timestamp, the spatial avoidance marker is automatically removed. Set an integer variable to record the number of times the cleaning cloth is used; increment the integer variable by 1 after each independent "wiping" operation is completed, which is recorded as the wear level indicator of the cleaning consumables; when a "replace cloth" instruction event is received and identified, the integer variable is forcibly cleared to zero, that is, the wear level indicator is reset to zero. The aforementioned real-time updated completion markers, space avoidance markers, and loss level markers are synchronously refreshed to the shared memory area of ​​the digital twin server at a preset frame rate (e.g., 30Hz) to form a real-time status record.

[0034] It should be noted that the above operations are all conventional technical means in this field, and implementers can adjust them according to actual applications, which will not be elaborated further.

[0035] Further, the robot's real-time pose and operation commands are collected based on the various monitoring and sensing devices built into the robot. This is a well-known technical means in the art and will not be elaborated further.

[0036] Step S2: At the current moment, extract the current end-effector position of the robot in the real-time pose. Determine spatial constraints based on the spatial distance between the current end-effector position and the prohibited boundary in the semantic mesh map. Determine sequential constraints based on the region where the current end-effector position is located and the knowledge graph. Determine temporal constraints based on the time interval between the historical cleaning operation time and the current time in the real-time status record and the distance between the historical cleaning operation position and the current end-effector position. Combine spatial constraints, sequential constraints and temporal constraints to determine the specification conflict intensity and constraint dominance direction of the robot end-effector.

[0037] In remote cleaning of vaccine workshops, the constraints faced by operators are not isolated. By extracting the current end position of the robot in real time, the static standard logic is transformed into multi-dimensional dynamic geometric constraints in the current state, which can transform complex biological cleaning standards from "paper operating procedures" to "online closed-loop control".

[0038] In one embodiment of the present invention, the current end-effector position of the robot is first extracted from the real-time pose at the current moment. It should be noted that calculating the end-effector position based on the robot's real-time pose is a well-known conventional technique in the field and will not be described in detail here.

[0039] In digital twin scenarios, there are strict barriers between areas of different cleanliness levels (such as Class A and Class B), and distance is the most direct physical indicator for measuring the risk of violation. By calculating the spatial distance between the current end position and the restricted boundary in real time, the "no crossing the boundary" instruction in the cleaning specifications is transformed into a quantifiable spatial constraint in the control process. Based on this, the spatial constraint is first determined according to the spatial distance between the current end position and the restricted boundary in the semantic grid map. Introducing spatial constraints can prevent physical crossing of the boundary during remote operation and avoid cross-contamination.

[0040] Preferably, in one embodiment of the present invention, considering the calculation of the Euclidean distance between the current end-effector position and the restricted boundary, the dangerous boundary closest to the robot and with the highest risk level can be identified, so as to concentrate computing resources to handle the most urgent boundary crossing risks; and the closer the robot's current end-effector position is to the restricted boundary, the stronger the obstacle constraint should be generated to intercept the operation. By clamping and negatively correlated mapping of the shortest Euclidean distance, a physical response mechanism with stronger constraint as the distance is closer can be established, quantifying the spatial constraint strength; furthermore, the direction from the restricted boundary to the current end-effector position is taken as the spatial constraint direction, establishing an "escape" direction for the robot away from the danger zone; through the coupling of strength and direction, a virtual constraint force field pointing to the safe area is constructed in the digital twin scenario, determining the spatial constraint that blocks any attempt to cross the restricted boundary; then the method for obtaining the spatial constraint includes: Calculate the Euclidean distance between the current end position and each restricted boundary, and select the shortest Euclidean distance. Perform clamping and negative correlation normalization on the shortest Euclidean distance to obtain the spatial constraint strength. Take the direction of the restricted boundary corresponding to the shortest Euclidean distance pointing to the current end position as the spatial constraint direction. Determine the spatial constraint based on the spatial constraint strength and spatial constraint direction.

[0041] As an example, the coordinates of the current end position in the semantic mesh map are first extracted, and the set of line segments corresponding to all forbidden boundaries in the semantic mesh map is extracted simultaneously. The implementer can further filter the set of line segments of forbidden boundaries in the neighborhood space of the current end position to reduce the amount of computation. Using the point-to-line distance formula, the Euclidean distance (shortest distance, referring to the distance from the point to the foot of the perpendicular of the line segment) between the current end position and each forbidden boundary is calculated, and the shortest Euclidean distance d is selected from them to characterize the degree of closeness of the robot to the nearest violation red line. Set an anti-singularity clamping threshold (e.g., 0.5mm) to clamp the shortest Euclidean distance d. If d < 0.5mm, force d = 0.5mm to prevent subsequent calculations from crashing. Remove the dimensions from the clamped shortest Euclidean distance (e.g., multiply by a preset adjustment factor of 1). After that, a negative correlation mapping is performed, for example, mapping to an exponential function exp(-x) with the natural constant as the base, and normalization is performed to obtain the spatial constraint strength; then the direction of the nearest point on the line segment corresponding to the forbidden boundary corresponding to the shortest Euclidean distance pointing to the current end position is taken as the spatial constraint direction; finally, the spatial constraint strength and the spatial constraint direction are combined to form a vector to obtain the spatial constraint (vector).

[0042] In digital twin scenarios, biological cleaning must adhere to the process compliance of the cleaning path (sequence). The current end position can help dynamically determine whether the current operation violates the preset process logic, transforming "logic violation" instructions into quantifiable sequence constraints in the control process. Based on this, the sequence constraints are determined according to the region and knowledge graph of the current end position. Introducing sequence constraints can lock the workflow and prevent the entire workshop cleaning from failing due to operator omissions.

[0043] Preferably, in one embodiment of the present invention, considering that the knowledge graph defines a mandatory sequential cleaning logic topology between areas within the workshop (area 1 must be cleaned before entering area 2), which is the basis for compliance review; by identifying whether the robot has skipped levels and entered a subsequent area, logical violations are accurately captured; if the preceding cleaning area is not cleaned, it indicates that there is a missing item in the current operation step, and constraints must be triggered; when the robot's current end position is in a subsequent cleaning area and the farther away from the preceding cleaning area, the greater the deviation from the correct task objective, the strength of the sequential constraint can be directly quantified based on Euclidean distance; furthermore, the direction pointing to the preceding cleaning area is used as the spatial constraint direction, establishing an operation direction that the robot should return to and complete; by coupling strength and direction, a virtual sequential constraint with guiding function (which can also be regarded as compliance gravity) is constructed to ensure that the cleaning task is executed according to the process sequence; then the method for obtaining the sequential constraint includes: Based on the knowledge graph, determine whether the current end position is located in a subsequent clean region. If it is located in any subsequent clean region and the corresponding preceding clean region has not yet been cleaned, apply a sequence constraint: The strength of the sequence constraint is determined by the Euclidean distance between the current end position and the previous cleaned area, and the direction from the current end position to the previous cleaned area is taken as the direction of the sequence constraint; the sequence constraint is determined based on the strength and direction of the sequence constraint.

[0044] As an example, firstly, the coordinates of the current endpoint in the semantic mesh map are extracted, and its region identifier and the clean region node to which it belongs are queried; then, a topological search is performed in the knowledge graph with this region representation as the endpoint to find the sequential edges pointing to the clean region node; in each sequential edge, it is determined whether the clean region node belongs to a subsequent clean region; if it is in any subsequent clean region and the preceding clean region corresponding to this subsequent clean region (within the same sequential edge) has not yet been cleaned, then it is determined that the current state is in a logical violation state, and sequence constraints are applied: Specifically, the preceding clean region corresponding to the subsequent clean region (within the same sequential edge) is extracted, the shortest Euclidean distance between the current end position and the preceding clean region is calculated, the dimension of the shortest Euclidean distance is removed (e.g., divided by a preset adjustment coefficient of 1mm), and then normalized by mapping it to the hyperbolic tangent function to obtain the order constraint strength; the direction from the current end position to the nearest point in the preceding clean region is taken as the order constraint direction; finally, the order constraint strength and the order constraint direction are combined to form a vector to obtain the order constraint (vector).

[0045] In digital twin scenarios, certain cleaning actions in biological cleaning, such as disinfectant spraying, require a specific surface contact time after spraying to kill pathogens. The current end-effector position helps determine whether the robot is attempting to intrude into sensitive areas where the sterilization cycle has not yet been completed, while the time interval between historical cleaning operation moments and the current moment helps assess whether the sterilization cycle has not been completed. Based on this, time constraints are determined according to the time interval between historical cleaning operation moments and the current moment, as well as the distance between historical cleaning operation positions and the current end-effector position, in the real-time status record. Time constraints are determined through spatiotemporal coupling. Introducing time constraints can forcibly intercept operator's accidental wiping behavior during the disinfectant's effective period, ensuring that each cleaning procedure meets the required biochemical indicators and guaranteeing the compliance of cleaning results.

[0046] Preferably, in one embodiment of the present invention, taking the spraying of disinfectant as an example, if the current moment is within the constraint period (i.e., the area has not yet been sprayed, or the disinfection waiting time has expired), no additional constraints need to be applied; conversely, within a specific window period from spraying to disinfection completion, time constraints must be activated to prevent violations from disrupting the film-forming sterilization process of the disinfectant; as time progresses, the sterilization degree of the disinfectant gradually increases, and by negatively mapping the time interval between the most recent spraying operation and the current moment, the process of sterilization risk gradually being released over time can be simulated, ensuring that the progress constraint contribution from the time perspective of the constraint period smoothly disappears when disinfection is completed; at the same time, the current end position is far from the most recent spraying The closer to the center, the higher the concentration of the disinfectant and the greater the risk of interference. By mapping the Euclidean distance negatively, a gradient protection field decreasing from the center to the periphery can be formed around the spray point, forcibly guiding the robot's end effector to avoid areas covered by disinfectant that have not yet taken effect. Furthermore, the direction from the most recent operation position to the current end effector position is used as the time constraint direction, constructing a direction for the robot to repel movement away from the most recent disinfection area. Through the coupling of intensity and direction, and the dual fusion of time progress and spatial avoidance, a virtual restricted area is constructed in the digital twin scenario to adaptively intercept illegal wiping behavior, ensuring the compliance of sterilization effect and the flexibility of system operation. The methods for obtaining the time constraint include: At the current moment, extract the most recent operation time and location from the historical cleaning operation data; based on the completion status indicator and spatial avoidance indicator, determine whether it is within the constraint period. If it is not within the constraint period, set the time constraint strength to 0; otherwise: The time interval between the most recent operation and the current operation is standardized and negatively correlated to obtain the schedule constraint contribution; the Euclidean distance between the most recent operation position and the current end position is standardized and negatively correlated to obtain the avoidance constraint contribution; the schedule constraint contribution and the avoidance constraint contribution are combined to obtain the time constraint strength. The direction from the most recent operation position to the current end position is used as the time constraint direction; the time constraint is determined based on the time constraint strength and the time constraint direction.

[0047] As an example, to retrieve the historical cleaning operation event of the last work instruction in the real-time status record, taking the spraying of disinfectant as an example: extract the end time of the spraying action and record it as the operation time; extract the centroid coordinates of the end position point set in the spraying action and record it as the operation position. Read the completion status and space avoidance status of the area where the spraying action is located at the current moment; if the completion status is updated to "cleaning completed" and the space avoidance status is removed, it is determined that it is not within the constraint period, and the time constraint strength is set to 0; otherwise, it is determined that it is within the constraint period (i.e., the disinfectant reaction protection period), and subsequent analysis is performed: Divide the time interval between the most recent operation time and the current time by the required duration of disinfectant residence (i.e., the duration of the constraint period, such as 30s) to standardize and remove the influence of dimensions. Then, perform a negative correlation mapping on the quotient, such as subtracting the quotient from 1 (wherein, the above time interval is calculated under the condition of being in the constraint period, so the quotient is always less than or equal to 1) to obtain the schedule constraint contribution. Calculate the Euclidean distance between the most recent job position and the current end position, and apply the Gaussian kernel function to the Euclidean distance. Standardization is performed to remove the influence of dimensions and negative correlation mapping, resulting in the mapping result. The mapping result is used as the contribution to the avoidance constraint; where, The coordinates of the current end position. The location of the most recent operation. For the preset distance (i.e., the radius of influence of the spraying, such as 100mm), exp() is an exponential function with the natural constant as the base; Then, the progress constraint contribution and the avoidance constraint contribution are fused together, such as by direct multiplication, to obtain the time constraint strength. Implementers may also use other fusion methods, such as weighted fusion, to achieve spatiotemporal coupling, which will not be elaborated here. The direction from the most recent operation position to the current end position is taken as the time constraint direction. Finally, the time constraint strength and the time constraint direction are combined to form a vector to obtain the time constraint (vector).

[0048] In the complex biopharmaceutical production process, the compliance requirements faced by robots are multi-dimensional, intertwined, and interactive. Therefore, after obtaining spatial constraints, sequential constraints, and temporal constraints, the intensity of regulatory conflict and the dominant direction of constraints of the robot end are determined by combining the three factors. The intensity of regulatory conflict can quantify the superposition effect of violation risks, and the dominant direction of constraints provides the most urgent and core compliance trend in the complex constraint field, providing a foundation for subsequent regulation of robots.

[0049] Preferably, in one embodiment of the present invention, the method for obtaining the intensity of normative conflict and the dominant direction of constraints includes: The spatial constraints, sequence constraints, and time constraints are vector summed. The specification conflict strength is obtained based on the magnitude of the vector sum and the constraint strength of each constraint. The direction of the vector sum is taken as the dominant constraint direction.

[0050] Specifically, the spatial constraints, sequence constraints, and time constraints are vector-summed using well-known techniques, which will not be elaborated further. The constraint strengths under each constraint are extracted, multiplied, and fused to obtain the constraint coupling term. Where i is the type label of the spatial constraint, sequence constraint and time constraint; m is the total number of type labels, which is 3 in this example; For each type of constraint, i.e., different values ​​of i correspond to the constraint strength under spatial constraints, sequential constraints, and temporal constraints, respectively; This represents the sum of all constraint strengths multiplied by a constant 1. The purpose of adding 1 is to prevent the sum from becoming smaller when the constraint strength values ​​are between 0 and 1; however, adding 1 might lead to excessively fast amplification, so it is necessary to divide by an exponential scaling factor. Finally, subtract the constant 1 from both the numerator and denominator, so that when When there are no constraints, the constraint coupling term is 0; In other embodiments, the constraint strengths under each constraint can be weighted and summed to obtain the constraint coupling term; The constraint coupling term is fused with the magnitude of the vector sum. In this example, they are multiplied to obtain the specification conflict strength. In other examples, a weighted sum can also be used, which will not be elaborated here. Then, the direction of the vector sum is taken as the dominant constraint direction.

[0051] Step S3: At the current moment, extract the robot's expected end-effector velocity from the operation command, determine the operation conflict probability based on the expected end-effector velocity and the dominant constraint direction, combined with the specification conflict intensity; intervene and correct the robot's operation command execution based on the operation conflict probability.

[0052] Since the risk of operational conflict cannot be determined solely by the current end-effector position, while the desired end-effector position and desired end-effector velocity in the operation command can represent the operator's active control intention, one embodiment of the present invention first extracts the desired end-effector position and desired end-effector velocity in the operation command at the current moment.

[0053] It should be noted that the extraction of the desired end position and the desired end velocity (vector) is a well-known conventional technique in the field and will not be elaborated further; in addition, when the desired end velocity is 0, subsequent corrections can be temporarily omitted.

[0054] Considering that analyzing the expected end velocity and the dominant direction of constraints can help assess whether the operator is conforming to the regulations (in terms of direction), thereby identifying the intention to violate regulations and conflict; at the same time, the intensity of regulation conflict quantifies the degree of violation conflict, and thus the probability of operational conflict can be comprehensively assessed by combining the intention to violate regulations and the quantified degree of conflict.

[0055] Preferably, in one embodiment of the present invention, considering that analyzing the gradient of the normative conflict intensity in the neighborhood of the current end-effector position can help predict the distribution of dangers in the surrounding space, providing a compliance basis for controlling the robot; and analyzing the consistency between the dominant constraint direction of the robot's end-effector and the direction of the desired end-effector velocity can help quantify the violation of the direction of the operational intent, thereby accurately identifying the operator's subjective tendency to violate the rules; further, combining the projection of the desired end-effector velocity onto the normative conflict intensity gradient can help assess the dangerous situation of the predicted operational intent; thus, the probability of operational conflict can be assessed by comprehensively considering the violation situation, the dangerous situation, and the degree of conflict quantification; the method for obtaining the probability of operational conflict includes: Obtain the canonical conflict intensity of each grid cell within a preset neighborhood of the current end position in the semantic grid map, and determine the gradient of the canonical conflict intensity within the preset neighborhood. The violation factor is determined based on the consistency between the dominant constraint direction of the robot end effector and the direction of the desired end effector velocity. The hazard factor is determined based on the projection of the desired end effector velocity onto the specification conflict intensity gradient. The operational conflict probability is determined by combining the violation factor, the hazard factor, and the specification conflict intensity of the robot end effector.

[0056] As an example, firstly, based on the method for obtaining the canonical conflict intensity at the current end position, the canonical conflict intensity of each grid cell in the semantic grid map is calculated; a preset neighborhood, such as 3×3×3, is constructed with the current end position as the center, and the canonical conflict intensity of a total of 26 grid cells within the preset neighborhood is extracted; using the central difference operator or trilinear interpolation algorithm, the partial derivatives of the canonical conflict intensity on the spatial coordinate axes are calculated to obtain the gradient (vector) of the canonical conflict intensity within the preset neighborhood; the calculation of the gradient is a conventional technique in this field and will not be elaborated further; implementers can also adjust the size of the preset neighborhood themselves. Then, using the dot product to measure directional consistency, the unit vector of the desired end-effector velocity is extracted, and the dot product of this unit vector and the unit vector corresponding to the dominant constraint direction of the robot's end-effector is calculated. The dot product is given a negative sign and truncated to obtain the violation factor. When the dot product is positive, it indicates directional consistency. After giving it a negative sign, the result is less than 0, so it is truncated to a negative value, forcing the result to 0, thus obtaining the violation factor. When the dot product is negative, it indicates directional inconsistency and the risk of violation. After giving it a negative sign, the result is greater than 0, thus obtaining the violation factor. Then, the projection of the desired end velocity onto the standard conflict intensity gradient is calculated, and the sign of the projection component is extracted. When the sign is positive, it indicates that the desired end velocity tends to move in the direction of increasing conflict intensity, which may lead to a more dangerous conflict. The risk factor is set to a preset high-risk parameter, such as 2. When the sign is negative, it indicates that the desired end velocity tends to move in the direction of decreasing conflict intensity, which reduces the conflict risk. The risk factor is set to a preset low-risk parameter, such as 0.5. When the projection component is 0, the risk factor is set to a constant 1. Finally, the violation factor, the hazard factor, and the intensity of the specification conflict at the robot end are multiplied and fused to obtain the operation conflict probability. In addition, the product is numerically clamped, that is, when the product is greater than 1, the operation conflict probability is forced to take the upper limit value of 1.

[0057] In another embodiment of the present invention, a penalty item based on the wear level indicator can also be introduced; if the wear level indicator of the cleaning consumable is 1 and the instruction to replace the consumable is still not executed, the value of the wear level indicator (i.e., the integer variable of the number of times the cleaning cloth is used) is divided by the number of times it should be used to obtain the penalty weight, and the penalty weight is used to weight the probability of operation conflict for subsequent intervention and correction.

[0058] Then, intervention and correction are performed on the robot's operation commands based on the probability of operation conflicts.

[0059] Preferably, in one embodiment of the present invention, the method for intervening in and correcting the execution of robot operation commands based on the probability of operation conflict includes: If the probability of an operation conflict is within the first preset range, it is determined to be a safe state, and the original operation command is executed. If the probability of an operation conflict is within the second preset range, it is determined to be a warning state, and the remote control terminal is prompted to modify the operation command. If the probability of operational conflict is within the third preset range, it is determined to be an intervention state. The damping coefficient is determined according to the mapping ratio of the probability of operational conflict within the third preset range. The damping coefficient is then used to reduce the motion component in the desired end velocity that deviates from the dominant direction of the constraint, and the corrected operational command is obtained. The first, second, and third preset intervals do not overlap, and the upper limits of the intervals increase sequentially.

[0060] As an example, the first preset interval is [0, 0.3), the second preset interval is [0.3, 0.7), and the third preset interval is [0.7, 1]. Implementers can also adjust the value range themselves; however, it is necessary to ensure that the first, second, and third preset intervals are continuous and do not overlap, and that the upper limit of the intervals increases sequentially. When the probability of an operation conflict is within the second preset range, it is determined to be a warning state, prompting the remote control terminal to modify the operation command. Specifically, the dominant direction of the constraint is highlighted on the digital twin interface, but the command is not forced to be modified. When the probability of operational conflict is within the third preset interval, it is determined to be an intervention state. The probability of operational conflict is subtracted by 0.7 and then divided by the interval length of 0.3 to obtain the damping coefficient. The motion component that deviates from the dominant constraint direction in the expected end velocity is extracted. The damping coefficient is multiplied by the motion component to obtain the velocity to be reduced. The vector difference between the expected end velocity and the velocity to be reduced is calculated to obtain the final corrected expected end velocity. Then, the corrected expected end position is solved by integrating the corrected expected end velocity. The corrected expected end velocity and the corrected expected end position constitute the corrected operational command.

[0061] The above are all conventional technical methods in this field, and will not be elaborated further. Implementers may also make their own adjustments.

[0062] In summary, this invention first constructs a digital twin scenario; then, it extracts the robot's current end-effector position, determines spatial constraints based on the spatial distance between the current end-effector position and the prohibited boundaries in the semantic grid map, determines sequential constraints based on the region where the current end-effector position is located and the knowledge graph, and determines temporal constraints based on the time interval between the historical cleaning operation time and the current time, as well as the distance between the historical cleaning operation location and the current end-effector position in the real-time status record; further, it comprehensively determines the regulatory conflict intensity and constraint dominance direction of the robot's end-effector; and further, it extracts the robot's desired end-effector speed, and determines the operational conflict probability based on the desired end-effector speed and the constraint dominance direction, combined with the regulatory conflict intensity, to intervene and correct the robot's operational command execution. This invention constructs a digital twin dynamic field that integrates spatial, sequential, and temporal three-dimensional constraints, and by quantifying the conflict probability between operational intentions and cleaning regulations, it introduces an adaptive hierarchical damping intervention mechanism, thereby improving the compliance and effectiveness of digital twin control of remotely operated robots in biomedical workshops.

[0063] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A digital twin control method for remotely operated robots integrating biomedical cleaning standards, characterized in that, include: Constructing a digital twin scenario includes: a semantic grid map containing restricted boundaries within the workshop; a knowledge graph recording the cleaning standards for the workshop area; and real-time status records of the robot's historical cleaning operations; and collecting the robot's real-time pose and operation commands. At the current moment, the current end-effector position of the robot in the real-time pose is extracted. Spatial constraints are determined based on the spatial distance between the current end-effector position and the no-go boundary in the semantic mesh map. Sequential constraints are determined based on the region where the current end-effector position is located and the knowledge graph. Temporal constraints are determined based on the time interval between the historical cleaning operation time and the current time in the real-time status record and the distance between the historical cleaning operation position and the current end-effector position. By combining spatial constraints, sequential constraints and temporal constraints, the intensity of the specification conflict and the dominant direction of the constraints of the robot end-effector are determined. At the current moment, the expected end-effector velocity of the robot is extracted from the operation instructions. Based on the expected end-effector velocity and the dominant constraint direction, the probability of operation conflict is determined in combination with the intensity of specification conflict. Based on the probability of operation conflict, the robot's operation instructions are intervened and corrected.

2. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The method for obtaining the semantic grid map includes: The environmental geometry model of the workshop is acquired and rasterized to generate grid cells. Each grid cell is configured with area identifiers and cleanliness level attributes. Pre-built no-entry boundaries are marked in the rasterized environmental geometry model to obtain a semantic grid map.

3. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The methods for obtaining the knowledge graph include: The cleaning specifications of the workshop area are analyzed, and cleaning area nodes, cleaning action nodes, and cleaning attribute nodes are extracted. Logical relationship edges connecting different nodes are constructed. The logical relationship edges include at least: temporal relationship edges representing the cleaning order between different cleaning area nodes and spatial mutual exclusion edges representing isolation rules, and conditional dependency edges representing triggering restrictions between cleaning action nodes and cleaning attribute nodes. A directed attribute graph is generated based on all nodes and all logical relationship edges to obtain a knowledge graph.

4. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The method for obtaining the real-time status record includes: Based on the robot's pose flow and command flow, historical cleaning operation events containing job type, job time, and job location are extracted. The status identifiers at each moment are dynamically maintained and determined according to the robot's cleaning operation process. The status identifiers include at least: a completion indicator for recording the cleaning progress of the area, a spatial avoidance indicator for recording the cleaning impact range, and a wear indicator for recording the wear level of cleaning consumables. The wear level indicator is cleared to zero when a consumable replacement instruction is received. Each historical cleaning operation event and its status identifiers at each moment in the process are encapsulated in real time to form a real-time status record.

5. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The method for obtaining the spatial constraints includes: Calculate the Euclidean distance between the current end position and each restricted boundary, and select the shortest Euclidean distance. Perform clamping and negative correlation normalization on the shortest Euclidean distance to obtain the spatial constraint strength. Take the direction of the restricted boundary corresponding to the shortest Euclidean distance pointing to the current end position as the spatial constraint direction. Determine the spatial constraint based on the spatial constraint strength and spatial constraint direction.

6. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 3, characterized in that, The method for obtaining the order constraint includes: Based on the knowledge graph, determine whether the current end position is located in a subsequent clean region. If it is located in any subsequent clean region and the corresponding preceding clean region has not yet been cleaned, apply a sequence constraint: The order constraint strength is determined by using the Euclidean distance between the current end position and the preceding cleaned area, and the direction from the current end position to the preceding cleaned area is taken as the order constraint direction; the order constraint is determined based on the order constraint strength and the order constraint direction.

7. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 4, characterized in that, The method for obtaining the time constraint includes: At the current moment, extract the most recent operation time and location from the historical cleaning operation data; based on the completion status indicator and spatial avoidance indicator, determine whether it is within the constraint period. If it is not within the constraint period, set the time constraint strength to 0; otherwise: The time interval between the most recent operation time and the current time is standardized and negatively correlated to obtain the schedule constraint contribution; the Euclidean distance between the most recent operation position and the current end position is standardized and negatively correlated to obtain the avoidance constraint contribution; the schedule constraint contribution and the avoidance constraint contribution are fused to obtain the time constraint strength. The direction from the most recent operation position to the current end position is used as the time constraint direction; the time constraint is determined based on the time constraint strength and the time constraint direction.

8. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The methods for obtaining the intensity of the normative conflict and the dominant direction of the constraint include: The spatial constraints, sequence constraints, and time constraints are vector summed. The specification conflict strength is obtained based on the magnitude of the vector sum and the constraint strength of each constraint. The direction of the vector sum is taken as the dominant constraint direction.

9. The digital twin control method for remotely operated robots integrating biomedical cleaning standards according to claim 1, characterized in that, The method for obtaining the probability of operation conflict includes: Obtain the canonical conflict intensity of each grid cell within a preset neighborhood of the current end position in the semantic grid map, and determine the gradient of the canonical conflict intensity within the preset neighborhood. The violation factor is determined based on the consistency between the dominant constraint direction of the robot end effector and the direction of the desired end effector velocity. The hazard factor is determined based on the projection of the desired end effector velocity onto the specification conflict intensity gradient. The operational conflict probability is determined by combining the violation factor, the hazard factor, and the specification conflict intensity of the robot end effector.

10. The digital twin control method for a teleoperated robot integrating biomedical cleaning standards according to claim 9, characterized in that, Methods for intervening in and correcting robot operation command execution based on the probability of operational conflict include: If the probability of the operation conflict is within the first preset range, it is determined to be a safe state, and the original operation instruction is executed. If the probability of the operation conflict is within the second preset range, it is determined to be a warning state, and the remote control terminal is prompted to modify the operation command. If the probability of the operation conflict is within the third preset range, it is determined to be an intervention state. The damping coefficient is determined according to the mapping ratio of the probability of the operation conflict within the third preset range. The damping coefficient is then used to reduce the motion component in the expected end velocity that deviates from the dominant direction of the constraint, thereby obtaining the corrected operation command. The first, second, and third preset intervals do not overlap, and the upper limits of the intervals increase sequentially.

Images