Multi-regulatory domain real-time compliance decision and data management method for humanoid robots

Abstract

The application provides a multi-regulatory domain real-time compliance decision and data management method for humanoid robots, and relates to the technical field of robot control and data security. The method comprises the following steps: acquiring environment perception, ontology state and target task instruction in real time; identifying the current regulatory domain based on the environment perception data and calling the corresponding regulatory constraint set; calculating the dynamic compliance influence factor according to the ontology state, the task instruction and the regulatory constraint set; generating an executable subtask sequence that conforms to the regulatory constraint set based on the factor to decompose and reconstruct the task instruction; monitoring the interactive data flow in real time during execution, calculating the regulatory domain operation entropy based on the data privacy rules and the dynamic compliance influence factor; judging whether the entropy exceeds the preset safety threshold; if yes, triggering the blocking, desensitization or degradation processing, and outputting the final control instruction. The application realizes multi-regulatory domain real-time compliance decision and dynamic data management, avoids cross-domain violations, guarantees privacy security, and continuously optimizes the decision accuracy through a self-learning mechanism.

Description

Technical Field

[0001] This invention relates to the field of robot control and data security technology, and in particular to a method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots. Background Technology

[0002] With the rapid development of humanoid robot technology, its application scenarios are becoming increasingly widespread, expanding from industrial manufacturing and medical care to home services, public safety, and many other fields. These scenarios often involve different geographical locations or management areas, each with its own specific laws, regulations, and behavioral norms. For example, traffic rules must be followed when walking in public places, patient privacy protection regulations must be followed in medical institutions, and data confidentiality regulations must be enforced in corporate office areas. When humanoid robots operate across regions, they must be able to perceive changes in the regulations of their environment in real time and adjust their behavior and data processing methods to ensure compliance.

[0003] However, existing humanoid robot task planning and control systems typically focus only on motion control and basic task execution, lacking the ability to perceive and adapt to multiple regulatory domains. When entering a new area, robots often fail to automatically recognize changes in regulatory domains or dynamically adjust their task execution strategies to adapt to new rule constraints. This can lead to robots unintentionally violating local regulations, causing legal risks or safety incidents. For example, a robot might continue collecting image data in an area where photography is prohibited, or perform movement tasks in a restricted area.

[0004] Regarding data privacy protection, most existing technologies employ static privacy strategies, meaning a fixed set of data access rules is preset during system deployment. This static strategy cannot adapt to dynamic changes in regulatory domains; the same data may have different privacy requirements in different regions. For example, facial image data may be allowed to be collected in public areas, but strictly restricted in medical areas or private residences. Existing systems lack the ability to quantitatively assess the real-time risks of data flows across domains and cannot dynamically adjust the scope of data dissemination, access frequency, and processing methods according to current regulatory domains, easily leading to data leaks or unauthorized use.

[0005] Furthermore, the regulatory domain itself is not static; ad hoc events, policy updates, or regional reclassification can all lead to changes in the set of regulatory constraints. Current technologies lack a mechanism to learn and optimize from historical execution data, and cannot adaptively adjust compliance decision parameters to adapt to the evolving regulatory environment. This makes it difficult for the robot to continuously optimize its compliance performance over long-term operation, and it may even become ineffective due to regulatory updates.

[0006] In summary, existing technologies have significant shortcomings in real-time compliance decision-making and data management for humanoid robots in multi-regulatory domain environments. They cannot effectively address issues such as regulatory adaptation when robots operate across regions, dynamic protection of data privacy, and continuous optimization for long-term system operation. These problems limit the reliability and safety of large-scale applications of humanoid robots in complex social environments. Summary of the Invention

[0007] To address the technical problems in existing technologies, such as the inability of humanoid robots to perceive changes in multiple regulatory domains in real time, the inability to dynamically quantify compliance risks, the inability to adaptively adjust data protection strategies, and the lack of continuous optimization capabilities, this invention provides a method for real-time compliance decision-making and data management across multiple regulatory domains for humanoid robots.

[0008] The technical solution provided by this invention is as follows:

[0009] The present invention provides a method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots, comprising:

[0010] S1: Real-time acquisition of environmental perception data of the humanoid robot's environment, the robot's body state data, and the target task instructions to be executed by the robot;

[0011] S2: Based on the environmental perception data, identify at least one regulatory domain in which the robot is currently located, and retrieve the regulatory constraint set corresponding to the at least one regulatory domain, wherein the regulatory constraint set contains at least one behavioral compliance rule;

[0012] S3: Based on the ontology state data, the target task instruction, and the set of regulatory constraints, calculate the dynamic compliance impact factor of the robot in response to the target task instruction at the current moment;

[0013] S4: Based on the dynamic compliance impact factor, decompose and reconstruct the target task instruction to generate at least one executable subtask sequence that conforms to the set of regulatory constraints;

[0014] S5: During the execution of the executable subtask sequence, the robot's interactive data stream is monitored in real time, and the regulatory domain operation entropy of the interactive data stream is calculated based on the preset data privacy rules and the dynamic compliance impact factor.

[0015] S6: Determine whether the entropy of the regulatory domain operation exceeds a preset security threshold;

[0016] S7: If the limit is not exceeded, the interactive data stream is allowed to be transmitted and processed according to the original path; if the limit is exceeded, the data control strategy is triggered to block, desensitize or downgrade the interactive data stream, and output the final control command that meets the real-time compliance requirements of multiple regulatory domains.

[0017] The beneficial effects of the technical solution provided by this invention include at least the following:

[0018] (1) In this invention, the current regulatory domain is identified by acquiring environmental perception data in real time, and the corresponding regulatory constraint set is retrieved. Combined with the ontology state data and the target task instructions, a dynamic compliance impact factor is calculated, thereby realizing a quantitative assessment of the risk of performing tasks in a multi-regulatory domain environment. This factor comprehensively considers multiple dimensions such as time decay, spatial distance, rule risk level, and cost of violation, enabling the robot to dynamically perceive changes in the regulatory environment and adjust its task execution strategy, effectively avoiding legal risks and safety accidents caused by cross-domain violations.

[0019] (2) In this invention, by monitoring the interactive data flow in real time during task execution and calculating the operational entropy of the regulatory domain based on data privacy rules and dynamic compliance impact factors, a quantitative assessment of data flow risks is achieved. When the operational entropy of the regulatory domain exceeds a preset security threshold, data control strategies such as blocking, desensitization, or downgrading are triggered in a timely manner. This mechanism ensures that the propagation and processing of data between different regulatory domains always comply with privacy requirements, prevents the leakage and unauthorized use of sensitive data, and improves the data security of the robot system.

[0020] (3) In this invention, by feeding back the dynamic compliance influence factors, regulatory domain operational entropy, and execution results during each execution process to the experience replay pool, and by using reinforcement learning algorithms to fine-tune the relevant parameters when the number of samples reaches a threshold, the compliance decision-making model is continuously optimized. This self-learning mechanism enables the system to adapt to the dynamic changes and updates of the regulatory domain, continuously improve decision-making accuracy and compliance performance, and ensure the reliability and adaptability of the humanoid robot in long-term operation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0022] Figure 1 This is a flowchart illustrating the multi-legal domain real-time compliance decision-making and data management method for humanoid robots provided in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the calculation process of the dynamic compliance influence factor in the multi-legal domain real-time compliance decision-making and data management method for humanoid robots provided in this embodiment of the invention.

[0024] Figure 3 This is a schematic diagram illustrating the calculation process of the regulatory domain operation entropy of the interactive data flow in the multi-regulatory domain real-time compliance decision-making and data management method for humanoid robots provided in an embodiment of the present invention. Detailed Implementation

[0025] This invention provides a method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots. The processing flow may include the following steps:

[0026] S1: Real-time acquisition of environmental perception data of the humanoid robot's environment, robot's physical state data, and target task instructions to be executed by the robot.

[0027] The system acquires real-time environmental perception data, robot body state data, and target task instructions for the humanoid robot's surroundings. Environmental perception data is collected by various sensors onboard the robot, describing the real-time physical environment around it. Body state data records the robot's kinematic and dynamic parameters, reflecting its current motion state and spatial pose. Target task instructions are high-level task descriptions that the robot needs to complete, issued by the upper-level control system.

[0028] S2: Based on environmental perception data, identify at least one regulatory domain in which the robot is currently located, and retrieve the regulatory constraint set corresponding to at least one regulatory domain, wherein the regulatory constraint set contains at least one behavioral compliance rule.

[0029] Based on environmental perception data, the system identifies at least one regulatory domain in which the robot is currently located and retrieves the regulatory constraint set corresponding to that domain. This regulatory constraint set contains at least one behavioral compliance rule. By analyzing the scene features and geographical location information contained in the environmental perception data, the system determines the legal jurisdiction or management area corresponding to the physical or logical area in which the robot is currently located. Each regulatory domain is pre-bound with a corresponding regulatory constraint set, which contains multiple specific behavioral compliance rules that specify the operations that the robot is allowed or prohibited from performing within that area.

[0030] S3: Based on the ontology state data, target task instructions, and regulatory constraint set, calculate the dynamic compliance impact factor of the robot in response to the target task instructions at the current moment.

[0031] The dynamic compliance impact factor of the robot in response to the target task instructions is calculated based on the robot's own state data, target task instructions, and regulatory constraints at the current moment. The dynamic compliance impact factor is a quantitative indicator used to assess the degree of compliance risk that may arise from executing the target task instructions in the current spatiotemporal context. The calculation of this factor comprehensively considers the robot's own motion state, the specific content of the task to be executed, and the rule requirements of the current regulatory domain, integrating multi-dimensional compliance impact factors into a unified numerical value.

[0032] S4: Based on dynamic compliance impact factors, decompose and reconstruct the target task instructions to generate at least one executable sub-task sequence that conforms to the set of regulatory constraints.

[0033] Based on dynamic compliance impact factors, the target task instructions are decomposed and reconstructed to generate at least one executable sub-task sequence that conforms to the set of regulatory constraints. The target task instructions are first decomposed into several smaller-granularity sub-task units. Then, using dynamic compliance impact factors as constraints, these sub-task units are recombined and ordered, eliminating sub-task units that conflict with the rules of the current regulatory domain, retaining and optimizing compliant sub-task units, and ultimately forming a task path that can be safely executed in the current environment.

[0034] S5: During the execution of the executable subtask sequence, monitor the robot's interactive data flow in real time, and calculate the regulatory domain operation entropy of the interactive data flow based on preset data privacy rules and dynamic compliance impact factors.

[0035] During the execution of an executable sub-task sequence, the robot's interactive data flow is monitored in real time, and the regulatory domain operational entropy of the interactive data flow is calculated based on preset data privacy rules and dynamic compliance impact factors. The interactive data flow refers to the set of data exchanged between the robot and external systems or devices during task execution. Data privacy rules specify the processing methods and dissemination restrictions for different types of data. Regulatory domain operational entropy is a quantitative indicator used to measure the degree of disorder or privacy leakage risk of data flow under the current regulatory domain; its value increases with the expansion of data dissemination scope and the increase in access frequency.

[0036] S6: Determine whether the entropy of the regulatory domain operation exceeds the preset safety threshold.

[0037] Determine whether the operational entropy of the regulatory domain exceeds a preset safety threshold. The safety threshold is a pre-defined critical value used to distinguish whether the data flow state is within an acceptable risk range. Compare the real-time calculated operational entropy of the regulatory domain with this safety threshold, and determine whether intervention measures are needed based on the comparison result.

[0038] S7: If the limit is not exceeded, the interactive data stream is allowed to be transmitted and processed along the original path. If the limit is exceeded, the data control policy is triggered to block, de-identify, or downgrade the interactive data stream, and output the final control command that meets the real-time compliance requirements of multiple regulatory domains.

[0039] If the data flow does not exceed the limit, the interactive data stream is allowed to be transmitted and processed according to the original path. If the data flow exceeds the limit, a data control strategy is triggered to block, anonymize, or downgrade the interactive data stream, and a final control command that complies with the real-time compliance requirements of multiple regulatory domains is output. Blocking operation means completely stopping the continued transmission of data. Anonymization operation means blurring or replacing sensitive information contained in the data. Downgrading operation means reducing the data sampling frequency or quantization accuracy. The final control command is the specific action instruction issued to the robot actuator after compliance verification.

[0040] In one possible implementation, such as Figure 2 As shown, step S3, which calculates the dynamic compliance impact factor, further includes:

[0041] S301: Extract the robot's current geographic coordinates and motion velocity vector from the body state data;

[0042] S302: Parse the task type, expected execution path, and target object attributes from the target task instructions;

[0043] S303: Filter out all behavioral compliance rules related to task type and target object attributes from the set of regulatory constraints. Each rule is associated with a predefined risk level coefficient and a violation penalty value.

[0044] S304: The dynamic compliance impact factor is calculated using the following formula. :

[0045] ;

[0046] in, The time decay coefficient, The time interval since the last compliance check. For boundary sensitivity coefficient, The shortest distance from the current geographic coordinates to the boundary of at least one regulatory domain. The preset safe distance threshold is denoted by n, where n is the number of selected behavioral compliance rules. Let be the risk level coefficient of the i-th rule. It is the maximum value among all rule risk level coefficients. The probability of violating the i-th rule is predicted based on the motion velocity vector and the expected execution path. As the baseline probability constant, As a penalty amplification factor, Let be the penalty value for violating the i-th rule.

[0047] When calculating the dynamic compliance impact factor, the robot's current geographic coordinates and motion velocity vector are first extracted from the ontology state data. The task type, expected execution path, and target object attributes are parsed from the target task instructions. All behavioral compliance rules related to the task type and target object attributes are selected from the regulatory constraint set; each rule is associated with a predefined risk level coefficient and a violation penalty value. Subsequently, a comprehensive formula is used to calculate the dynamic compliance impact factor, which integrates the time decay effect, the spatial boundary sensitivity effect, and the contributions of each rule. The time decay effect is represented by a negative exponential function, using the time interval since the last compliance check as the variable. The spatial boundary sensitivity effect is represented by an S-shaped function, using the difference between the shortest distance from the current geographic coordinates to the regulatory domain boundary and a preset safe distance threshold as the variable. The contributions of each rule are summed, with each term obtained by multiplying the normalized risk level coefficient of the rule, the logarithmic amplification term of the violation probability, and the exponential amplification term of the penalty value. The normalized risk level coefficient is the ratio of the risk level coefficient of the rule to the maximum value of the risk level coefficients of all rules. The logarithmic amplification term for the violation probability is calculated based on the probability of violating the rule predicted from the motion velocity vector and the expected execution path. The exponential amplification term for the penalty value has a base of the natural constant and an exponent of the product of the penalty amplification factor and the violation penalty value.

[0048] Specifically, the probability of violating the i-th rule The prediction process is as follows:

[0049] First, determine the corresponding spatial constraint area or behavioral constraint condition based on the content of the i-th rule, such as a geofenced area where entry is prohibited, a speed-restricted section, or a sensitive area where the collection of specific data is prohibited.

[0050] Then, based on the robot's current motion velocity vector and expected execution path, a kinematic model is used to predict the robot's motion trajectory within a preset time window in the future. This prediction takes into account the robot's maximum acceleration and turning ability, and generates a possible motion trajectory range.

[0051] Next, spatial intersection analysis is performed between the predicted trajectory and the rule-constrained region to calculate the probability that the trajectory enters the constrained region. This probability is obtained by mapping the proportion of trajectory points falling into the constrained region or the shortest distance between the trajectory and the boundary of the constrained region.

[0052] For speed-related rules, the instantaneous speed on the predicted trajectory is directly compared with the speed limit value of the rule, and the proportion of trajectory points that exceed the speed limit is the violation probability.

[0053] For data acquisition rules, the probability of acquiring sensitive data is calculated based on the time the expected execution path passes through sensitive areas and the sensor's field of view coverage.

[0054] Finally, the above analysis yields a violation probability value between 0 and 1. This value reflects the probability of violating the i-th rule under the current trend of movement.

[0055] In one possible implementation, such as Figure 3 As shown, step S5, calculating the regulatory domain operation entropy of the interactive data stream, further includes:

[0056] S501: Analyze the interactive data stream and identify at least one data element contained therein, each data element having a preset privacy sensitivity level;

[0057] S502: Based on preset data privacy rules, determine the permissible propagation range of each data element under the current regulatory domain. The permissible propagation range is represented by the maximum number of accessible nodes.

[0058] S503: Real-time monitoring of the actual access frequency of each data element per unit time and the number of nodes actually covered by the propagation.

[0059] S504: The operational entropy of the regulatory domain is calculated using the following formula. :

[0060] ;

[0061] Where m is the total number of data elements, Let F be the actual access frequency of the k-th data element, and let F be the sum of the actual access frequencies of all data elements. The frequency-range adjustment factor for the k-th data element is obtained from the privacy sensitivity level through a preset mapping function. This represents the number of nodes actually covered by the propagation of the k-th data element. This represents the maximum number of accessible nodes corresponding to the allowed propagation range of the k-th data element under the current regulatory domain.

[0062] When calculating the operational entropy of the regulatory domain, the interactive data stream is first parsed to identify at least one data element contained within it, each with a preset privacy sensitivity level. Based on preset data privacy rules, the permissible propagation range of each data element under the current regulatory domain is determined, represented by the maximum number of accessible nodes. The actual access frequency and the number of nodes actually covered by each data element per unit time are monitored in real time. Subsequently, a formula based on information entropy theory is used to calculate the operational entropy of the regulatory domain. This formula calculates the contribution of each data element separately, sums them, and takes a negative value. The contribution of each data element is obtained by multiplying its actual access frequency percentage by a weighted information term. This weighted information term includes two parts: frequency information entropy and range information entropy, which are weighted by a frequency-range adjustment factor. The frequency information entropy is calculated based on the logarithm of the actual access frequency percentage of the data element, and the range information entropy is calculated based on the logarithm of the ratio of the number of nodes actually covered by the data element's propagation to the maximum number of nodes within the permissible propagation range.

[0063] In one possible implementation, acquiring environmental perception data in step S1 specifically includes:

[0064] S101: The robot collects image data, point cloud data, and body posture data through its onboard vision sensors, lidar, and inertial measurement unit.

[0065] S102: Perform time synchronization and spatial coordinate transformation on the acquired image data, point cloud data, and body posture data to generate fused environmental perception information.

[0066] The specific process of acquiring environmental perception data includes: collecting image data, point cloud data, and body posture data through the robot's onboard vision sensors, LiDAR, and inertial measurement unit, respectively. The acquired image data, point cloud data, and body posture data are then synchronized in time and transformed in spatial coordinates to generate fused environmental perception information. Time synchronization ensures that data collected by different sensors correspond to the same moment, while spatial coordinate transformation unifies the data from different sensor coordinate systems into the robot's body coordinate system or the world coordinate system.

[0067] In one possible implementation, step S2, identifying at least one regulatory domain in which the robot is currently located, specifically includes:

[0068] S201: Input environmental perception data into a pre-trained convolutional neural network to extract semantic labels for the scene;

[0069] S202: Compare the semantic tags with the regional attributes in the high-precision map stored locally to determine the geofence, enterprise management area or temporary event control area corresponding to the robot's current physical location, as the regulatory domain.

[0070] The specific process of identifying at least one regulatory domain in which the robot is currently located includes: inputting environmental perception data into a pre-trained convolutional neural network to extract semantic labels for the scene. The convolutional neural network is pre-trained to identify scene categories with specific semantic meanings from environmental images or point clouds. The semantic labels are compared with regional attributes in a pre-stored high-precision map to determine the geofenced area, enterprise management area, or temporary event control area corresponding to the robot's current physical location; these areas are then designated as regulatory domains.

[0071] In one possible implementation, step S4 involves decomposing and reconstructing the target task instructions, specifically including:

[0072] S401: Decompose the target task instruction into a series of atomic actions based on the behavior library;

[0073] S402: Using the dynamic compliance impact factor as part of the cost function, atomic actions are sorted and filtered, atomic actions that directly conflict with any behavioral compliance rule in the regulatory constraint set are removed, and the remaining atomic actions are combined into an executable subtask sequence.

[0074] The specific process of decomposing and reconstructing the target task instructions includes: decomposing the target task instructions into a series of atomic actions based on a behavior library. The behavior library stores the basic action units that the robot can execute and their combinations. Using a dynamic compliance impact factor as part of the cost function, the atomic actions are sorted and filtered, removing atomic actions that directly conflict with any behavior compliance rule in the regulatory constraint set, and combining the remaining atomic actions into executable sub-task sequences. The cost function is used to evaluate the merits of different action sequences, and the dynamic compliance impact factor acts as a penalty for high-risk actions.

[0075] In one possible implementation, triggering the data control policy in step S7 specifically includes:

[0076] S701: Parse the source address, destination address, and data type of the interactive data stream;

[0077] S702: Based on preset data privacy rules, perform blocking operations to immediately stop the transmission of interactive data streams, or perform desensitization operations to obfuscate or replace sensitive fields in the data, or perform degradation operations to reduce the sampling frequency or quantization accuracy of the data.

[0078] The specific process of triggering data control policies includes: parsing the source address, destination address, and data type of the interactive data stream to determine the data's origin, destination, and content nature. Based on preset data privacy rules, if the operational entropy of the regulatory domain exceeds a security threshold, a blocking operation is performed to immediately halt the transmission of the interactive data stream; an anonymization operation is performed to obfuscate or replace sensitive fields in the data; or a degradation operation is performed to reduce the data's sampling frequency or quantization accuracy. The specific operation taken is determined by the data privacy rules based on the data type and the current regulatory domain.

[0079] In one possible implementation, the violation penalty value used in step S304 It is a dynamic variable, and its update process further includes:

[0080] S3041: Records the actual penalties incurred in the history of violating rule i, including the penalty amount, number of warnings, or duration of permission restrictions;

[0081] S3042: Update the violation penalty value according to the penalty record and the exponentially weighted moving average method. This allows it to reflect the average severity of penalties within the most recent time window.

[0082] The violation penalty value is a dynamic variable, and its update process further includes: recording historical penalty records for violations of rule i, including penalty amounts, number of warnings, or duration of access restrictions. Based on these penalty records, the violation penalty value is updated using an exponentially weighted moving average method to reflect the average penalty intensity within the most recent time window. The exponentially weighted moving average method assigns higher weight to recent data, ensuring that the updated value responds promptly to changes in penalty intensity.

[0083] In one possible implementation, the frequency-range adjustment factor in step S504 The privacy sensitivity level is determined in the following way:

[0084] S5041: Divide the privacy sensitivity level into several levels, each level corresponding to a preset adjustment factor baseline value;

[0085] S5042: If the privacy sensitivity level is the highest, then set... At this point, the calculation of the operational entropy of the regulatory domain is simplified to only the portion assessed by the propagation scope; if the privacy sensitivity level is the lowest level, then set... At this point, the calculation of the operational entropy of the regulatory domain is simplified to a portion evaluated only by the access frequency; for intermediate levels, It takes a value between 0 and 1, and decreases as the level increases.

[0086] The frequency-range adjustment factor is determined based on the privacy sensitivity level as follows: The privacy sensitivity level is divided into several levels, each corresponding to a preset adjustment factor baseline value. If the privacy sensitivity level is the highest, the frequency-range adjustment factor is set to zero; in this case, the calculation of the regulatory domain operational entropy is simplified to the portion evaluated only by the propagation range. If the privacy sensitivity level is the lowest, the frequency-range adjustment factor is set to one; in this case, the calculation of the regulatory domain operational entropy is simplified to the portion evaluated only by the access frequency. For intermediate levels, the frequency-range adjustment factor takes a value between zero and one, decreasing as the level increases.

[0087] In one possible implementation, after step S7, a post-optimization step is also included:

[0088] S8: Feed back the dynamic compliance impact factors, regulatory domain operation entropy, and execution effect of the final control instructions generated during this execution process to an experience replay pool;

[0089] S9: When the number of samples in the experience replay pool reaches a preset threshold, the parameters for calculating the dynamic compliance impact factor in step S3 and the parameters for calculating the operational entropy of the regulatory domain in step S5 are fine-tuned using reinforcement learning algorithms to optimize the decision-making accuracy in subsequent time steps.

[0090] The dynamic compliance impact factor, regulatory domain operational entropy, and the execution effect of the final control instruction generated during this execution process are fed back into an experience replay pool. The experience replay pool stores historical execution data for subsequent learning. When the number of samples in the experience replay pool reaches a preset threshold, a reinforcement learning algorithm is used to fine-tune the parameters for calculating the dynamic compliance impact factor in step S3 and the parameters for calculating the regulatory domain operational entropy in step S5, in order to optimize the decision accuracy at subsequent time points. The reinforcement learning algorithm analyzes the execution effects of historical data and adjusts the model parameters to achieve better decision results.

[0091] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0092] (1) In this invention, the current regulatory domain is identified by acquiring environmental perception data in real time, and the corresponding regulatory constraint set is retrieved. Combined with the ontology state data and the target task instructions, a dynamic compliance impact factor is calculated, thereby realizing a quantitative assessment of the risk of performing tasks in a multi-regulatory domain environment. This factor comprehensively considers multiple dimensions such as time decay, spatial distance, rule risk level, and cost of violation, enabling the robot to dynamically perceive changes in the regulatory environment and adjust its task execution strategy, effectively avoiding legal risks and safety accidents caused by cross-domain violations.

[0093] (2) In this invention, by monitoring the interactive data flow in real time during task execution and calculating the operational entropy of the regulatory domain based on data privacy rules and dynamic compliance impact factors, a quantitative assessment of data flow risks is achieved. When the operational entropy of the regulatory domain exceeds a preset security threshold, data control strategies such as blocking, desensitization, or downgrading are triggered in a timely manner. This mechanism ensures that the propagation and processing of data between different regulatory domains always comply with privacy requirements, prevents the leakage and unauthorized use of sensitive data, and improves the data security of the robot system.

[0094] (3) In this invention, by feeding back the dynamic compliance influence factors, regulatory domain operational entropy, and execution results during each execution process to the experience replay pool, and by using reinforcement learning algorithms to fine-tune the relevant parameters when the number of samples reaches a threshold, the compliance decision-making model is continuously optimized. This self-learning mechanism enables the system to adapt to the dynamic changes and updates of the regulatory domain, continuously improve decision-making accuracy and compliance performance, and ensure the reliability and adaptability of the humanoid robot in long-term operation.

[0095] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots, characterized in that... ,include: S1: Real-time acquisition of environmental perception data of the humanoid robot's environment, the robot's body state data, and the target task instructions to be executed by the robot; S2: Based on the environmental perception data, identify at least one regulatory domain in which the robot is currently located, and retrieve the regulatory constraint set corresponding to the at least one regulatory domain, wherein the regulatory constraint set contains at least one behavioral compliance rule; S3: Based on the ontology state data, the target task instruction, and the set of regulatory constraints, calculate the dynamic compliance impact factor of the robot in response to the target task instruction at the current moment; S4: Based on the dynamic compliance impact factor, decompose and reconstruct the target task instruction to generate at least one executable sub-task sequence that conforms to the set of regulatory constraints; S5: During the execution of the executable subtask sequence, the robot's interactive data stream is monitored in real time, and the regulatory domain operation entropy of the interactive data stream is calculated based on preset data privacy rules and the dynamic compliance impact factor; S6: Determine whether the operational entropy of the regulatory domain exceeds a preset safety threshold; S7: If the limit is not exceeded, the interactive data stream is allowed to be transmitted and processed according to the original path; if the limit is exceeded, the data control strategy is triggered to block, de-identify, or downgrade the interactive data stream, and output the final control command that meets the real-time compliance requirements of multiple regulatory domains. The calculation of the regulatory domain operation entropy of the interactive data stream in step S5 further includes: S501: Analyze the interactive data stream and identify at least one data element contained therein, each data element having a preset privacy sensitivity level; S502: Based on the preset data privacy rules, determine the permissible propagation range of each data element under the current regulatory domain, wherein the permissible propagation range is represented by the maximum number of accessible nodes; S503: Real-time monitoring of the actual access frequency of each data element per unit time and the number of nodes actually covered by the propagation; S504: The operational entropy of the regulatory domain is calculated using the following formula. : ； Where m is the total number of data elements, Let F be the actual access frequency of the k-th data element, and F be the sum of the actual access frequencies of all data elements. The frequency-range adjustment factor for the k-th data element is obtained from the privacy sensitivity level through a preset mapping function. The number of nodes actually covered by the propagation of the k-th data element. This represents the maximum number of accessible nodes corresponding to the allowed propagation range of the k-th data element under the current regulatory domain.

2. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... Step S3, calculating the dynamic compliance impact factor, further includes: S301: Extract the robot's current geographic coordinates and motion velocity vector from the body state data; S302: Parse the task type, expected execution path, and target object attributes from the target task instruction; S303: Filter out all behavioral compliance rules related to the task type and the target object attributes from the set of regulatory constraints. Each rule is associated with a predefined risk level coefficient and a violation penalty value. S304: The dynamic compliance impact factor is calculated using the following formula. : ； in, The time decay coefficient, The time interval since the last compliance check. For boundary sensitivity coefficient, The shortest distance from the current geographic coordinates to the boundary of at least one regulatory domain. The preset safe distance threshold is n, where n is the number of selected behavioral compliance rules. Let the risk level coefficient be the i-th rule. The maximum value among all rule risk level coefficients. The probability of violating the i-th rule is predicted based on the motion velocity vector and the expected execution path. As the baseline probability constant, As a penalty amplification factor, Let be the penalty value for violating the i-th rule.

3. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... The step S1 of acquiring environmental perception data specifically includes: S101: Image data, point cloud data, and body posture data are collected by the visual sensors, lidar, and inertial measurement unit mounted on the robot, respectively; S102: Perform time synchronization and spatial coordinate transformation on the acquired image data, point cloud data, and body posture data to generate fused environmental perception information.

4. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... Step S2, which identifies at least one regulatory domain in which the robot is currently located, specifically includes: S201: Input the environmental perception data into a pre-trained convolutional neural network to extract semantic labels for the scene; S202: Compare the semantic tags with the regional attributes in the high-precision map stored locally in advance to determine the geofence area, enterprise management area or temporary event control area corresponding to the current physical location of the robot, as the regulatory domain.

5. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... Step S4 involves decomposing and reconstructing the target task instructions, specifically including: S401: Decompose the target task instruction into a series of atomic actions according to the behavior library; S402: Using the dynamic compliance impact factor as part of the cost function, sort and filter the atomic actions, remove atomic actions that directly conflict with any behavioral compliance rule in the regulatory constraint set, and combine the remaining atomic actions into the executable subtask sequence.

6. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... The data management strategy is triggered in step S7, specifically including: S701: Parse the source address, destination address, and data type of the interactive data stream; S702: According to the preset data privacy rules, perform a blocking operation to immediately stop the transmission of the interactive data stream, or perform a desensitization operation to obfuscate or replace sensitive fields in the data, or perform a degradation operation to reduce the sampling frequency or quantization accuracy of the data.

7. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 2, characterized in that... The violation penalty value used in step S304 It is a dynamic variable, and its update process further includes: S3041: Record historical penalties actually incurred for violating rule i, including penalty amount, number of warnings, or duration of access restrictions; S3042: Update the violation penalty value according to the penalty record using the exponentially weighted moving average method. This allows it to reflect the average penalty intensity within the most recent time window.

8. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 3, characterized in that... The frequency-range adjustment factor in step S504 The privacy sensitivity level is determined as follows: S5041: Divide the privacy sensitivity level into several levels, each level corresponding to a preset adjustment factor baseline value; S5042: If the privacy sensitivity level is the highest level, then set At this point, the calculation of the operational entropy of the regulatory domain is simplified to only the portion assessed by the propagation range; if the privacy sensitivity level is the lowest level, then set... At this point, the calculation of the operational entropy of the regulatory domain is simplified to a portion evaluated only by the access frequency; for intermediate levels, It takes a value between 0 and 1, and decreases as the level increases.

9. The method for real-time compliance decision-making and data management across multiple legal domains for humanoid robots according to claim 1, characterized in that... Following step S7, a post-optimization step is also included: S8: Feed the dynamic compliance impact factor, the regulatory domain operation entropy, and the execution effect of the final control instruction generated during this execution process into an experience replay pool; S9: When the number of samples in the experience replay pool reaches a preset threshold, the parameters for calculating the dynamic compliance impact factor in step S3 and the parameters for calculating the operational entropy of the regulatory domain in step S5 are fine-tuned using a reinforcement learning algorithm to optimize the decision-making accuracy at subsequent times.

Images