Multi-layer dynamic biometric identification, adaptive access control, and global authentication and control integrated system for physical ai-based intelligent robots

Abstract

The present invention provides a physical AI robot authentication system that generates a single robot unique ID by fusing the robot's mechanical fingerprint, behavior ID, and model DNA. The single robot unique ID is verified through a three-layer structure of cloud, edge, and on-device. In an offline environment, multi-identification reliability is ensured by performing temporary authentication based on P2P mutual communication between adjacent robots and an edge consensus algorithm.

Description

Technology Field

[0001] The present invention relates to a multi-layered identification and integrated control system for an embedded AI-based intelligent robot, and more specifically, to a system that generates a single robot unique identifier by fusing mechanical fingerprints based on mechanical deviations of the robot, behavioral pattern data, and AI model data, performs distributed authentication in cloud, edge, and on-device layered structures and peer-to-peer (P2P) mutual verification in offline environments based on the generated single robot unique identifier, and integrally provides adaptive authorization control according to environmental conditions, dynamic identifier updating based on blockchain history, and phased control measures to prepare for risk situations. Background Technology

[0002] With the recent advancement of artificial intelligence technology, the utilization of intelligent robots based on Physical AI—which perceive surrounding conditions and act autonomously in physical environments—is expanding to various industrial sites and special-purpose settings. As these intelligent robots exert direct influence on the physical environment based on advanced computational capabilities, the establishment of an accurate identification system for individual robots and a sophisticated security authentication framework is essential.

[0003] Conventional authentication systems for intelligent robots and the Internet of Things (IoT) primarily rely on the MAC address of a communication module, a static identifier assigned during manufacturing, or a physical unclonable function (PUF) at the semiconductor level to identify entities. For example, Prior Art 1 (Korean Published Patent No. 10-2023-0122642) discloses a technology for dynamically authenticating a device using a physical unclonable function; however, this is limited to the semiconductor chip level of a general IoT device and fails to present an identification mechanism utilizing mechanical operational deviations such as dynamic motor torque or joint friction coefficients of an intelligent robot. Furthermore, Prior Art 2 (Korean Published Patent No. 10-2021-7035637) discloses a method for generating an identity by combining a secret copy injected during device manufacturing with a software hash; however, since this is a static injection method at the manufacturing stage, it has limitations in complex identification that fuses dynamic physical data generated during robot operation or artificial intelligence behavior patterns.

[0004] Furthermore, most existing robot control and authentication systems are designed to operate based on a constant network connection with a cloud or a central control server. Prior art document 3 (Korean Published Patent No. 10-2019-0171024) discloses a system that authenticates a device via the cloud using a hardware security module, but it is merely a general-purpose method that relies on a separate encryption chip. Consequently, in offline environments where external network infrastructure is compromised, such as disaster rescue sites or communication blind spots, the real-time authentication process is completely halted due to the disconnection of communication with the central server. In other words, the absence of a distributed environment-adaptive edge consensus function, which verifies reliability and grants temporary work authority by directly exchanging mechanical characteristics or behavioral data between adjacent robots deployed at the site, leads to a problem where collaboration among multiple robots is restricted.

[0005] Furthermore, conventional technology lacks a system for actively controlling system access rights by reflecting physical variations, such as changes in the robot's operating environment, abnormal operation of artificial intelligence models, or component replacement. Prior art document 4 (International Patent Publication No. WO 2023014985) discloses a mechanism for controlling the operation of an artificial intelligence system, and prior art document 5 (US Patent Publication No. US 2013-0086261) discloses a system for detecting behavioral patterns and anomalies of a device by evaluating activity data. However, these primarily focus on software-based model regulation or network activity log analysis. Consequently, there is no disclosure of a dynamic control concept that extracts the unique mechanical characteristics of robot hardware to physically bind them to an artificial intelligence model execution license, or identifies hardware variations by comparing the history of physical component replacement. As a result, there are structural difficulties in fundamentally blocking unauthorized operation of artificial intelligence models on unauthorized hardware or in performing multi-stage control measures, including on-device autonomous shutdown in the event of communication failure or dangerous situations. The problem to be solved

[0006] The technical problem that the present invention aims to solve is to dynamically identify multiple elements unique to physical AI-based intelligent robots that involve physical and software variability, and to provide seamless distributed authentication through mutual verification between adjacent robots even in offline environments where the central communication network is disconnected.

[0007] Furthermore, the technical objective of the present invention is to fundamentally prevent the unauthorized use of AI algorithms and illegal copying of hardware on unauthorized hardware by combining the inherent hardware deviations of intelligent robots with the execution license of internal AI software.

[0008] Furthermore, the technical objective of the present invention is to detect the dynamic activity environment of an intelligent robot to perform situation-specific adaptive authorization control, and to detect dangerous robots based on real-time reliability obtained by weighted summing the consistency of multiple authentication elements and authentication history, thereby providing integrated step-by-step control and control measures.

[0009] Furthermore, the technical objective of the present invention is to ensure the continuity of a single unique robot ID by distinguishing between malicious tampering and normal variations through a comprehensive analysis of maintenance history and software consistency when physical changes in hardware characteristics occur due to the replacement of robot parts or aging. means of solving the problem

[0010] A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to one embodiment of the present disclosure for realizing the aforementioned tasks includes: a mechanical fingerprint generation module that generates a mechanical fingerprint (MFP) by analyzing the physical deviation of the robot; a behavioral ID generation module that generates a behavioral ID (BID) by analyzing the learned behavioral pattern of the robot; a model DNA generation module that generates a model DNA based on parameter information of the robot's internal AI model; an integrated ID fusion module that generates a unified robot identity by fusing the mechanical fingerprint, the behavioral ID, and the model DNA; and a multi-layer authentication unit that verifies the unified robot identity based on a three-layer authentication structure of cloud, edge, and on-device, and performs temporary authentication based on P2P mutual verification through mutual communication between adjacent robots in an offline environment and an edge consensus algorithm.

[0011] According to one embodiment of the present invention, the mechanical fingerprint generation module generates the mechanical fingerprint based on at least one sensor data among motor torque deviation, joint friction coefficient, sensor noise pattern, and current waveform collected in real time during a calibration operation performed during the booting of the robot or during the performance of a daily task; the behavior ID generation module generates the behavior ID by analyzing the robot's walking pattern, object manipulation habit, balance maintenance method, and environment exploration pattern as time-series data, and adaptively updates the behavior ID by reflecting behavioral changes according to the learning of the AI ​​model; and the model DNA generation module generates the model DNA by combining the parameter hash value of the AI ​​model, the architecture signature, and the learning data source information, updates the hash value when the parameters of the AI ​​model are updated, and can generate a tampering event in the event of unauthorized change.

[0012] According to one embodiment of the present invention, the multi-layer authentication unit performs global policy and authorization management and cross-reference with the entire ID database at the cloud layer, performs regional robot group authentication at the edge layer, and performs real-time self-verification at the on-device layer, wherein each layer operates independently so that authentication can be maintained at the remaining layers even in the event of a failure. Additionally, the edge consensus algorithm performs group authentication at the robot team level by having multiple adjacent robots exchange mechanical fingerprints and behavior IDs in an offline environment where communication with the central server is disconnected, and can grant temporary work authority only when a preset number of robots or more agree on mutual verification.

[0013] A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to one embodiment of the present invention may further include an adaptive authorization control module that detects the activity environment of the intelligent robot and dynamically assigns a contextual ID, and differentially applies motor output, movement speed, and functional access rights according to the environment, and a trust score module that calculates a real-time trust score of the robot by weighted summing the degree of match of the mechanical fingerprint, the degree of match of the behavior ID, the integrity of the model DNA, maintenance history, and authentication history, and dynamically adjusts the authority assigned according to the calculated trust score.

[0014] A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to one embodiment of the present invention may further include a hardware-software binding module that combines the mechanical fingerprint with the execution license of the AI ​​model to allow the operation of the AI ​​model algorithm only on registered unique hardware and blocks operation when an attempt is made to execute it on unauthorized hardware.

[0015] A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to one embodiment of the present invention may further include a hybrid ID update module that, when the mechanical fingerprint value changes beyond an allowable error range due to the aging or replacement of parts of the intelligent robot, compares it with the maintenance history recorded on the blockchain and comprehensively verifies the consistency of the behavior ID and the model DNA to distinguish between malicious tampering and natural aging or normal replacement, and updates the mechanical fingerprint while ensuring the continuity of the single robot unique ID when determined to be a normal variation.

[0016] A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to one embodiment of the present invention may further include a global kill switch unit that sequentially performs four-stage control measures, including warning, function restriction, remote emergency stop, and complete network blocking, for a dangerous robot in which tampering is detected as a result of verification by the multi-layer authentication unit or the reliability score of the robot is below a threshold value, wherein the control command of the global kill switch unit is transmitted through all three layers of cloud, edge, and on-device, and can perform an autonomous stop function based on the on-device self-determination of the intelligent robot in the event of communication interruption. Effects of the invention

[0017] According to the present invention, by generating a single unique robot ID through the fusion of a mechanical fingerprint quantifying the physical deviation of a robot, a behavior ID analyzing learned behavior patterns, and a model DNA based on artificial intelligence model parameters, it is possible to precisely identify attempts at physical replication of hardware and unauthorized theft of artificial intelligence software. In particular, through hardware-software binding that combines the mechanical fingerprint and the execution license of the artificial intelligence model, the operation of algorithms on unauthorized devices other than the registered unique hardware can be fundamentally blocked, thereby preventing the leakage of core technology and data.

[0018] In addition, by applying a three-layer authentication structure of cloud, edge, and on-device, single points of failure can be eliminated and the availability of system authentication can be improved. Even in offline environments where communication with the central server is disconnected, collective authentication is performed through P2P mutual verification between adjacent robots and edge consensus algorithms, thereby enabling stable operation of robot missions without interruption or delay, even in communication blind spots or disaster situations.

[0019] In addition, the robot's activity environment is detected to dynamically assign a situation layer, and a real-time reliability score can be calculated by weighted summing the consistency of multi-identification data and authentication history. By dynamically applying differential motor output, movement speed, and functional access rights to the robot based on the calculated reliability score and the situation layer, unnecessary system resource consumption can be reduced and the physical safety of the surrounding work environment can be ensured.

[0020] Furthermore, if the mechanical fingerprint value exceeds the allowable tolerance range due to the robot's natural aging or parts replacement, natural variation and malicious tampering can be accurately distinguished by comparing it with the maintenance history recorded on the blockchain and verifying the consistency of the behavior ID and model DNA. In other words, since the continuity of a single robot's unique ID is guaranteed through an automated identifier update procedure, maintenance and management costs can be reduced in industrial environments operating large-scale robot swarms.

[0021] Consequently, for dangerous robots in which tampering is detected as a result of multi-layer verification or the reliability score drops below a threshold, a global kill switch unit can be immediately activated to sequentially perform four-stage control measures: warning, function restriction, remote emergency stop, and complete network disconnection. By controlling the system to execute an autonomous stop function based on on-device judgment in the event of communication interruption, the spread of secondary physical damage caused by hacking or artificial intelligence malfunction can be prevented, and an integrated safety network for the operation of large-scale intelligent robots can be established.

[0022] The various technical and economic effects of the present invention as described above are presented as examples to aid in understanding the invention, and the embodiments according to the present invention do not limit the scope of the invention, nor is the scope of the rights of the present invention limited to the specific effects presented. Brief explanation of the drawing

[0023] FIG. 1 is an overall configuration diagram showing a global authentication and control integrated system for a physical AI-based intelligent robot according to one embodiment of the present invention. FIG. 2 is a block diagram illustrating a single robot unique ID generation process according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the operation of a multi-layer authentication and offline P2P edge consensus algorithm according to one embodiment of the present invention. FIG. 4 is a conceptual diagram illustrating adaptive authority control and hardware-software binding operations according to an embodiment of the present invention. FIG. 5 is a flowchart showing the four-step control measures of the hybrid ID update and global kill switch unit according to one embodiment of the present invention. Specific details for implementing the invention

[0024] Preferred embodiments of the present invention will be described in more detail, but it is clarified that the presented content is illustrative and not limited thereto. Terms such as "computer program," "component," "module," and "system" used in this specification refer to entities implemented in hardware, software, or a combination thereof, and may be used interchangeably. Furthermore, the term "or" used in this specification is used in a comprehensive sense rather than an exclusive sense, and the order of steps described in the claims does not limit the scope of the invention. In addition, each component of the invention described below may be driven by a high-performance processor such as a GPGPU (General-Purpose computing on Graphics Processing Units), TPU (Tensor Processing Unit), or NPU (Neural Processing Unit) to process deep learning and artificial intelligence computations at high speed.

[0025] The key terms used in this specification are defined as follows.

[0026] In this specification, "Mechanical Fingerprint (MFP)" refers to identification data that quantifies minute inherent deviations in physical hardware, such as motor torque deviation, joint friction coefficient, and sensor noise of a robot.

[0027] In this specification, "Physical AI" collectively refers to Human AI robots, Humanoid robots, and robots equipped with Embodied AI.

[0028] In this specification, "Behavioral ID (BID)" refers to an identifier obtained by analyzing unique behavioral characteristics, such as a robot's walking patterns and object manipulation habits, in a time series according to AI model learning.

[0029] In this specification, "Model DNA" refers to identification information that proves the uniqueness and integrity of AI software by combining the parameter hash value, architecture signature, etc., of an AI model inside a robot.

[0030] In this specification, "Unified Robot Identity" refers to the integrated identifier of a robot generated by fusing the mechanical fingerprint, behavior ID, and model DNA.

[0031] In this specification, "Contextual ID" refers to identification information that is dynamically assigned by detecting the activity environment of a robot and provides a standard for authorization control based on the environment.

[0032] In this specification, "edge consensus algorithm" refers to an algorithm that performs collective authentication by having multiple adjacent robots exchange mutual identification information in an offline environment where communication with a central server is disconnected.

[0033] FIG. 1 is an overall configuration diagram of a global authentication and control integration system for a physical AI-based intelligent robot according to one embodiment of the present disclosure. Referring to FIG. 1, a dynamic biometric multi-layer identification, adaptive authorization control, and global authentication and control integration system for a physical AI-based intelligent robot according to one embodiment of the present disclosure for realizing the aforementioned objectives comprises an integrated ID generation unit (100), a multi-layer authentication unit (200), an adaptive authorization control module (300), a hardware-software binding module (400), a hybrid ID update module (500), and a global Kill Switch unit (600). The integrated ID generation unit (100) functions as a top-level module that generates a single identifier by fusing the hardware, software, and behavioral characteristics of the robot. The multi-layer authentication unit (200) flexibly verifies the legitimacy of the robot according to cloud, edge, and on-device network conditions. The adaptive authorization control module (300) detects the robot's activity environment, assigns a situation layer, and applies permissions differentially, while the hardware-software binding module (400) combines a mechanical fingerprint and an AI model execution license to block unauthorized hardware operation. Additionally, the hybrid ID renewal module (500) ensures the continuity of the unique ID by comparing it with the blockchain history when parts are replaced, and the global kill switch unit (600) performs four-stage control measures upon detection of a dangerous robot. This system fundamentally blocks threats that may occur when a physical AI robot operates in a complex physical environment and enables efficient global control.

[0034] FIG. 2 is a block diagram illustrating a process for generating a unique single robot ID according to an embodiment of the present disclosure. As shown in FIG. 2, the integrated ID generation unit (100) includes a mechanical fingerprint generation module (110), a behavior ID generation module (120), a model DNA generation module (130), and an integrated ID fusion module (140). The mechanical fingerprint generation module (110) generates a mechanical fingerprint by analyzing the physical deviation of the robot. In one embodiment, the mechanical fingerprint generation module (110) extracts at least one sensor data among motor torque deviation, joint friction coefficient, sensor noise pattern, and current waveform collected in real time during calibration operations performed when booting the robot or during routine task performance. Even if parts are produced in the same manufacturing process, unique physical deviations occur due to minute assembly errors, and the mechanical fingerprint generation module (110) analyzes this using signal processing techniques such as Fast Fourier Transform (FFT) to produce a numerical mechanical fingerprint.

[0035] The behavior ID generation module (120) generates a behavior ID by analyzing the learned behavior patterns of the robot as time-series data. Even if an intelligent robot equipped with physical AI is equipped with the same basic AI model, it exhibits a unique behavior trajectory as it performs reinforcement learning in different work environments. The behavior ID generation module (120) analyzes joint angle changes and acceleration data related to the robot's walking pattern, object manipulation habits, balance maintenance methods, and environment exploration patterns using a Long Short-Term Memory (LSTM) network or a Transformer-based deep learning algorithm. Through this, the behavior ID generation module (120) adaptively updates the behavior ID by reflecting behavioral changes resulting from the continuous learning of the AI ​​model.

[0036] The above model DNA generation module (130) generates model DNA based on parameter information and architecture signatures of the AI ​​model inside the robot. According to one embodiment of the present invention, the model DNA generation module (130) applies a cryptographic hash function to the main layer weights of the neural network model and derives model DNA by combining this with the model's architecture signature and training data source information. The model DNA generation module (130) updates the hash value when AI parameters are officially updated through federated learning, etc., but immediately generates a tampering event if the model weights are changed without authorization due to hacking.

[0037] Next, the integrated ID fusion module (140) fuses the generated mechanical fingerprint, behavior ID, and model DNA to generate a single robot unique ID. As an embodiment of the present disclosure, the case of a 6-axis articulated robot arm deployed in a manufacturing plant can be assumed. Even if multiple robot arms of the same model are deployed on the same line, the integrated ID fusion module (140) extracts the vibration frequency of the first axis motor of each robot as a mechanical fingerprint, the unique component operation trajectory as a behavior ID, and the parameter hash finely tuned to the environment as model DNA, and vector-combines them to issue a single robot unique ID specific to that robot. This multi-element combination makes it impossible to forge the legitimacy of the entire single robot unique ID through the replication of specific hardware components or software theft alone.

[0038] As another embodiment of the present disclosure, the case of a medical assistive robot or a home care robot deployed in a hospital or home can be considered. Since these robots interact closely with patients or users, high reliability and privacy protection are required. In such an embodiment, the 'Single Robot Unique ID' of the present system may be linked with a 'Robot Unique ID' issued by a separate blockchain-based authentication system. Specifically, the identification value of the Robot Unique ID system, generated based on a hardware integrity proof combining dynamic biometric data (motor current waveforms, sensor fusion noise, etc.) and static PUF values, can be mapped to the Single Robot Unique ID of the present system. In this process, by recording a Zero Knowledge Proof (ZKP) value in the blockchain ledger instead of the original identification data, the leakage of personal information from robot data is prevented. Furthermore, by providing an extended function that manages the robot's lifecycle history—such as registration, renewal, transfer, and disposal—in a way that prevents tampering, the scope of rights protection and authentication transparency can be enhanced.

[0039] FIG. 3 is a flowchart of the operation of a multi-layer authentication and offline P2P edge consensus algorithm according to an embodiment of the present disclosure. Referring to FIG. 3, the multi-layer authentication unit (200) verifies the single robot unique ID based on a three-layer authentication structure of cloud, edge, and on-device. In the cloud layer, the highest level of security verification is performed by comparing the entire ID database along with global policy and authorization management. In the edge layer, robot group authentication is performed through an edge server built within a specific regional network to alleviate cloud communication delays and bottlenecks. In the on-device layer, the matching of identification information is performed in the robot's internal security area through real-time self-verification. Each layer is designed to operate independently so that even if a failure occurs in the upper layer network, authentication can be maintained without interruption in the remaining layers.

[0040] In another embodiment of the present invention, the operation is described in the case where multiple intelligent robots deployed to a fire suppression or disaster rescue site are placed in an offline environment where communication with a central server is cut off. When the wide-area communication network is disconnected, the multi-layer authentication unit (200) performs temporary authentication through P2P mutual verification via mutual communication between adjacent robots based on an edge consensus algorithm. The rescue robots at the disaster site form a short-range wireless communication network and exchange encrypted fragments generated based on their respective mechanical fingerprints and behavior IDs by broadcasting them. The multiple robots perform collective authentication at the robot team level based on the exchanged identification information, and grant temporary work authority to the robot only when a pre-configured majority of robots agree on mutual verification that the identification information of a specific robot is valid. In this way, even in the event of hostile jamming or physical network disconnection, only authorized robots can continue safe cooperative missions under mutual trust.

[0041] FIG. 4 is a conceptual diagram illustrating adaptive authority control and hardware-software binding operations according to an embodiment of the present disclosure. As shown in FIG. 4, the adaptive authority control module (300) detects the activity environment of an intelligent robot, dynamically assigns a situation layer, and differentially applies authority. The adaptive authority control module (300) determines the current environment in real time based on vision sensor and LiDAR sensor data mounted on the robot. For example, in the case of a delivery robot moving inside a logistics center, the adaptive authority control module (300) recognizes that it is a controlled environment and assigns a situation layer that allows the use of maximum motor output and high-speed movement. However, when it is detected that the robot leaves the logistics center and enters an external road where ordinary pedestrians are present, the situation layer is immediately switched to a pedestrian zone, the movement speed is limited to a safe speed, and the priority of the AI ​​inference function for avoiding collisions with humans is raised.

[0042] In addition, the reliability score module (310) calculates the real-time reliability score of the robot by weighted summing the consistency and history of various authentication elements. The reliability score module (310) updates the continuous reliability score by reflecting in real-time the consistency of the mechanical fingerprint, the consistency of the behavior ID against past patterns, the integrity of the model DNA, recent maintenance history, and multi-layer authentication history. If the calculated reliability score falls below a threshold, the adaptive authority control module (300) temporarily revokes the high-risk physical operation authority of the robot or dynamically adjusts the autonomous judgment mode to a manual control mode.

[0043] Additionally, the hardware-software binding module (400) cryptographically combines the mechanical fingerprint with the execution license of the artificial intelligence model to allow execution only on registered unique hardware. The hardware-software binding module (400) measures the mechanical fingerprint of the current hardware via the system bus and compares it in real-time with the original reference value signed within the license file. If an industrial spy steals high-performance AI software and attempts to execute it on unauthorized hardware, it detects the mechanical fingerprint mismatch and immediately blocks the execution of the AI ​​algorithm at the kernel level to prevent technology leakage.

[0044] FIG. 5 is a flowchart of the four-step control measures for the hybrid ID renewal and global kill switch unit according to one embodiment of the present disclosure. Referring to FIG. 5, the hybrid ID renewal module (500) operates when the mechanical fingerprint value changes beyond an allowable error range (e.g., a deviation rate of 5% or more compared to the initial setting) due to the aging of the intelligent robot or replacement of parts. When the mechanical fingerprint changes because the robot's joint parts are worn out or the core motor is replaced for repair, the hybrid ID renewal module (500) performs a comparison operation by querying the repair shop maintenance history and parts order history distributed and recorded on the blockchain network. At the same time, it comprehensively verifies whether the consistency of the behavior ID and model DNA remains the same as before, thereby clearly distinguishing between attempts at malicious hardware tampering and natural aging or normal parts replacement. If it is determined to be a normal change, a link guaranteeing the continuity of the single robot unique ID is left on the blockchain to maintain the connection with the previous ID, and the mechanical fingerprint is safely renewed to match the new hardware physical characteristics.

[0045] The Global Kill Switch (600) issues a four-stage control measure to a dangerous robot in which serious tampering is detected as a result of verification by the multi-layer authentication unit (200) or the reliability score calculated by the reliability score module (310) is below a preset threshold. The Global Kill Switch (600) can sequentially or immediately, depending on the severity of the issue, perform a first-stage warning measure to propagate the dangerous state to the target robot, the central control center, and surrounding robots; a second-stage function limitation measure to minimize motor torque and movement speed to reduce the physical destructive power of the robot; a third-stage remote emergency stop measure to hardware-cut off power or activate the brake if abnormal behavior persists; and a fourth-stage complete network blocking measure to physically and logically isolate all communication networks to prevent the spread of malicious code or data leakage. These control commands from the Global Kill Switch (600) are designed to be transmitted redundantly via multiple paths through all three layers: cloud, edge, and on-device. In addition, even in extreme hacking situations where communication with the outside is completely cut off, the system is configured to forcibly execute an auto-stop function when critical tampering is detected by judgment logic within the on-device security area, thereby perfectly guaranteeing fail-safety for the entire system.

[0046] The methods and systems according to the various embodiments of the present invention described above are not intended to limit the scope of the patent claims, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. The operation and step sequence of each module described in this specification are intended to explain the flow of functional logic and are not necessarily limited to a chronological order; they may be performed in parallel or the order may be changed according to technical necessity. Accordingly, the scope of protection of the present invention is determined by the patent claims set forth below rather than by the detailed description of the embodiments above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0047] 100: Integrated ID Generation Section 110: Mechanical fingerprint generation module 120: Behavior ID generation module 130: Model DNA Generation Module 140: Integrated ID Fusion Module 200: Multilayer Authentication Unit 300: Adaptive Authority Control Module 310: Reliability Score Module 400: Hardware-Software Binding Module 500: Hybrid ID Renewal Module 600: Global Kill Switch Division

Claims

Claim 1 A mechanical fingerprint generation module that generates a mechanical fingerprint (MFP) by analyzing the physical deviation of a robot; a behavioral ID generation module that generates a behavioral ID (BID) by analyzing the learned behavioral patterns of a robot; a model DNA generation module that generates model DNA based on parameter information of an internal AI model of a robot; an integrated ID fusion module that generates a unified robot identity by fusing the mechanical fingerprint, the behavioral ID, and the model DNA; and a multi-layer authentication unit that verifies the unified robot identity based on a three-layer authentication structure of cloud, edge, and on-device, while performing temporary authentication based on P2P mutual verification through mutual communication between adjacent robots in an offline environment and an edge consensus algorithm. A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot, comprising: a mechanical fingerprint generation module that generates the mechanical fingerprint based on at least one sensor data among motor torque deviation, joint friction coefficient, sensor noise pattern, and current waveform collected in real time during a calibration operation performed at the booting of the robot or during the performance of a daily task; a behavior ID generation module that generates the behavior ID by analyzing the robot's walking pattern, object manipulation habit, balance maintenance method, and environment exploration pattern as time-series data, and adaptively updates the behavior ID by reflecting behavioral changes according to the learning of the AI ​​model; and a model DNA generation module that generates the model DNA by combining the parameter hash value of the AI ​​model, an architecture signature, and learning data source information, and updates the hash value when the parameter of the AI ​​model is updated and generates a tampering event when an unauthorized change occurs. Claim 2 delete Claim 3 A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to claim 1, wherein the multi-layer authentication unit performs global policy and authority management and cross-reference with the entire ID database at the cloud layer, performs regional robot group authentication at the edge layer, and performs real-time self-verification at the on-device layer, wherein each layer operates independently so that authentication is maintained at the remaining layer even in the event of a failure, and the edge consensus algorithm performs group authentication at the robot team level by having multiple adjacent robots exchange the mechanical fingerprint and the behavior ID in an offline environment where communication with the central server is cut off, and grants temporary work authority only when a preset number of robots or more agree on mutual verification. Claim 4 A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot, further comprising: an adaptive authorization control module that detects the activity environment of the intelligent robot and dynamically assigns a contextual ID, and differentially applies motor output, movement speed, and functional access rights according to the environment; and a trust score module that calculates a real-time trust score of the robot by weighted summing the match of the mechanical fingerprint, the match of the behavior ID, the integrity of the model DNA, maintenance history, and authentication history, and dynamically adjusts the rights assigned according to the calculated trust score. Claim 5 A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot, wherein, in claim 1, the system further comprises a hardware-software binding module that combines the mechanical fingerprint with the execution license of the AI ​​model to allow the operation of the AI ​​model algorithm only on registered unique hardware and blocks operation when an attempt is made to execute it on unauthorized hardware. Claim 6 A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to claim 1, further comprising a hybrid ID update module that, when the mechanical fingerprint value changes beyond an allowable error range due to aging or parts replacement of the intelligent robot, compares it with the maintenance history recorded on the blockchain and comprehensively verifies the consistency of the behavior ID and the model DNA to distinguish between malicious tampering and natural aging or normal replacement, and updates the mechanical fingerprint while ensuring the continuity of the single robot unique ID when determined to be a normal variation. Claim 7 A dynamic multi-factor identification and global authentication system for a physical AI-based intelligent robot according to claim 1, further comprising a global kill switch unit that sequentially performs four-stage control measures—warning, function restriction, remote emergency stop, and complete network blocking—for a dangerous robot in which tampering is detected as a result of verification by the multi-layer authentication unit or the reliability score of the robot is below a threshold, wherein the control command of the global kill switch unit is transmitted through all three layers of cloud, edge, and on-device, and performs an autonomous stop function based on the on-device self-determination of the intelligent robot in the event of communication interruption.

Images