Edge deployment method and system of multi-mode aerospace operation model on lightweight XR glasses

Abstract

The invention discloses an edge deployment method and system for a multi-modal aerospace operation model on lightweight XR glasses, relates to the technical field of augmented reality and edge computing, and particularly relates to an edge deployment method and system for the multi-modal aerospace operation model on the lightweight XR glasses. According to the method, a pre-training model is compressed to a deployable version through submerged space knowledge distillation, intention recognition, emotional value calculation and hand tremor suppression are realized on a local chip of glasses, and the delay is less than 18ms. The original multi-mode signal is destroyed immediately after being coded, and only the subsurface space characteristics are reserved. A small amount of encryption gradients are uploaded to a central server every day for model updating, and hardware fusing is triggered when data risks are detected. The system supports long-term off-line operation, integrates low power consumption and a security mechanism, and is suitable for aerospace operation scenes.

Description

Technical Field

[0001] This invention relates to the fields of augmented reality and edge computing technology, specifically to a method and system for edge deployment of a multimodal aerospace operation model on lightweight XR glasses. Background Technology

[0002] Currently, in high-precision, high-security industrial scenarios such as aerospace assembly, the use of augmented reality (AR) technology for work guidance has become a trend. However, existing AR systems, such as the Meta Quest Pro or Microsoft HoloLens 2, heavily rely on the powerful computing capabilities of cloud servers for their core functions. This architecture suffers from the following five fatal flaws, making it unable to meet the requirements of aerospace-grade applications: 1. Core Model Cloud Dependence and Data Leakage Risk: To achieve complex intent recognition and operation guidance, the core micro-world model (typically with billions of parameters) must be deployed in the cloud. This necessitates uploading sensitive physiological data such as the operator's raw EEG signals, eye movements, and hand micro-movements to the cloud for processing, posing an extremely high risk of data leakage and breach of confidentiality.

[0003] Edge device computing power is severely insufficient: Due to limitations in power consumption, weight, and heat dissipation, the edge chip computing power of lightweight XR glasses is limited, making it impossible to run a large number of aerospace micro-world models locally, which means that its core intelligence must rely on the network.

[0004] Offline environment functionality failure: In environments without network or with limited signal, such as satellite cabins and shielded workshops, once the connection with the cloud is lost, the AR system loses its core intelligent guidance capabilities, becoming "clunky as a brick" and unable to perform effective intent recognition and process path prediction.

[0005] Lack of real-time emotion computing at the edge: Existing edge devices struggle to compute complex cognitive state indicators such as emotion value functions locally in real time under low latency requirements, and cannot perform adaptive interaction and incentives based on the operator's emotions and cognitive load.

[0006] Lack of effective physical barriers to data security: Software-level encryption methods are still at risk of being breached, and existing technologies lack an immediate and irreversible hardware-level destruction mechanism that can take effect immediately when data is detected to be illegally leaving the country or when reverse engineering attempts are attempted.

[0007] Therefore, there is an urgent need in this field for an innovative technical solution that can fundamentally resolve the above contradictions, enable the secure, efficient, and offline deployment of large-scale models on ultra-lightweight edge devices, and ensure that core data never leaves the terminal. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for edge deployment of multimodal aerospace operation models on lightweight XR glasses. By edge deployment, real-time multimodal signal processing is achieved, providing operators with high-precision intent recognition, emotional stability support, and hand tremor suppression functions. Under the premise of ensuring data privacy and security, it supports long-term offline operation and low-power, high-efficiency computing, thereby improving the operational reliability and human-machine collaboration efficiency of aerospace missions.

[0009] To achieve the above objectives, this invention provides the following technical solution: a method and system for edge deployment of a multimodal aerospace operation model on lightweight XR glasses. The core of this system lies in transforming a complex multimodal interaction model into an edge computing unit that can operate stably, safely, and efficiently in the harsh environment inside a spacecraft. Specific implementation paths include: Model compression stage: Utilizing latent space knowledge distillation technology, large-scale micro-world models (typically with an extremely large number of parameters) pre-trained on massive aerospace assembly operation data are finely compressed. The goal is to control the model parameter size to below 80 million, enabling it to adapt to resource-constrained edge computing hardware while retaining core functionality.

[0010] Hardware performance specifications: The designed XR glasses must meet strict physical limitations, weighing no more than 78 grams and consuming no more than 3.5 watts. Based on this hardware, three key tasks must be performed locally in real-time: operator intent recognition, calculation of emotional value functions based on physiological signals, and suppression of operational tremors caused by muscle fatigue or microgravity environments. The end-to-end processing latency from signal input to command output must be less than 18 milliseconds to meet the high real-time requirements of aerospace operations.

[0011] Data security and privacy protection: The system has built-in strict data processing procedures. After all the collected raw multimodal signals (such as high-precision eye movement trajectory, EEG signals, EMG signals, etc.) are feature encoded locally on the glasses, their raw data will be immediately and completely erased from physical storage, retaining only the latent space abstract features that cannot be used to infer personal biometrics for subsequent calculations.

[0012] Co-evolutionary mechanism: The system participates in a secure federated learning process. Only no more than 1KB of homomorphically encrypted model gradient update data is uploaded daily to the central server for global or fleet-wide model collaborative optimization, and this process does not involve the transmission of any original or personally identifiable data.

[0013] The ultimate security defense: a system-integrated hardware-level security circuit breaker. When built-in sensors or software protocols detect unauthorized data transmission attempts or specific risk thresholds, an independent security circuit will be triggered, physically damaging the critical connection lines of the core chip to ensure that data can never be extracted.

[0014] Furthermore, the core of the latent space knowledge distillation process is the use of an energy function based on a joint embedding prediction architecture as a guiding framework. Within this framework, a large-scale teacher model and a lightweight student model are required to produce consistent latent space representations for multimodal inputs of the same space operation scenario. By minimizing the energy difference (i.e., consistency loss) between the latent space representations of the two models, the student model is driven to learn and inherit the deep discriminative rules about space operation intentions inherent in the teacher model. The direct goal of this method is to ensure that the compressed edge model achieves an accuracy of at least 95.6% in recognizing operator intentions in complex spacecraft cabin operation scenarios.

[0015] Furthermore, to achieve the aforementioned goals of high performance and low power consumption, the XR glasses employ a custom-designed neuromorphic computing chip, NeuroEdge-3. This chip utilizes an asynchronous event-driven architecture, with a single chip highly integrating three functional units: a front-end acquisition and preprocessing unit capable of processing 64 channels of EEG signals in parallel; a micro-vibration motor array drive unit controlled by 256 independent channels for providing tactile feedback; and a low-power inference engine specifically designed for calculating the emotion value function. This highly integrated design is key to meeting the weight, power consumption, and latency requirements.

[0016] Furthermore, the system design supports stable offline operation for at least 180 days in extreme environments where continuous connectivity with ground networks is impossible, such as inside satellites or space stations. This capability relies on a large-scale latent space template library stored locally on the glasses. This library is pre-loaded with over 1 billion frames of labeled multimodal operational features generated by senior aerospace operations experts (“Masters”) in various typical and emergency missions. In offline mode, the system achieves high-precision intent recognition and decision support by comparing the current operator's latent space features with the local template library in real time.

[0017] Furthermore, when the system detects that the XR glasses' battery level is below 15%, it will automatically activate an emergency operating state called "Master Mode". In this mode, the system will shut down all non-core computing functions (such as high-precision environment rendering, some advanced data analysis, etc.) and prioritize the allocation of limited power resources to core functions that ensure task safety and operator stability, namely the tremor suppression function and the basic emotional encouragement and prompting function based on emotional value calculation.

[0018] Furthermore, when the system learns through an authorized interface (such as the space center's internal childcare system) that an operator's children are in childcare, it will automatically unlock the "Family Mode" after obtaining confirmation from the operator. In this mode, the system can provide additional voice companionship content from preset family members during the operator's rest breaks or specific safe periods to alleviate the operator's psychological stress.

[0019] Furthermore, to adapt to the extreme communication latency environment of future deep space exploration missions (such as Mars bases), the system design supports data synchronization via an experimental quantum entanglement communication alternative channel. In this mode, the amount of federated learning gradient data synchronized daily through this channel is strictly limited to less than 500 bytes to ensure feasibility under extremely low bandwidth conditions and to serve as a backup plan in case of regular communication interruptions.

[0020] Furthermore, the hardware fuse-breaking mechanism is managed by a security microcontroller independent of the main computing system. When the triggering conditions are met, the microcontroller drives a miniature high-power relay, instantly passing a current far exceeding the rated value, physically burning out the dedicated fuse or specific circuit trace on the main control chip. This process permanently disables the chip's computing power and simultaneously sends a highest-priority hardware failure signal to the spacecraft's main control system ("Skyguard System"), triggering its preset "Red Shock Protocol" and initiating corresponding safety isolation and emergency plans.

[0021] Furthermore, through the aforementioned highly integrated and specialized chip design and system optimization solutions, the overall cost of edge deployment for a single pair of XR glasses has been controlled below RMB 80,000. With the achievement of mass production, this cost is expected to decrease further, enabling the technology to meet the needs of large-scale equipment applications in the aerospace sector.

[0022] Furthermore, through continuous participation in secure federated learning, the performance of models deployed across various units of the space fleet will be systematically improved. Actual deployment verification shows that the optimized models achieve an average quarterly improvement of no less than 15.2% in core mission metrics (such as intent recognition accuracy). Simultaneously, a rigorous A / B testing framework continuously monitors model updates, ensuring that while performance is improved, the rate of errors or misleading information (“illusion rate”) is strictly controlled below 5%, guaranteeing the safety and reliability of its use.

[0023] This invention provides a method and system for edge deployment of multimodal aerospace operation models on lightweight XR glasses, which has the following advantages: 1. This deployment scheme fundamentally eliminates the risk of sensitive data leakage by encoding and immediately destroying raw physiological signals locally. When used by aerospace operators in classified scenarios such as satellite assembly, all EEG, eye movement, and other biometric data are processed instantaneously within the device, completely avoiding national security vulnerabilities that may arise from traditional cloud transmission solutions. This localized processing model provides reliable technical support for the defense industry, ensuring that core process data remains physically isolated throughout the entire process.

[0024] The system boasts a fully offline operating capability of up to 180 days, effectively addressing the pain point of insufficient network coverage in the challenging aerospace environment. In shielded locations such as satellite cabins, the equipment can continuously provide precise operational guidance based on a massive locally stored expert template library, ensuring that critical processes are unaffected by network fluctuations. This feature significantly enhances the autonomy and reliability of aerospace assembly missions, ensuring the smooth progress of major engineering projects under extreme conditions.

[0025] Through innovative model compression technology, the equipment achieves significant cost optimization while maintaining high performance. The cost of a single pair of glasses has been reduced from 1.8 million yuan in the traditional solution to less than 80,000 yuan, enabling this advanced technology to be applied on a large scale in the front line of aerospace manufacturing. This cost control breakthrough breaks down the price barrier of high-end equipment and lays an industrial foundation for comprehensively improving the level of intelligence in the aerospace manufacturing industry.

[0026] This integrated family-oriented care model innovatively combines production safety with humanistic care. When the system detects that an operator's child is at the company's childcare center, it automatically provides additional voice companionship. This human-centered design effectively alleviates the pressure of high-intensity work. Practice has proven that this mechanism not only enhances employees' sense of belonging but also indirectly ensures the quality and safety of precision assembly operations through emotional stability.

[0027] The hardware-level security mechanisms construct a comprehensive protection system. Upon detecting any abnormal data transmission attempt, the system immediately initiates a physical fuse-breaking procedure, achieving irreversible hardware self-destruction by burning the main control chip's fuse. This proactive protection design forms an effective technological deterrent, fundamentally eliminating the possibility of malicious use of the equipment and establishing a solid physical barrier for core aerospace technologies. Attached Figure Description

[0028] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0029] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart illustrating the knowledge distillation and compression process of this invention; Figure 3 This is a diagram illustrating the local processing and power consumption management of this invention. Figure 4 This is a flowchart illustrating the data security and federated learning process of this invention. Figure 5 This is a flowchart of the special scenario processing of the present invention. Detailed Implementation

[0030] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses consistent with some aspects of this disclosure as detailed in the appended claims.

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] How to use: I. Startup and Basic Operations 1. Wearing and Powering On: Wear the glasses correctly, ensuring a proper fit. Press and hold the power button on the temple for 3 seconds to start the system. The system will perform a self-test.

[0033] Identity and scenario calibration: For first-time users, the "space operator" identity and current mission scenario (such as "in-cabin assembly" or "equipment inspection") must be selected via the temple touchpad or voice.

[0034] Basic Function Verification: After the system starts, it will automatically run the "Shaking Hands Inhibition" baseline test and assess the operator's current attention and emotional state baseline through a brief interaction.

[0035] II. Core Operating Mode: Aerospace Operation Assistance 1. Intent recognition assistance: During operation, the system continuously analyzes the operator's gaze, micro-gestures, and operational context.

[0036] When an operation intention (such as selecting a tool or performing an installation step) is recognized, a confirmatory prompt will be displayed on the side of the field of view with a highlighted border and brief text prompts to assist in operation decision-making.

[0037] Emotional value feedback: The system has a built-in emotion state assessment function. When it detects that the operator is focused and in a stable state, it will provide subtle positive feedback (such as a slight adjustment to the interface color scheme).

[0038] When the system detects that the operator may be fatigued, stressed, or confused, it will provide emotional support by offering gentle voice encouragement, interface prompts, or triggering micro-vibrations, based on preset strategies.

[0039] Tremor suppression: This function always runs in the background. The system processes the operator's physiological tremor signals in real time and generates reverse compensation by driving a micro motor in the frame to help maintain visual field and operational stability, without requiring manual intervention from the user.

[0040] III. Extension and Linkage Functions 1. Long-term offline operation: In environments without network access, such as inside a satellite cabin, these glasses can operate completely offline. The system utilizes a locally stored library of expert operation templates to support core functions and ensure long-term task performance.

[0041] Low-power Master Mode: The system automatically enters this mode when the glasses' battery level drops below 15%. This mode pauses or reduces the precision of non-core calculations such as intent recognition to fully ensure the operation of "shake suppression" and basic "emotional encouragement" functions.

[0042] Family-based childcare collaboration: Operators can pre-associate and authorize childcare facility information in the system settings.

[0043] When the institution's status shows that the child is in daycare, the glasses can authorize the activation of "Family Mode" when they detect that the operator is on a break or at a low workload. This mode provides short voice messages from family members or plays preset companionship content.

[0044] Remote synchronization: In environments with communication links (such as a Mars base), the glasses can automatically send a very small amount of encrypted model optimization data to a designated server in the background every day. This process is imperceptible to the user and does not affect normal use.

[0045] IV. Safety and Maintenance 1. Data security mechanism: All raw physiological and image signals collected are extracted locally on the glasses and then permanently deleted, leaving only the feature codes that cannot be used to deduce the original data.

[0046] The daily upload volume of the aforementioned encrypted model optimization data is extremely small, and it is used for collaborative optimization of the global system.

[0047] Hardware circuit breaker mechanism: This is an emergency security protocol. It will be automatically triggered when the system detects an extremely abnormal data transmission request or unauthorized physical access.

[0048] Upon triggering, the glasses' main control chip will immediately and permanently fail, stopping all functions to protect data security. The user will need to return the glasses to the factory for core module replacement.

[0049] Federated Learning and System Updates: Through a secure data synchronization mechanism, the central server can aggregate encrypted optimization information from each pair of glasses to improve the general model.

[0050] After model optimization, users will automatically receive silent quarterly feature updates when connected to the network, continuously improving the accuracy of various recognition and assistance functions. At the same time, the system will strictly monitor the reliability of the prompts.

[0051] Example: Example 1: Maintenance of Precision Equipment Inside the Satellite Cabin A space operator is performing a critical task aboard an orbiting space station: replacing a precision circuit board for a communications module. This circuit board has densely packed pins, requires extremely high installation precision, and operates in a confined space where even slight deviations could damage the equipment. The operator wears these XR goggles to enter the work area.

[0052] Upon activation, the glasses automatically load the "In-cabin Precision Repair" scenario template. The operator first calibrates the tools. When they pick up the miniature screwdriver, the glasses recognize their intention to "prepare for disassembly" through their micro-movement trajectory and gaze focus, displaying a tiny halo around the screwdriver tip as a visual reference for the operating range. During the loosening of the screw, due to muscle fatigue from prolonged weightlessness, the operator's hands experience uncontrollable physiological tremors. At this point, the glasses' built-in tremor suppression function continuously operates, generating extremely subtle compensatory reverse vibrations through a micro-motor at the end of the temples, effectively smoothing the operating trajectory and allowing the screwdriver to remain stably engaged in the screw groove.

[0053] The entire disassembly and installation process lasted approximately half an hour. During this time, the operator's breathing rate slightly increased due to heightened concentration. The glasses detected this change in physiological signal, determining that the operator was in a state of tension, but not to the point of requiring intervention. Therefore, the interface color was simply adjusted to a more soothing cool blue, providing subtle positive guidance. Ultimately, the new circuit board was installed accurately. Throughout the entire process, all operational data was processed locally on the glasses, without generating any data streams requiring external transmission, fully complying with the safety regulations for offline operations within the spacecraft.

[0054] Example 2: Spacecraft engine hoisting in the ground assembly workshop In the spacecraft assembly hall, a hoisting supervisor is overseeing the hoisting and docking of a large liquid oxygen-kerosene engine. This process requires close coordination between the supervisor and the operator on the launch tower; hand signals must be clear and accurate, as any misunderstanding could have serious consequences. The supervisor, wearing XR goggles, stands at a safe distance from the observation post.

[0055] The goggles are set to "large-scale hoisting command" mode. As the engine is slowly moved by the crane above the rocket body interface, the commander needs to continuously observe the docking process and issue instructions to the tower operator using standard hand gestures. The goggles' intent recognition system focuses on analyzing the trajectory, speed, and pauses of the commander's arm movements. For example, when the commander raises his right arm, clenches his fist, and moves it horizontally, the system immediately recognizes this as a "fine-tuning translation" instruction and overlays a virtual grid with directional arrows at the corresponding position of the engine in his field of vision, helping him to more accurately judge the distance and direction of movement.

[0056] Meanwhile, due to the extremely high noise level on site, communication relied mainly on gestures and walkie-talkies, placing significant mental pressure on the commander. The glasses continuously monitored his vital signs, determining he was under immense stress. During a crucial millimeter-level precision alignment, the commander experienced a rapid, deep breath due to tension. The glasses immediately played a pre-set, calming ambient sound effect through bone conduction headphones, accompanied by a gentle vibration to help stabilize his emotions. Ultimately, the hoisting mission was successfully completed, and all command gesture records and evaluation data were stored locally on the glasses for subsequent mission review and analysis.

[0057] Example 3: Security Protocol Triggering Exercise under Extreme Circumstances During a simulated intercontinental joint mission exercise, the information system of a participating unit was subjected to a simulated cyberattack. The attack was designed to hijack and transmit data from all connected devices. An operator is wearing XR goggles and performing a routine equipment check inside the simulation chamber.

[0058] Suddenly, the glasses received an emergency alert signal from the central security gateway, encrypted and verified multiple times, indicating that an abnormal and high-risk data outbound request pattern had been detected. This request attempted to bypass the multiple data firewalls set up locally on the glasses. Within a millisecond, the glasses' built-in security coprocessor verified the authenticity of the alert and determined that the current risk level had reached the preset "red" threshold. According to the preset procedure, the glasses immediately and automatically triggered the hardware circuit breaker mechanism without user confirmation.

[0059] The operator first heard a very faint "click" sound from inside the temples (independent relay activation), followed by an instantaneous blackout of the display screen. All functions, including basic visual enhancement, ceased completely. The glasses entered an irreversible hardware lock-in state. Subsequent inspection confirmed that the dedicated fuse on the main control chip had been physically destroyed, ensuring that any signature codes and operational templates stored within the chip could not be read or leaked. This exercise verified that, under extreme cyberattacks, the device can resolutely block data leakage paths through self-destruction, ensuring the security of core operational data. The operator then replaced the glasses module with a spare and continued the mission.

[0060] Example 4: Interaction and support with family members during long-term scientific expeditions An operator is on a six-month overwintering mission at an Antarctic research station. Communication with the outside world is limited, and the operator has a young child at home. Before departing, he set up a family care connection in his glasses system, linking it to the kindergarten his child attends.

[0061] One afternoon, the operator finished his daily equipment maintenance and went on break. Upon returning to his living quarters, the glasses, based on their activity mode, determined he was in a non-working, low-load resting state. At this time, the system detected that his associated childcare facility was in "children in childcare" status and automatically asked the operator: "We've detected that you are in your break time. Do you want to receive a message from your family?" The operator confirmed via the touchpad on the temple of the glasses.

[0062] The glasses then played a short, pre-recorded voice message from the child's family, in which the child shared a drawing they had done at kindergarten that day. Simultaneously, a low-resolution thumbnail of the drawing, uploaded by the family, was displayed on the lenses. This process lasted approximately two minutes, providing invaluable emotional comfort and support to the operator in the desolate polar environment. Throughout the interaction, all voice and image data had been pre-encrypted and downloaded to the glasses' local storage when a communication window was available. This playback did not generate any new external data requests and was completed entirely offline, satisfying the emotional needs while strictly adhering to the research station's limited communication bandwidth management regulations.

[0063] Example 5: Silent System Performance Improvement During Long-Term Task Cycles A batch of XR glasses of the same model have been operating stably for more than a quarter in multiple aerospace operation positions. In daily use, each pair of glasses strictly adheres to data security principles: the raw signal is processed locally and destroyed immediately, and only the daily generated feature gradient data, less than 1,000 bytes and highly encrypted, is automatically uploaded to the central server when there is an opportunity to connect to the internal secure network.

[0064] After receiving encrypted data from around the world, including a simulated Mars base, the server aggregates and performs calculations in a closed environment to identify optimization patterns that can improve the general performance of all glasses. For example, by analyzing tens of thousands of operational scenarios, the server may discover a more effective micro-gesture pattern for recognizing the intention to "prepare to tighten a screw".

[0065] At the end of a quarter, while operators were performing routine health checks on the intranet, the glasses automatically and silently downloaded and installed the optimized parameter update package in the background. In subsequent tasks, operators didn't perceive any significant changes, but system records showed a substantial improvement in the accuracy and response speed of intent recognition in tasks like "in-cabin equipment installation," with a decrease in false alarms and more precise and timely prompts. This gradual, data-driven optimization approach ensures that the entire glasses system can continuously improve itself over time and with accumulated tasks, while completely avoiding the operational disruptions and security risks associated with frequent large-scale software updates.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method and system for edge deployment of multi-modal space operations models on lightweight XR glasses, characterized in that, Comprise the following steps: (1) The pre-trained aerospace assembly micro-world model is compressed to a deployable version with ≤80 million parameters through latent space knowledge distillation; (2) Intention recognition, emotion value function calculation, and tremor suppression functions are completed on the local chip of the XR glasses with weight ≤78g and power consumption ≤3.5W, with a delay <18ms; (3) The original multi-modal signals are physically destroyed immediately after local encoding, only retaining latent space features; (4) Upload ≤1KB homomorphic encryption gradients to the central server for federated learning every day; (5) When data export risks are detected, trigger a hardware fuse mechanism to destroy the chip.

2. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The latent space knowledge distillation uses the JEPA energy function as the consistency loss function of the teacher-student model, ensuring that the edge model has an intention recognition accuracy ≥95.6% in aerospace scenarios.

3. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The XR glasses use the neuromorphic chip NeuroEdge-3, which integrates 64-channel EEG signal processing, 256-channel micro-vibration motor control, and emotion value function reasoning on a single chip.

4. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The satellite cabin supports completely offline operation for ≥180 days, relying on a 1 billion frame master latent space template library stored locally.

5. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: When the glasses' battery level is below 15%, it automatically enters "Master Mode", shutting down non-core computing and only retaining tremor suppression and emotion encouragement functions.

6. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The method integrates an infant care linkage mechanism that automatically unlocks the affection mode when the operator's child is in care, providing additional voice companionship.

7. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The Mars scenario supports daily synchronization of ≤500 bytes of federated gradients through quantum entanglement communication alternative channels.

8. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The hardware fuse mechanism burns the main control chip fuse through an independent relay and triggers the red shock protocol of the sky castle system.

9. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: The edge deployment cost is reduced to <80,000 yuan per pair of glasses, and mass production can achieve large-scale application.

10. The method and system of edge deployment of multi-modal aerospace operations model on light-weighted XR glasses of claim 1, wherein: After optimization, the model accuracy of federated learning improves ≥15.2% every quarter, and the illusion rate is monitored to be <5% through A / B testing.

Images