An intelligent agent adaptive control system and method based on personal account binding cloud memory

Abstract

This invention discloses an intelligent agent adaptive control system and method based on personal account binding to cloud memory, belonging to the field of artificial intelligence and intelligent control technology. The system achieves cross-device inheritance of personalized user memories through the unique binding of accounts to cloud memory banks; the cloud-based collaborative decision-making and scheduling module operates based on the principle of "user memory-driven decision-making," balancing low latency and high-complexity tasks through flexible switching of three scheduling modes; offline memory mode ensures continuous intelligence even when offline; a memory data migration module enables the trading and inheritance of memory assets; and a brain-cerebellum collaborative interface forms a closed loop of decision-making-execution-feedback-memory update. This invention can be widely applied in humanoid robots, drones, autonomous vehicles, and other fields, possessing extremely high commercial value and strategic significance.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and intelligent control technology. Specifically, it relates to an intelligent agent adaptive control system and method based on personal account binding to cloud memory, which is applicable to personalized control and collaborative operation of various intelligent agent terminals such as humanoid robots, drones, autonomous vehicles, smart home devices, and wearable devices. Background Technology

[0002] With the development of artificial intelligence technology, various intelligent terminal devices (such as robots, drones, and self-driving cars) have gradually entered daily life and industrial production. However, existing technologies have the following shortcomings: First, they lack personalized memory capabilities. Existing intelligent agent terminals are typically "memoryless," requiring users to re-teach or reconfigure them each time they are used, making it impossible to form a behavioral model tailored to each user. When users switch to new devices, all their usage habits and experiences cannot be inherited.

[0003] Second, there is insufficient collaboration between the cloud and local systems. Existing solutions either rely entirely on cloud-based decision-making (which fails when the network is interrupted) or rely entirely on local decision-making (which cannot utilize cloud computing power), lacking a flexible multi-mode scheduling mechanism.

[0004] Third, cross-device memory cannot be shared. Multiple smart terminal devices owned by the same user (such as home robots, car assistants, and mobile phones) cannot share memory, and each device is an information silo.

[0005] Fourth, memory data cannot be transferred or inherited. Users cannot transfer their memory data to others, nor can they pass on their memories to their family members after their death.

[0006] Fifth, the lack of privacy protection mechanisms. Users' personalized memory data contains a large amount of private information, and existing solutions lack effective access control and privacy protection mechanisms.

[0007] To address the aforementioned technical problems, this invention provides an intelligent agent adaptive control system and method based on personal account binding to cloud memory. Summary of the Invention

[0008] Purpose of the invention This invention aims to solve the technical problems in the prior art, such as the lack of personalized memory in intelligent agent terminals, insufficient cloud-local collaboration, inability to share memory across devices, inability to migrate memory data, and lack of privacy protection. It provides an intelligent agent adaptive control system and method based on personal account binding to cloud memory. Technical solution

[0009] The present invention adopts the following technical solution: An intelligent agent adaptive control system based on personal account-linked cloud memory includes the following core modules: (a) User account management module: used to establish a unique binding relationship between user personal account and cloud memory instance.

[0010] (ii) Cloud memory iterative storage module: It adopts a three-layer memory architecture, including short-term memory units, long-term memory units and structured knowledge units, which respectively store user real-time interaction data, core behavioral habits and reusable task strategies.

[0011] (III) Cloud-based collaborative decision-making and scheduling module: It operates based on the user memory-driven decision-making principle, takes the user's personalized memory data as the core basis, and supports three scheduling modes: action first and calculation later, calculation first and action later, and action and calculation simultaneously.

[0012] (iv) Local data adaptive update module: used to perform basic actions and adjust parameters according to cloud instructions.

[0013] (v) Offline mode management module: used to switch to offline memory mode that calls local pre-stored memory data when the network is interrupted.

[0014] (vi) Memory Data Migration Module: Used to respond to migration commands, encrypt and package the memory data under the source account and transmit it to the target account to establish a new binding relationship.

[0015] (vii) Multi-agent collaboration module: used to support multiple intelligent agent terminals under the same user account to share or differentiate and synchronize cloud memory data.

[0016] (viii) Brain and cerebellum collaborative interface module: used to convert high-level action instructions generated by the cloud decision module into standardized motion control instructions and send them to the local motion controller.

[0017] (ix) Memory data access control module and privacy protection module: used to support differentiated permission authorization and combined with differential privacy protection to ensure user privacy and security. Beneficial effects

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. Cross-device memory inheritance: The account and memory bank are uniquely bound, so users do not need to retrain when changing devices.

[0019] 2. Three scheduling modes: based on the principle of user memory-driven decision-making, taking into account both low-latency response and high-complexity computation requirements.

[0020] 3. Continuous Intelligence Even When the Network is Offline: Offline mode ensures that personalized services can still be provided when the network is interrupted.

[0021] 4. Memory assetization: Support memory migration, inheritance, and trading to form a skills market.

[0022] 5. Multi-device collaborative evolution: All devices under the same account share learning results.

[0023] 6. Privacy Protection and Compliance: Differential privacy, zero-knowledge proof, and attribute encryption technologies are adopted to comply with relevant data security regulations.

[0024] 7. Scalable architecture: The system adopts a layered modular design. The cloud decision-making module and the memory storage module communicate through a standardized interface, without relying on a specific computing paradigm or network protocol.

[0025] 8. Human-centered architecture: Shifting from device-centric to human-centric, so that users are not locked into any hardware. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below.

[0027] Example 1: Overall System Structure This embodiment provides an intelligent agent adaptive control system based on personal account binding to cloud memory. Users register personal accounts through the user account management module, and these accounts are uniquely bound to instances in the cloud memory repository. The cloud memory iterative storage module adopts a three-layer memory architecture: short-term memory units store real-time interaction data from the most recent 24 hours, long-term memory units store core behavioral habits that have been repeated multiple times or explicitly marked by the user, and structured knowledge units store reusable task strategies.

[0028] Example 2: Complete Interaction Process Example Taking a user learning to fold a shirt using a humanoid robot as an example, the complete system workflow is as follows: Step 1: User Login and Memory Loading. Users log in to the humanoid robot with their accounts. The cloud memory database instance loads the user's historical folding behavior parameters into the local cache based on the unique binding relationship of the account.

[0029] Step 2: Real-time interaction and short-term memory generation. During the user's demonstration of folding, the robot collects joint angle and force feedback data to generate short-term memory entries.

[0030] Step 3: Automatic switching of scheduling mode. The system automatically switches between three scheduling modes—action before calculation, calculation before action, and calculation while action—based on response time requirements and task complexity.

[0031] Step 4: Long-term memory consolidation. When the same action is repeatedly performed or the user explicitly instructs to save, short-term memory is written into long-term memory units after feature extraction.

[0032] Step 5: Cross-device synchronization. Other devices under the same user account can automatically synchronize the user's structured knowledge of preferences without retraining.

[0033] Example 3: Automatic switching logic for three scheduling modes The system executes the following decision logic in each decision cycle: If the task requires a response latency of less than 30ms, the mode is switched to "action first, calculation later," with local execution and asynchronous optimization in the cloud. If the task computation volume exceeds the local computing power threshold and the allowable latency is greater than 500ms, the mode is switched to compute first and then act, waiting for the cloud to make a complete decision; In other cases, the mode switches to on-the-fly calculation, combining local sub-step execution with cloud-based streaming updates.

[0034] Example 4: Specific Implementation Method of Differential Privacy Protection Before the memory data is uploaded to the cloud, the local privacy protection module adds noise to the behavioral sequence data using the Laplace mechanism and processes the frequency statistics data using the exponential mechanism, so that the data is usable but not visible.

[0035] Example 5: Memory Migration and Inheritance Process Migration trigger condition: The source account user initiates a migration command through identity authentication, specifying the target account and migration range.

[0036] Migration execution steps: Export memory data package from the cloud → Encrypt using the target account's public key → Generate migration hash value for storage → Target account automatically decrypts and rebuilds the memory database.

[0037] Inheritance scenario: The user pre-designates a digital executor. After verification, the system transfers the memory data to the inheritor's account, and the original account is marked as inherited.

[0038] Example 6: Offline Mode Management When the network is working properly, the system automatically pre-caches frequently used user data locally. If the network is interrupted, it automatically switches to offline memory mode to continue providing personalized services. Once the network is restored, the local data is automatically merged and uploaded to the cloud.

[0039] Example 7: Multi-agent cooperation example With a single user account, multiple smart devices can be linked together. The behavioral preferences taught by the user on any device will be synchronized to the cloud, and other authorized devices can directly access them without needing to repeat the learning process.

[0040] Example 8: Brain-Cephalon Collaborative Interface The cloud-based high-level decision-making instructions are converted into standardized motion control instructions via an interface and sent to the local controller for execution. After execution, status information is returned, forming a complete closed loop of decision-making, execution, feedback, and memory update.

[0041] Example 9: Continuous Service in Network Outage Scenarios When the device enters a no-signal area, it automatically uses locally stored user preference data and executes the control strategy according to the user's habits, without restoring the factory default mode.

Claims

1. An intelligent agent adaptive control system based on personal account-linked cloud memory, characterized in that, include: The user account management module is used to establish a unique binding relationship between a user's personal account and a cloud memory instance; The cloud-based memory iterative storage module is used to save and iteratively update personalized memory data bound to the user's account; The cloud-based collaborative decision-making and scheduling module is used to allocate computing resources according to task requirements and supports three scheduling modes: action before calculation, calculation before action, and calculation while action is performed. The local data adaptive update module is used to perform basic actions and adjust parameters according to cloud instructions; The offline mode management module is used to switch to offline memory mode, which calls locally pre-stored memory data, when the network is interrupted. The memory data migration module is used to respond to migration commands, encrypt and package the memory data under the source account, and transfer it to the target account to establish a new binding relationship.

2. The system according to claim 1, characterized in that, The cloud-based memory iterative storage module adopts a three-layer memory architecture, including: a short-term memory unit for storing real-time interactive data, a long-term memory unit for storing core behavioral habits, and a structured knowledge unit for storing reusable task strategies; the three-layer memory architecture logically covers at least temporary storage functions, long-term solidification functions, and structured knowledge functions.

3. The system according to claim 2, characterized in that, The three-layer memory architecture supports the flow of data in any direction between short-term, long-term, and structured knowledge units; the flow mode includes at least one of the following: solidification flow, refinement flow, activation flow, and cyclical flow.

4. The system according to claim 1, characterized in that, The cloud-based collaborative decision-making and scheduling module automatically switches scheduling modes according to task latency requirements: low-latency tasks use the "action first, calculation later" mode, high-complexity tasks use the "calculation first, action later" mode, and regular tasks use the "action and calculation simultaneously" mode.

5. The system according to claim 1, characterized in that, The offline memory mode is implemented through a local pre-caching mechanism: when the network is normal, the user's high-frequency scene memory data is synchronized to the local machine, and when the network is disconnected, the cached data is called to provide personalized services.

6. The system according to claim 1, characterized in that, The memory data migration module supports cross-account memory inheritance scenarios, including: user-preset memory inheritance strategies, user-initiated memory gifts, and memory export and archiving before account cancellation.

7. The system according to claim 1, characterized in that, It also includes a multi-agent collaboration module, which supports multiple agent terminals under the same user account to share or synchronize cloud memory data in a differentiated manner.

8. The system according to claim 7, characterized in that, The multi-agent collaboration module also supports temporary memory fusion of agents from different users after obtaining authorization; and uses edge computing nodes as local coordinators. When the cloud connection is interrupted, the edge nodes take the lead in coordination, and end-to-end encrypted communication is used between the edge nodes and the agent terminals.

9. The system according to claim 1, characterized in that, It also includes a brain-cerebellum collaborative interface module, which is used to: convert high-level action instructions generated by the cloud collaborative decision-making and scheduling module into standardized motion control instructions, send them to the local motion controller, and receive the status feedback after execution and write it to the cloud memory iterative storage module.

10. The system according to claim 9, characterized in that, The high-level motion commands include: target pose, desired velocity, and / or task priority; the motion control commands include: target joint angle values ​​or torque commands.

11. The system according to claim 1, characterized in that, It also includes a memory data access control module, which supports users in granting differentiated memory reading permissions to third-party applications, devices or smart agents. The permissions include: one-time authorization, time-limited authorization and / or scene-triggered authorization.

12. The system according to claim 11, characterized in that, The memory data access control module employs attribute-based encryption or zero-knowledge proof technology to enable users to complete third-party authorization verification without exposing the original memory data.

13. The system according to claim 1, characterized in that, It also includes a differential privacy protection module, which adds controllable noise during the cloud-based memory iterative training process to prevent the inference of users' original privacy data from the behavioral model.

14. The system according to claim 1, characterized in that, It also includes a migration evidence module, which generates a blockchain evidence record after each memory data migration is completed. The evidence record includes at least the source account identifier, the target account identifier, the migration timestamp, and the data hash value.

15. The system according to claim 1, characterized in that, The cloud-based collaborative decision-making and scheduling module supports access to various heterogeneous computing resources, including GPU clusters, TPU clusters, supercomputers, quantum computing processors, and / or edge computing nodes; and dynamically selects the optimal computing resources based on the computational characteristics of the task.

16. The system according to claim 1, characterized in that, It also includes a multi-protocol communication module to support simultaneous access to multiple heterogeneous networks, including: 5G / 6G cellular networks, Wi-Fi, Bluetooth and / or satellite communication networks; and supports automatic switching or parallel transmission between multiple networks.

17. The system according to claim 1, characterized in that, The user account management module also supports binding user biometrics to the account. The biometrics include fingerprints, facial recognition, and / or voiceprint recognition. After the user passes biometric verification, the bound cloud memory is automatically activated.

18. The system according to claim 1, characterized in that, The binding relationship between the cloud memory instance and the user account is unique. Any intelligent agent terminal only serves as an access terminal for the cloud memory and does not directly establish an exclusive binding relationship with the user account. Multiple access terminals under the same user account share the same cloud memory instance.

19. An adaptive control method for intelligent agents based on personal account-linked cloud memory, characterized in that, include: Uniquely bind the user's personal account to the cloud memory instance; Collect user interaction data in real time and iteratively update personalized memory data in the cloud; Choose one of the following scheduling modes based on task requirements: act first, calculate later, calculate first, or calculate while acting. Perform basic actions locally, while adjusting parameters according to instructions from the cloud. When the network is interrupted, it automatically switches to offline memory mode and calls the locally cached memory data.

20. The method according to claim 19, characterized in that, It also includes access control steps: responding to access requests from third-party applications or devices, authenticating permissions based on user-preset differentiated permissions, including one-time authorization, time-limited authorization, or scenario-triggered authorization; and recording access logs.

21. The method according to claim 19, characterized in that, It also includes privacy protection steps: adding controllable noise during the cloud-based memory iterative training process to prevent the inference of users' original privacy data from the behavioral model.

22. The method according to claim 19, characterized in that, The cloud-based iterative update adopts a three-layer memory architecture: short-term memory uses a sliding window update strategy, long-term memory uses a periodic solidification mechanism, and structured knowledge units are stored and retrieved using a graph embedding method.

23. The method according to claim 19, characterized in that, It also includes a memory data migration step: after receiving the migration instruction, the selected memory data under the source account is encrypted, packaged, and transmitted end-to-end to the target account. The target account then decrypts the data and establishes a new binding relationship. The memory data migration supports two modes: incremental migration and full migration. After the migration is completed, a blockchain evidence record is generated.

24. The method according to claim 19, characterized in that, It also includes a multi-agent collaboration step: detecting the cloud connection status; if the connection is normal, the cloud decision takes precedence; if the connection is interrupted, the edge computing node is activated as a local coordinator, and end-to-end encrypted communication is used between the edge node and the agent.

25. A method for coordinated control of the cerebellum and cerebrum, applied to the system described in claim 9 or 10, characterized in that, include: The brain module generates high-level action commands based on the user's personalized memory data in the cloud memory bank; The high-level action commands are converted into motion control commands through a standardized interface; The command is sent to the cerebellum module for execution. The cerebellum module outputs status information after execution; The brain module stores the feedback information in a cloud memory bank for subsequent decision optimization.

26. The method according to claim 25, characterized in that, The process of converting high-level action commands into motion control commands includes: parsing the target pose or desired velocity in the command, combining it with a preset kinematic model, and generating target values ​​for joint angles or torque commands; the status information fed back by the cerebellum module after execution includes joint angles, torques, or task completion; after the brain module stores the feedback information in the cloud memory bank, it updates the user's personalized memory data, forming a closed loop of decision-making-execution-feedback-memory update.

27. The method according to claim 25, characterized in that, The aforementioned cerebellum-cerebrum collaborative control method is applicable to at least one of the following scenarios: gait control of humanoid robots and / or trajectory tracking of autonomous vehicles.

28. The method according to claim 19, characterized in that, Also includes: In response to a user account cancellation command, the system performs memory data export and archiving, generating a portable memory data package. The memory data package uses a standardized format and supports importing into other accounts or systems.

29. The method according to claim 28, characterized in that, The memory data packet is encrypted end-to-end before being exported, and the encryption key is bound to the biometrics of the user account. When importing the target account, it can only be decrypted after the biometrics of the target account are verified.