Physical entropy source-based embodied intelligence control system and method

Abstract

The application discloses a kind of somatic intelligent control system and method based on physical entropy source, using " one source double use " innovative framework, rely on single physical entropy source generates single entropy seed, simultaneously realizes two major core functions: on one hand, entropy seed is converted into disturbance parameter, and the dynamic disturbance of artificial intelligence model output is carried out, break the certainty output defect of traditional somatic intelligent model, effectively solve the industry pain point that agent behavior pattern is rigid, dynamic environment adaptability is poor, on the other hand, based on the same entropy seed, dynamic key is generated through one-way hash algorithm, and once one secret mode encryption transmission is realized to equipment control instruction, completely solve the security hidden danger that somatic device, intelligent terminal control instruction communication is easy to steal, easy to crack, the overall framework of the application has strong universality and adaptability, can comprehensively cover independent computing device remote control terminal, integrated robot / industrial terminal, automatic driving / smart vehicle three kinds of mainstream application forms.

Description

Technical Field

[0001] This invention relates to the fields of embodied intelligent control, artificial intelligence algorithm optimization, industrial communication security, and autonomous driving control technology, and more specifically to an embodied intelligent control system and method based on physical entropy sources. Background Technology

[0002] With the large-scale deployment of embodied intelligence, industrial robots, and autonomous driving technologies, various physical intelligent terminals have been widely used in dynamic scenarios such as industrial production, intelligent services, and vehicle driving, gradually replacing traditional manual labor to complete complex interactive tasks. Currently, embodied intelligent control systems mainly focus on the optimization and iteration of AI decision-making algorithms, motion control logic, and environmental perception technology. However, in practical engineering applications, the industry has long faced two core common pain points: rigid behavior patterns of intelligent agents and weak environmental adaptability, as well as poor communication security of control commands, which makes them easy to be cracked and tampered with. These issues severely restrict the large-scale and highly reliable commercial deployment of embodied intelligent devices. At the level of intelligent control decision-making, existing embodied intelligence models are all deterministic output architectures. After the model is trained, the same environmental input will always output the same control decisions and behavioral instructions. Such deterministic control modes can operate stably in structured and static scenarios, but they have the following drawbacks when facing unstructured disturbances, dynamic environmental changes, and unknown interaction scenarios in the real physical world: The behavior of intelligent agents is rigid and lacks flexibility. They cannot adaptively adjust their work strategies according to subtle changes in the environment, and are prone to problems such as work lag, failure of action adaptation, and low tolerance for environmental interaction. The current industry standard optimization method is to train iteratively with massive amounts of data and optimize for reinforcement learning scenarios. However, this method is highly dependent on labeled data, has high iteration costs, and limited generalization ability. It cannot break the inherent defect of deterministic output of the model from the bottom up, and it has never been able to completely solve the defects of poor adaptability to dynamic environments and rigid behavior of embodied intelligent agents. In terms of device communication security, existing intelligent terminals, industrial robotic arms, and autonomous vehicles mostly use fixed keys, periodic static keys for encryption, or plaintext transmission for control command transmission. The long-term reuse of fixed keys makes them highly vulnerable to being captured, cracked, and replicated by malicious devices, leading to tampering with control commands and unauthorized control of the terminal. Conventional dynamic encryption schemes often rely on software pseudo-random number generation for keys. Pseudo-random numbers are predictable, replicable, have low entropy, and weak security, failing to achieve true secure encryption. Furthermore, existing technologies use two independent architectures for intelligent decision optimization and communication security encryption, requiring separate random sources, algorithm modules, and hardware resources. This not only results in redundant system architecture, high resource consumption, and high power consumption, but also leads to timing discrepancies and poor adaptability when the two modules operate independently, further reducing overall system stability. Furthermore, existing technologies have not yet developed a technical solution for reusing "decision perturbation + dynamic encryption" based on a single physical entropy source. Physical entropy sources are only used in single scenarios such as random number generation and encryption key generation, resulting in extremely low resource utilization. At the same time, existing solutions lack versatility and cannot be adapted to various mainstream application forms such as remote control of independent computing devices, integrated smart terminals, and in-vehicle autonomous driving, exhibiting significant limitations in scenario adaptation.

[0003] In summary, current embodied intelligent control systems suffer from numerous technical bottlenecks, including rigid behavior, weak environmental adaptability, low communication security, redundant system architecture, poor scenario adaptability, and low resource utilization. A novel embodied intelligent control scheme based on a physical entropy source homogeneous reuse architecture is needed to simultaneously address the two core issues of intelligent decision-making flexibility and control communication security. Summary of the Invention

[0004] Addressing the shortcomings of existing technologies and common pain points in the industry, the present invention aims to provide an embodied intelligent control system and method based on physical entropy sources. The core of the invention is an innovative "one source, two uses" architecture, which relies on a single physical entropy seed to synchronously realize the dynamic encryption of model decision-making disturbances and control commands. Without increasing system redundancy, it significantly improves the environmental adaptability and control communication security of the embodied intelligent agent, while being compatible with multiple mainstream intelligent terminal forms. This solves the problems of rigid behavior, poor security, redundant architecture, and weak scene adaptability in existing technologies.

[0005] The technical solution adopted by the present invention to achieve the technical objective is: an embodied intelligent control system based on physical entropy source, the system including a physical entropy source acquisition module, an entropy seed distribution module, an intelligent decision-making disturbance module, a dynamic encryption module and a terminal control execution module; The physical entropy source acquisition module is used to acquire real random noise from the physical world as the original physical entropy source, and to perform filtering, noise reduction, and quantization processing on the original physical entropy source to generate a unique single physical entropy seed. The entropy seed processing module is connected to the physical entropy source acquisition module and is used to synchronously distribute a single physical entropy seed to the intelligent decision-making disturbance module and the dynamic encryption module to realize the reuse of signals from the same source. The intelligent decision perturbation module is used to receive physical entropy seeds, convert the entropy seeds into dynamic perturbation parameters, and adjust the output layer parameters of the embodied intelligent AI model in real time to break the deterministic output of the AI ​​model and optimize the dynamic environment adaptability of the intelligent agent. The dynamic encryption module is used to receive physical entropy seeds from the same source, perform iterative calculations on the entropy seeds based on a one-way hash algorithm, generate a real-time dynamic encryption key, and use a one-time pad encryption method to encrypt the control commands output by the AI ​​model. The terminal control execution module is used to receive encrypted control commands, complete command decryption and verification, and drive the corresponding physical terminal to perform corresponding operation actions. Furthermore, the system adopts a "one source, two uses" core architecture. Through a single entropy seed generated from the same physical entropy source, it simultaneously supports two core functions: intelligent decision-making disturbance and dynamic encryption of control commands. There are no additional redundant entropy sources or algorithm modules. The system is compatible with three types of application forms: independent computing device remote control terminals, integrated industrial intelligent terminals, and autonomous intelligent vehicles.

[0006] An embodied intelligent control method based on physical entropy sources, applied to the aforementioned control system, includes the following steps: S1. Physical Entropy Source Acquisition: Real random noise from the physical world is collected through a preset acquisition unit as the original physical entropy source. The original entropy source data is preprocessed to remove environmental interference noise and quantify to generate a standardized single physical entropy seed. S2, Same Source Signal Distribution: The generated single physical entropy seed is synchronously and in parallel distributed to the intelligent decision-making disturbance link and the dynamic encryption link to realize dual-function reuse of a single entropy source; S3, Intelligent Model Dynamic Perturbation: Based on the received physical entropy seed, a dynamic perturbation value is generated to dynamically fine-tune the output probability distribution of the Softmax function of the embodied intelligent AI model, breaking the fixed output mode of the model and generating dynamic control decisions with environmental adaptability. S4. Dynamic encryption of control commands: Based on the same physical entropy seed, a real-time dynamic key is generated through one-way hash iteration operation. The dynamic control commands generated in step S3 are encrypted frame by frame using a one-time pad mechanism. Each set of control commands corresponds to a unique key. S5. Command Transmission and Execution: The encrypted control commands are transmitted to the terminal control execution module. The terminal completes the command decryption and legality verification through the synchronization entropy verification mechanism. After the verification is successful, the terminal is driven to complete the corresponding operation.

[0007] Furthermore, the physical entropy seed is a binary data stream, which is first converted into a floating-point perturbation value through linear mapping. The mapping range is limited to [-0.05, 0.05] (which can be adaptively adjusted according to the device type) to ensure that the perturbation amplitude is reasonable and will not cause the agent to lose control.

[0008] For different behavioral dimensions (such as robotic arm displacement, robot steering, vehicle speed / obstacle avoidance, etc.), the one-dimensional entropy seed data is expanded into a perturbation array with the same output dimension as Softmax, and the probability fine-tuning is completed one by one.

[0009] In each perception-decision cycle, a brand-new perturbation factor is generated using the latest entropy seed, achieving frame-by-frame dynamic perturbation with no repetitive perturbation patterns.

[0010] Furthermore, supplemented by scenario: 1. Independent computing device + industrial robotic arm scenario The AI ​​model Softmax output dimension corresponds to the four types of actions of the robotic arm: grasping, moving, lifting, and waiting. The entropy seed generates four sets of independent perturbation values, which are used to correct the output probability of the four types of actions, allowing the robotic arm to flexibly switch between operation actions under complex working conditions.

[0011] 2. Integrated robot scenario The robot's obstacle avoidance, movement, and interaction behaviors correspond to Softmax multi-class output, and the entropy seed generated by hardware thermal noise is used to disturb the probability distribution in real time, thus realizing dynamic obstacle avoidance and flexible operation.

[0012] 3. Autonomous vehicle scenarios The Softmax function is used for decision classification such as path selection, speed adjustment, and emergency avoidance. The disturbance factor generated by vehicle sensor noise fine-tunes the decision probability and improves the decision diversity under complex road conditions.

[0013] Furthermore, the industry-standard one-way hash algorithm is selected, and the algorithm's operation logic is embedded in the hardware processing unit / embedded program; the number of hash iteration rounds and the key length can be configured according to the communication security level. The system operates with strict timing synchronization: entropy seed generation → model perturbation + key generation are executed in parallel → instruction encryption → transmission decryption, ensuring that the perturbation, encryption, and decryption use the same frame of entropy seed, with no timing deviation; One-time key rule: Regardless of whether the instruction content is repeated, the key will be updated as long as the entropy seed is updated; the physical entropy source continuously outputs random data, so the key has the inherent non-repeatability. Furthermore, supplemented by scenario: 1. Remote control scenario for independent computing devices In a remote network transmission environment, each remote control command sent to the robotic arm is independently encrypted. The hash key is generated from the ambient noise collected by the camera / microphone, which resists network packet capture and password cracking attacks.

[0014] 2. Integrated industrial terminal scenarios All calculations are performed locally on the device. Circuit thermal noise generates an entropy seed and a local key, which encrypts the control commands flowing within the device to prevent malicious local tampering, debugging, and intrusion.

[0015] 3. Autonomous vehicle scenarios Core vehicle control commands such as steering, braking, and acceleration within the vehicle CAN bus / vehicle network are encrypted frame by frame, and hash keys are generated based on vehicle sensor noise to ensure the security of the driving control link.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention features a core architectural innovation, pioneering a "one source, two uses" co-source reuse mechanism. It breaks through the limitations of the single function of traditional entropy sources, and achieves dual functions of AI model decision perturbation and control command one-time encryption through a single physical entropy seed. It eliminates the need to configure two independent random sources and algorithm architectures, greatly simplifying the system structure, reducing hardware resource consumption and operating power consumption, and solving the problems of redundancy, low resource utilization, and poor timing adaptability in existing system architectures.

[0017] 2. This invention completely solves the problem of rigid behavior of embodied intelligent agents. By perturbing the output of the AI ​​model's Softmax function through physical entropy seeds, it breaks the inherent defect of deterministic output of traditional models. Without the need for iterative training with massive amounts of data, the intelligent agent can have the ability to dynamically adjust its behavior, which greatly improves the flexibility, fault tolerance and generalization ability of the intelligent agent in unstructured and dynamically disturbed environments, and optimizes the environmental adaptability of embodied intelligence from the bottom layer.

[0018] 3. This invention has the effect of comprehensively improving the security level of control command communication. Based on the real physical entropy source combined with the one-way hash algorithm to generate dynamic keys, it realizes the one-time pad encryption mechanism. Compared with traditional fixed keys and software pseudo-random keys, it has the advantages of being unpredictable, uncopyable, and extremely unique. It can effectively prevent control commands from being stolen, tampered with, cracked, or illegally hijacked, and completely solve the security risks of industrial terminals and autonomous vehicle control communication.

[0019] 4. This invention has a highly versatile effect, covering all mainstream application forms. It is fully adaptable to three mainstream embodied intelligent applications: remote control scenarios for independent computing devices, integrated robot / industrial terminal scenarios, and autonomous intelligent vehicle scenarios. It has a wide range of applications, strong feasibility, and comprehensive protection scope.

[0020] 5. This invention avoids the defects of abstract algorithms, explicitly retains refined technical features such as Softmax function perturbation and one-time hash padding, forming a complete and engineering-implementable closed-loop technology. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of an embodied intelligent control system architecture based on physical entropy sources.

[0022] Figure 2 This is a flowchart illustrating an embodied intelligent control method based on physical entropy sources.

[0023] Figure 3 This is a flowchart illustrating the process of remote control of an independent computing device.

[0024] Figure 4This is a flowchart illustrating a system in industrial, service, and transportation scenarios. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings: Example 1:

[0026] Remote control implementation scenarios for stand-alone computing devices: This embodiment is applied to scenarios where independent computing devices such as computers and industrial control hosts remotely control robots and industrial robotic arm terminals. The physical entropy source acquisition module is connected to general acquisition units such as cameras and microphones. It collects ambient light noise and acoustic random noise as the original physical entropy source. After filtering and noise reduction and data quantization, it generates a standardized single physical entropy seed. The entropy seed processing module synchronously distributes the entropy seed to the intelligent decision-making disturbance module and the dynamic encryption module. The intelligent decision-making disturbance module uses the entropy seed to generate dynamic disturbance parameters and fine-tunes the probability distribution of the output of the Softmax function of the built-in embodied intelligent model of the industrial control equipment in real time. This breaks the fixed decision output of the model and generates dynamic control commands that adapt to the subtle disturbances in the environment for tasks such as grasping and displacement of industrial robotic arms, thus avoiding the problems of rigid robotic arm movements and poor adaptability. Meanwhile, the dynamic encryption module, based on a common entropy seed, iteratively generates a real-time dynamic key using a one-way hash algorithm. Each remote control command for the robotic arm is encrypted with a unique key, ensuring that different time sequences and different operation commands correspond to different keys. The encrypted commands are transmitted to the remote industrial terminal via the network. The terminal decrypts and verifies the commands using a synchronous entropy verification mechanism; if valid, the corresponding operation is executed. This embodiment achieves dual protection of equipment operational flexibility and remote communication security in remote control scenarios. Example 2:

[0027] Integrated robot / industrial terminal implementation scenarios: This embodiment is applied to intelligent robots and industrial integrated terminal scenarios where the acquisition module, processing unit, and execution terminal are integrated. In this scenario, no external acquisition device is required. The physical entropy source acquisition module integrated in the terminal body uses physical random signals such as thermal noise of the device's operating circuit and hardware jitter as the original entropy source to quantify and generate a single local physical entropy seed. The local synchronous distribution of entropy seeds serves two purposes. First, it perturbs the Softmax output layer of the terminal's built-in AI model, enabling the integrated robot to adaptively adjust its action strategies based on the dynamic environment during autonomous obstacle avoidance, flexible operation, and precision work. This prevents operational failures caused by fixed action patterns. Second, it generates a local dynamic encryption key based on the same entropy seed, encrypting and verifying the control commands generated locally on the terminal. This prevents malicious tampering of local commands and unauthorized control of the local terminal, achieving a closed loop of autonomous intelligent control and local security protection for the integrated terminal. This embodiment features a highly integrated architecture that requires no external devices, making it suitable for the lightweight operation requirements of embedded industrial intelligent terminals. Example 3:

[0028] Autonomous driving / intelligent vehicle implementation scenarios: This embodiment is applied to autonomous driving and intelligent vehicle control scenarios. It uses the real sensing noise and random signal disturbances generated during the operation of vehicle cameras, radar, and body sensors as the vehicle physical entropy source. After preprocessing by the vehicle processing unit, a unique physical entropy seed is generated. Dual-path multiplexing of the same-source entropy seed: The first path is used for perturbation of the autonomous driving decision model, dynamically fine-tuning the Softmax output of the vehicle path planning, obstacle avoidance control, and speed adjustment models. This breaks the fixed decision logic of the autonomous driving model, improves the vehicle's decision-making flexibility and driving adaptability in complex and sudden road conditions, and reduces the risk of decision rigidity in the autonomous driving system. The second path is used for encryption of vehicle control commands. Based on a one-way hash one-time pad mechanism, core vehicle control commands such as steering, braking, and acceleration are encrypted in real time. Each frame of driving command corresponds to a unique dynamic key, effectively preventing the security risks of vehicle communication being cracked and vehicle control commands being hijacked and tampered with, and ensuring the driving safety of autonomous vehicles.

[0029] Any technical solution that achieves the above-mentioned technical effects by utilizing the technical solutions described in this invention, or by designing similar technical solutions by those skilled in the art under the inspiration of the technical solutions described in this invention, falls within the protection scope of this invention.

Claims

1. An embodied intelligent control system based on physical entropy sources, characterized in that: It includes a physical entropy source acquisition module, an entropy seed distribution module, an intelligent decision-making perturbation module, a dynamic encryption module, and a terminal control execution module; The physical entropy source acquisition module is used to acquire real random noise in the physical world as the original physical entropy source, preprocess the original physical entropy source, and generate a single physical entropy seed. The entropy seed distribution module synchronously distributes the single physical entropy seed to the intelligent decision-making perturbation module and the dynamic encryption module; The intelligent decision perturbation module is used to generate dynamic perturbation parameters from the received physical entropy seed, and to adjust the output parameters of the embodied intelligent AI model to break the deterministic output of the AI ​​model. The dynamic encryption module is used to generate a dynamic encryption key from the same physical entropy seed, and to perform one-time one-pad encryption on the control commands output by the AI ​​model. The terminal control execution module is used to receive and decrypt the encrypted control commands and drive the physical terminal to perform the corresponding operation. The system adopts a dual-source architecture, using a single physical entropy seed generated from the same physical entropy source to simultaneously support the dynamic encryption function of intelligent decision-making disturbances and control commands, adapting to various embodied intelligent terminal application scenarios.

2. The embodied intelligent control system based on physical entropy source according to claim 1, characterized in that: The system is adapted to scenarios of remote control terminals for independent computing devices. The physical entropy source acquisition module is connected to an external camera or microphone to collect random environmental noise and generate physical entropy seeds, so as to realize remote intelligent decision optimization and encrypted transmission of remote control commands.

3. The embodied intelligent control system based on a physical entropy source according to claim 1, characterized in that: The system is adapted to integrated robot / industrial terminal scenarios. The physical entropy source acquisition module is integrated into the terminal body, which collects hardware thermal noise or operational jitter to generate physical entropy seeds, enabling the terminal to make local autonomous decision-making disturbances and local command encryption protection.

4. The embodied intelligent control system based on a physical entropy source according to claim 1, characterized in that: The intelligent decision perturbation module is specifically used to: generate dynamic perturbation values ​​based on physical entropy seeds, dynamically fine-tune the probability distribution output by the Softmax function of the embodied intelligent AI model, and optimize the agent's ability to adapt to dynamic environments. The intelligent decision-making disturbance module specifically performs the following operations: The embodied AI model's final output layer uses the Softmax function to calculate the probability distribution of each behavior category. The original output probability vector is denoted as P = [p1, p2, ..., Pn], where n is the total number of pre-defined behavior categories in the model. The standardized entropy seed output by the physical entropy source acquisition module is converted into a range-controllable random perturbation factor Δε = [ε1, ε2, ..., En]. The perturbation factor value range is preset according to the equipment operating conditions to avoid abnormal behavior. The disturbance factor is superimposed element by element on the original probability vector to obtain the corrected probability vector: P'=P+△E. The corrected probability vector is then subjected to Softmax normalization to obtain the final output probability distribution. The AI ​​model outputs control decisions based on this probability distribution.

5. This method introduces real random quantities through physical entropy sources, continuously breaking the fixed output mode of the Softmax function, enabling the agent's behavior to dynamically and adaptively adjust with the environment, thus solving the problem of rigid behavior patterns.

6. The embodied intelligent control system based on a physical entropy source according to claim 1, characterized in that: The dynamic encryption module is specifically used to: generate a real-time dynamic key based on the physical entropy seed through one-way hash iteration operation, and independently encrypt each frame of control command using a one-time pad mechanism; The dynamic encryption module uses a one-way hash algorithm to generate a dynamic key based on a homogeneous physical entropy seed, and combines this with a one-time pad mechanism to complete instruction encryption. The specific process is as follows: Key generation: Using the physical entropy seed of the current frame as the original input, a one-way hash function is called to perform multiple rounds of iterative calculation to generate a fixed-length hash digest, which is then used as the exclusive encryption key for the current frame. The one-way hash function has the property of being irreversible, so attackers cannot deduce the original entropy seed and algorithm logic from the key. Command splitting: The control commands output by the AI ​​model are split into frames, and each frame of control commands corresponds to a set of keys independently; One-time pad encryption: The plaintext of a single frame control command is XORed with the corresponding dynamic key bitwise to generate ciphertext; different frames of commands use different keys, and the keys are not repeated or reused cyclically. Synchronous decryption: After receiving the ciphertext, the terminal execution module adopts the same source synchronous entropy verification mechanism, uses the same original entropy seed to reproduce the hash key, and completes the decryption and instruction legality verification. The key in this scheme is generated entirely based on real-time physical entropy seeds, and each control command corresponds to a unique key, completely eliminating the risks of key reuse, prediction, and cracking.

7. An embodied intelligent control method based on a physical entropy source, and a system according to any one of claims 1-5, characterized in that: Includes the following steps: S1. Collect real random noise from the physical world as the original physical entropy source, preprocess the original physical entropy source, and generate a standardized single physical entropy seed. S2. The single physical entropy seed is distributed synchronously and in parallel to the intelligent decision-making perturbation link and the dynamic encryption link. S3. Based on the physical entropy seed, a dynamic perturbation value is generated to adjust the output of the embodied intelligent AI model and generate a dynamically adapted control decision. S4. Based on the same physical entropy seed, a dynamic key is generated by a one-way hash algorithm to encrypt the control command corresponding to the control decision once using a one-pad encryption. S5. Transmit the encrypted control commands to the terminal, and after decryption and verification, drive the physical terminal to perform the operation.

Images