A deterministic state continuous execution method and system for autonomous intelligent system and storage medium

Abstract

The application discloses a kind of determinacy state continuous execution methods, systems and storage medium for autonomous intelligent system.The method comprises: obtaining target state information generated by autonomous intelligent system, and state entity is constructed based on target state information, execution intention information, state identity identification information and execution proof association information;According to the authorization judgment of target state, at least including physical constraint space, system running constraint space and intelligent decision constraint space, multiple continuous migration states are converted from the target state after authorization, so that the change between adjacent migration states meets the preset maximum change constraint;State identity identification and migration state are continuously associated during migration process, and execution proof chain based on previous association is generated.The application realizes the determinacy continuous execution control of autonomous intelligent system target state before entering physical execution link by the above technical scheme.

Description

Technical Field

[0001] This invention relates to the field of autonomous intelligent system execution control technology, and in particular to a method, system and storage medium for deterministic constraints, continuous migration, identity binding and execution proof of the state change process before the target state generated by the autonomous intelligent system enters the physical execution link.

[0002] This invention can be applied to, but is not limited to, the following scenarios: artificial intelligence agent execution control, heterogeneous computing system state management, robot motion control, autonomous driving decision execution, spacecraft autonomous control, industrial control systems, and energy system state management. Background Technology

[0003] Autonomous intelligent systems (such as AI agents, autonomous driving systems, robot control systems, and aerospace autonomous control systems) are able to autonomously generate control strategies or target states based on environmental perception and mission objectives. With the rapid development of artificial intelligence technology, autonomous intelligent systems are moving from the "suggestion generation" stage to the "direct execution" stage.

[0004] However, existing technologies generally suffer from the following technical deficiencies when the target state generated by an autonomous intelligent system enters the physical execution link:

[0005] 1. Lack of authorization constraint mechanism before the target state enters the execution layer

[0006] Existing systems typically assume that the target state generated by the autonomous intelligent system is the executable state, directly sending it to the actuator or lower-level controller. When the autonomous intelligent system generates an unreasonable target state due to model bias, environmental changes, or boundary condition variations, this state may still directly enter the execution chain, lacking a multi-layered constraint adjudication mechanism before the target state enters physical execution. For example, existing security mechanisms mostly involve anomaly detection during execution or result verification after execution, rather than constraining and authorizing the target state itself during the execution admission phase.

[0007] 2. Lack of continuity constraints for state changes

[0008] Existing systems typically transition directly from the current state to the target state during state transitions, or use simple interpolation / smoothing algorithms. These methods only focus on the rate of change of state values ​​and do not establish deterministic continuous transition constraints at the state transition path level. In scenarios involving high dynamic response, high energy systems, or safety-critical scenarios, unconstrained state transitions can lead to system instability, actuator overload, or irreversible physical damage.

[0009] 3. Lack of identification and traceability of state change processes.

[0010] Existing systems typically only record changes in state values ​​without assigning a persistent identifier to the state change process. When the system state changes at multiple points in time, it is difficult to determine whether the preceding and following states belong to the same execution intent, and it is also impossible to establish a causal relationship between the state changes. This is particularly prominent in scenarios involving multi-agent collaboration, long-duration task execution, or fault attribution.

[0011] 4. Lack of verifiable proof of state execution

[0012] Existing system logs are primarily used for post-event auditing, and their evidentiary strength relies on credible assumptions at the software layer. When it is necessary to prove that a state change has indeed undergone authorization, constraints, and continuous migration, existing log mechanisms cannot provide a verifiable chain of execution proof. In safety-critical systems (such as autonomous driving, spacecraft, and industrial control systems), the verifiability of the execution process is becoming a fundamental requirement for regulatory review and liability determination.

[0013] 5. Existing trusted computing technologies do not solve the problem of trusted state execution.

[0014] Existing trusted computing technologies such as Trusted Execution Environments (TEEs), Trusted Platform Modules (TPMs), and Secure Boot primarily address the trustworthiness of the program's runtime environment, platform identity, and startup chain. Their protection targets the computing environment and static data, rather than dynamic state changes. Specifically, Secure Boot addresses startup trustworthiness, TPM addresses platform identity trustworthiness, and TEE addresses runtime environment trustworthiness. However, none of these technologies address whether the target state generated by an autonomous intelligent system is authorized during execution, whether it undergoes continuous migration, whether identity continuity is maintained, or whether execution proof can be generated. Therefore, a clear technological gap exists between existing trusted computing technologies and the execution trustworthiness requirements of autonomous intelligent systems.

[0015] 6. The need for reliable state execution arising from the large-scale deployment of autonomous intelligent systems

[0016] With the development of artificial intelligence models, intelligent agent systems, and autonomous decision-making technologies, the application scope of autonomous intelligent systems is gradually expanding from information processing scenarios to execution scenarios with complex physical constraints. In such systems, autonomous intelligent modules may simultaneously affect multiple heterogeneous resources or physical execution objects, including but not limited to the states of computing resources, energy, mechanical motion, spatial operation, and industrial processes. Because different execution objects have different physical constraints, operational constraints, and security boundaries, the target state generated by the intelligent system cannot be directly equated to an executable state. Existing technologies typically optimize from the perspectives of improving computing power, task scheduling, control algorithms, or security protection, but lack a reliable execution mechanism located between the intelligent decision-making layer and the physical execution layer to uniformly constrain, continuously migrate, maintain identity, and prove execution of the target state. Therefore, as autonomous intelligent systems scale up and enter complex physical environments, a deterministic and reliable mechanism for state execution processes is needed.

[0017] In summary, the existing technological systems share common technological gaps in the following aspects:

[0018] Enforcement access authorization: There is a lack of a multi-layered constraint adjudication mechanism before the target state enters the execution process.

[0019] Continuous state transitions: lack of deterministic continuous constraints at the level of state transition paths.

[0020] Continuous identity maintenance: lack of persistent identity identification and causal association during state change processes.

[0021] Execution proof generation: Lack of verifiable execution proof chain generation capability.

[0022] State execution trustworthiness: Existing trusted computing technologies do not cover the execution trustworthiness of state change processes.

[0023] This invention provides a deterministic state continuous execution method, system, and storage medium for autonomous intelligent systems, aiming to fill the aforementioned technological gap. Summary of the Invention

[0024] (a) Technical problems to be solved

[0025] Existing autonomous intelligent systems can generate complex control strategies or target states, but they lack a technical mechanism that can deterministically constrain, continuously control, bind identities, and verify execution of state change processes before the intelligent decision results enter the physical execution system.

[0026] Specifically, the existing technology has the following technical defects:

[0027] First, there is a lack of an execution access authorization mechanism. The target state generated by the autonomous intelligent system is usually sent directly to the executor without setting multi-layered constraint adjudication during the execution access phase, making it impossible to judge the legality of the state before it enters physical execution.

[0028] Second, there is a lack of continuity constraints on state changes. Existing state switching or smooth transition methods only focus on the rate of change of state values ​​and do not establish deterministic continuous transition constraints at the level of state transition paths. In highly dynamic or safety-critical scenarios, this may lead to system instability or actuator overload.

[0029] Third, there is a lack of identification and traceability of state change processes. Existing systems only record changes in state values, without assigning continuous identification to the state change process, making it difficult to establish causal relationships between state changes.

[0030] Fourth, there is a lack of verifiable proof of state execution. Existing logging mechanisms rely on trust assumptions at the software layer and cannot provide a verifiable chain of execution proofs, making it difficult to meet the regulatory review and accountability requirements of safety-critical systems.

[0031] This invention provides a deterministic state-continuous execution mechanism for autonomous intelligent systems and their physical execution environments, aiming to solve at least one of the aforementioned technical problems.

[0032] (II) Technical Solution

[0033] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0034] A deterministic state continuous execution method for autonomous intelligent systems includes the following steps:

[0035] Step S1: State Entity Construction

[0036] Obtain the target state information generated by the autonomous intelligent system, and construct a state entity Ω based on the target state information, execution intention information, state identity information, and execution proof association information:

[0037]

[0038] Wherein, S represents the target state parameter information, used to represent the state value or configuration parameter to be executed; I represents the execution intent information, used to represent the source or task identifier that triggers the state change; V represents the state identity information, used to maintain the continuous identity association of the state entity during the state transition process; and T represents the execution proof association information, used to associate the execution verification process of the state entity.

[0039] Step S2: State Constraint Authorization

[0040] Obtain multi-layered constraint information, including at least the physical constraint space, the system operation constraint space, and the intelligent decision constraint space;

[0041] Based on the multi-layer constraint information, the target state is authorized and authorization state information is generated.

[0042] The physical constraint space has a higher priority than the system operation constraint space, and the system operation constraint space has a higher priority than the intelligent decision constraint space.

[0043] Step S3: Continuous State Transition

[0044] Based on the authorized status information, the target state is transformed into multiple consecutive migration states. ;

[0045] Among them, the change between adjacent migration states satisfies the preset maximum change constraint:

[0046]

[0047] M is a preset maximum state change threshold; the state is a single-dimensional state parameter or a multi-dimensional state vector.

[0048] Step S4: State and Identity Preservation

[0049] During state transition, the state identity information is continuously associated with the current transition state, so that the multiple consecutive transition states correspond to the same target state entity.

[0050] Step S5: Perform proof generation

[0051] Execution proof information is generated according to the state transition process. The execution proof information includes at least state information before state transition, authorization judgment information, transition state information, and preceding proof association information. The execution proof information is generated during the state transition process and is used to describe the authorization relationship, state change relationship, and preceding association relationship during the state transition process.

[0052] Subsequent state entities establish a connection between the preceding proof information and the current proof information, forming a non-skippable state execution proof chain;

[0053] Each piece of proof information contains a related reference to the preceding proof information. If any intermediate proof information is missing, the subsequent proof information cannot pass the related verification.

[0054] A deterministic state-based continuous execution system for autonomous intelligent systems includes:

[0055] The state entity construction module is used to obtain the target state information to be executed generated by the autonomous intelligent system, and construct a state entity based on the target state information, execution intention information, state identity information and execution proof association information;

[0056] The constraint authorization module is used to make authorization judgments on the target state based on the priority relationship between the multiple constraint spaces, including at least physical constraint space, system operation constraint space and intelligent decision constraint space, and generate authorization state information.

[0057] The state transition module is used to convert the target state into multiple consecutive transition states according to the authorized state information, wherein the amount of change between adjacent transition states satisfies a preset maximum change constraint.

[0058] The identity binding module is used to maintain the continuous association between the state identity identification information and the current migration state during the state transition process, so that the multiple consecutive migration states correspond to the same target state entity.

[0059] The proof generation module is used to generate execution proof information based on the state transition process, and to establish a connection between the previous proof information and the current proof information to form a non-skippable state execution proof chain.

[0060] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0061] (III) Beneficial Effects

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] First, improve the execution security of autonomous intelligent systems. By setting up multi-layered constraint and adjudication mechanisms during the execution access phase, unauthorized target states can be prevented from entering the physical execution chain, reducing execution risks caused by intelligent decision-making biases.

[0064] Second, ensure the continuity of state transitions. By decomposing the target state into multiple consecutive transition states that satisfy the maximum change constraint, system instability or actuator overload caused by state jumps can be avoided.

[0065] Third, establish the continuity of identity during state changes. By continuously associating state identities during the transition process, the previous and subsequent states can be identified as having the same execution intent, providing a causal basis for state changes.

[0066] Fourth, it provides verifiable execution proofs. By generating a non-skippable chain of state execution proofs, each state change can be traced back to its authorization basis, migration path, and execution result, providing a technical foundation for the regulatory review and liability determination of safety-critical systems.

[0067] Fifth, it fills a gap in existing trusted computing technologies. This invention complements existing trusted computing technologies, further addressing the issue of trusted state execution by building upon Secure Boot, TPM, and TEE, which respectively solve the problems of trusted startup, trusted platform identity, and trusted runtime environment. This provides a complete trusted execution chain for autonomous intelligent systems.

[0068] Sixth, it is applicable to a variety of autonomous intelligent execution scenarios. This invention can be applied to complex heterogeneous execution environments driven by autonomous intelligent systems, including computing infrastructure, robotic systems, aerospace systems, energy systems, and industrial control systems, providing a unified and reliable mechanism for state execution processes under different physical constraints. Attached Figure Description

[0069] Figure 1 This is a diagram illustrating the overall architecture of the deterministic state continuous execution method provided in this embodiment of the invention.

[0070] Figure 2 A schematic diagram of the structure of the state entity Ω=(S,I,V,T) provided in an embodiment of the present invention.

[0071] Figure 3 This is a schematic diagram of multi-layer constraint space relationships provided for an embodiment of the present invention.

[0072] Figure 4 This is a schematic diagram of continuous state transition relationships provided in an embodiment of the present invention.

[0073] Figure 5 This is a schematic diagram illustrating the continuous association of state and identity in an embodiment of the present invention.

[0074] Figure 6 This is a schematic diagram of the execution proof chain structure provided in an embodiment of the present invention.

[0075] Figure 7 A schematic diagram of the module structure of a deterministic state continuous execution system provided in an embodiment of the present invention.

[0076] Figure 8 This is a schematic diagram of the hardware execution control architecture provided in an embodiment of the present invention. Detailed Implementation

[0077] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to illustrate the present invention and should not be regarded as limiting the scope of protection of the present invention.

[0078] Example 1: Basic Flow of the Deterministic State Continuous Execution Method

[0079] like Figure 1As shown, the deterministic state continuous execution method provided in this embodiment of the invention includes the following steps:

[0080] Step S101: State Entity Construction

[0081] After generating the target status information, the autonomous intelligent system 101 obtains the target status information to be executed, and constructs the status entity Ω based on the target status information, execution intention information, status identity information and execution proof association information.

[0082] like Figure 2 As shown, state entity It comprises four components: S represents target state parameter information, including the state value or configuration parameters to be executed; I represents execution intent information, including the source or task identifier that triggers the state change; V represents state identity information, used to maintain the continuous identity association of state entities during state transitions; and T represents execution verification association information, used to associate the execution verification process of state entities. The state identity V is distinct from timestamp information or version number information generated based on time sequence, and it is bound to the state value, execution intent, and execution transaction.

[0083] Step S102: State Constraint Authorization

[0084] The system acquires multi-layered constraint information, including physical constraint space, system operation constraint space, and intelligent decision constraint space.

[0085] like Figure 3 As shown, the priority relationship among the constraint layers is as follows: the physical constraint space has a higher priority than the system operation constraint space, and the system operation constraint space has a higher priority than the intelligent decision constraint space. The system performs authorization judgment on the target state based on the above multi-layered constraint information, generating authorization state information.

[0086] Step S103: Continuous state transition

[0087] Based on the authorization status information, the system transforms the target state into multiple consecutive transition states. .

[0088] like Figure 4 As shown, the changes between adjacent migration states satisfy a preset maximum change constraint: , where M is a preset maximum state change threshold. In this embodiment, the state can be a one-dimensional state parameter or a multi-dimensional state vector.

[0089] Step S104: State and Identity Maintenance

[0090] During state transition, the system maintains a continuous association between the state identity V and the current transition state.

[0091] like Figure 5 As shown, from S0 to S n Throughout the entire migration process, the state identity V remains unchanged, enabling external systems to recognize that these successive states belong to the same execution intent.

[0092] Step S105: Perform proof generation

[0093] The system generates execution proof information based on the state transition process.

[0094] like Figure 6 As shown, the system generates corresponding proof information for each migration state. Each proof contains a reference to a preceding proof. Subsequent states are linked through these preceding proofs, forming a non-skippable chain of state execution proofs. The chain is broken if any intermediate proof node is missing.

[0095] Example 2: Application in heterogeneous computing systems

[0096] This embodiment uses the resource status management of a heterogeneous computing system as an example to describe the application of the present invention in computing resource status management.

[0097] When the intelligent scheduling module proposes to adjust the computing resource operating parameters from the first state value to the second state value, the system performs the following operations:

[0098] The system constructs a state entity Ω, which includes target state parameters S, execution intent information I, state identity identifier V, and execution proof association information T.

[0099] The system acquires multi-layered constraint information: the physical constraint space consists of the maximum allowed operating parameter values ​​and the maximum single change amount; the system operation constraint space determines whether the current operating state is within a safe range; and the intelligent decision constraint space consists of the suggested values ​​from the intelligent scheduling module. After authorization and judgment, the system decomposes the target state into multiple consecutive transition states, with the change amount in each transition step satisfying the maximum change constraint. Throughout the entire transition process, the state identity identifier V remains associated, and the system generates a corresponding execution proof chain.

[0100] Example 3: Application in Autonomous Space Systems

[0101] This example uses an autonomous spacecraft attitude adjustment mission. The autonomous spacecraft needs to adjust its attitude from the current angle to the target angle in order to establish a communication link.

[0102] The system constructs a state entity Ω, containing target attitude information, execution intent information, state identity identifier, and execution proof association information. The system acquires multi-layered constraint information: the physical constraint space includes maximum angular acceleration and maximum angular velocity along a single axis; the system operation constraint space includes the angle between the current solar vector and the attitude angle, and energy balance constraints; the intelligent decision constraint space includes the maneuver path suggested by the autonomous planning module. After authorization and judgment, the system decomposes the attitude adjustment process into multiple consecutive transition states, each intermediate state satisfying angular velocity change rate constraints and energy constraints. Throughout the entire attitude adjustment process, the state identity identifier remains associated, and the system generates a complete proof chain containing the authorization basis, transition path, and execution result for task attribution and post-event analysis.

[0103] Example 4: State execution control in large-scale heterogeneous intelligent infrastructure

[0104] This embodiment uses a large-scale heterogeneous intelligent computing system as a scenario to describe the application of the present invention in a complex execution system that includes multiple computing nodes, storage nodes, network nodes, and energy management units.

[0105] The system includes an autonomous intelligent scheduling module that can generate resource adjustment target states based on task requirements. These target states include at least one of the following: computing resource allocation state, power consumption state, network bandwidth state, or storage access state. Due to power consumption limitations, temperature limitations, communication limitations, and hardware reliability limitations inherent in large-scale intelligent infrastructure, the target states generated by the autonomous intelligent scheduling module cannot be directly executed.

[0106] The system performs the following operations: acquires target state information and constructs a state entity Ω; acquires the physical constraint space (total system power consumption limit, single node power consumption limit, temperature safety range), system operation constraint space (current load distribution, resource availability, network congestion status), and intelligent decision constraint space (target value output by the autonomous intelligent scheduling module), and makes authorization judgments based on priority relationships; decomposes the target state into multiple consecutive transition states, ensuring that the changes between adjacent transition states meet the preset maximum change constraint; maintains the continuous association of state identity identifiers during state transitions; and generates execution proof information during execution, forming an unskippable execution proof chain.

[0107] Through the above steps, the system can safely transform the target state generated by the autonomous intelligent scheduling module into actual execution actions while meeting power consumption, temperature, communication, and hardware reliability limitations.

[0108] Example 5: Hardware Execution Architecture

[0109] like Figure 8 As shown, the present invention includes the following functional units at the hardware level:

[0110] The status receiving interface 104A is used to receive the status information of the target to be executed generated by the autonomous intelligent system.

[0111] The constraint arbitration unit 104B is used to make authorization judgments on the target state based on multi-layered constraint information, including at least physical constraint space, system operation constraint space, and intelligent decision constraint space. The authorization decisions are executed in the order that physical constraint space has higher priority than system operation constraint space, and system operation constraint space has higher priority than intelligent decision constraint space. When the target state does not meet the physical constraint space requirements, the constraint arbitration unit directly prohibits the execution of the target state.

[0112] The state transition control unit 104C is connected to the constraint arbitration unit and is used to receive authorized state information, convert the target state into multiple consecutive transition states, and the amount of change between adjacent transition states satisfies the preset maximum change constraint.

[0113] The identity binding unit 104D is connected to the state transition control unit and is used to maintain the continuous association between the state identity information and the current transition state during the state transition process, so that multiple consecutive transition states correspond to the same target state entity.

[0114] The proof generation unit 104E is connected to the state transition control unit and the identity binding unit. It is used to generate execution proof information according to the state transition process and establish a connection between the previous proof information and the current proof information to form a non-skippable state execution proof chain.

[0115] The execution output interface 104F is used to control the target state's entry into the physical execution link based on the authorization status information. The target state is only allowed to enter the physical execution link after it has passed constraint authorization, continuous migration, and identity preservation. The above units can be implemented through hardware logic circuits, programmable logic devices, application-specific integrated circuits, processor execution logic, or a hardware-software co-operation method.

[0116] Through the aforementioned hardware execution architecture, the state execution process is executed independently by the hardware execution control layer, without relying on the output of the autonomous intelligent system or subsequent processing by the upper-level software modules, thereby forming an independent, hardware-isolated, and trustworthy execution boundary between the intelligent decision-making layer and the physical execution layer.

[0117] Industrial applicability

[0118] This invention can be widely applied in the following fields:

[0119] I. AI Computing Infrastructure: This includes AI training clusters, inference clusters, heterogeneous computing systems, CXL memory expansion systems, Chiplet architecture systems, etc., which are used to deterministically control changes in resource states such as GPU frequency, HBM bandwidth, power consumption status, and computing topology.

[0120] II. Autonomous Systems: These include autonomous driving systems, robot control systems, unmanned aerial vehicle autonomous control systems, spacecraft autonomous control systems, etc., which are used to continuously transfer and reliably execute motion states, decision states, and mission states.

[0121] III. Industrial Control Systems: These include intelligent manufacturing systems, energy management systems, and process control systems, which are used to perform deterministic execution control of equipment operating status, production parameters, and energy distribution status.

[0122] IV. Safety-critical systems: These include medical auxiliary equipment control systems, rail transit signaling systems, energy control systems, etc., used for authorized verification and execution proof of state change processes.

[0123] V. Emerging Intelligent Systems: These include AI agent execution systems, digital twin control systems, and autonomous experimental systems, which are used to establish a trustworthy execution boundary between intelligent decision-making and physical execution.

[0124] This invention is not dependent on any specific application scenario, but rather addresses the general reliable execution requirements of target states generated by autonomous intelligent systems entering the physical execution link. The technical solution of this invention can function independently of upper-layer intelligent algorithms and lower-layer physical devices, serving as a deterministic execution control mechanism in the middle layer, and is compatible with various autonomous intelligent systems and physical execution systems.

[0125] Explanation of reference numerals in the attached figures

[0126] 101 Autonomous Intelligent System 102 State execution trust layer 103 Physical execution system 104 Deterministic state continuous execution control module 104A Status receiving interface 104B Constraint Arbitration Unit 104C State transition control unit 104D Identity binding unit 104E Proof generation unit 104F Execute output interface Ω State Entity S Target state parameter information I Execution Intent Information V Status identification information T Execution certificate related information IS Physically constrained space OS System operating constraint space MS Intelligent decision constraint space M Maximum state change threshold

Claims

1. A deterministic state continuous execution method for autonomous intelligent systems, characterized in that, include: Obtain the target status information to be executed generated by the autonomous intelligent system, and construct a status entity based on the target status information, execution intention information, status identity information and execution proof association information; Based on multi-layered constraint information, including at least physical constraint space, system operation constraint space, and intelligent decision constraint space, authorization judgment is made on the target state based on the priority relationship between the multi-layered constraint spaces, and authorization state information is generated. Based on the authorized status information, the target state is converted into multiple consecutive migration states, wherein the amount of change between adjacent migration states satisfies a preset maximum change constraint. During state transition, the state identity information is continuously associated with the current transition state, so that the multiple consecutive transition states correspond to the same target state entity; Execution proof information is generated based on the state transition process, and a non-skippable state execution proof chain is formed by establishing a connection between the preceding proof information and the current proof information.

2. The method according to claim 1, characterized in that, The state entity includes: Target status parameter information (S) is used to represent the status value or configuration parameter to be executed; Execution intent information (I) is used to indicate the source or task identifier that triggered the state change; State identity information (V) is used to maintain the continuous identity association of state entities during state transitions; The execution verification information (T) is used to verify the execution of the associated state entity.

3. The method according to claim 1, characterized in that: The physical constraint space has a higher priority than the system operation constraint space; The system operation constraint space has a higher priority than the intelligent decision constraint space; When the target state does not satisfy the physical constraint space, the execution of the target state is prohibited.

4. The method according to claim 1, characterized in that, The preset maximum change constraint can be expressed as: in, To preset the maximum state change threshold, This is the current migration status. The next transition state is defined as either a one-dimensional state parameter or a multi-dimensional state vector.

5. The method according to claim 1, characterized in that, The state identity information is associated with at least one of the state parameter information, execution intent information, and execution proof association information, and maintains a unique association relationship among multiple consecutive migration states.

6. The method according to claim 1, characterized in that, The execution proof information includes at least state information before state transition, authorization judgment information, transition state information, and prior proof association information. The execution proof information is generated during the state transition process and is used to describe the authorization relationship, state change relationship, and prior association relationship during the state transition process. In the state execution proof chain: Each proof information contains related references to preceding proof information; If any intermediate proof information is missing, subsequent proof information cannot pass the associated verification.

7. The method according to claim 1, characterized in that, The physical constraint space includes at least one of the following constraints: System physical limits; Actuator physical limitations; Safety boundary constraints.

8. The method according to claim 1, characterized in that, The system operation constraint space includes at least one of the following constraints: Current system operating status; Resource availability constraints; Load condition constraints.

9. The method according to claim 1, characterized in that, The intelligent decision constraint space includes at least one of the following constraints: The target state output by the autonomous intelligent system; The suggested values ​​output by the autonomous intelligent system; The optimization target output by the autonomous intelligent system.

10. A deterministic state continuous execution system for autonomous intelligent systems, characterized in that, include: The state entity construction module is used to obtain the target state information to be executed generated by the autonomous intelligent system, and construct a state entity based on the target state information, execution intention information, state identity information and execution proof association information; The constraint authorization module is used to make authorization judgments on the target state based on the priority relationship between the multiple constraint spaces, including at least physical constraint space, system operation constraint space and intelligent decision constraint space, and generate authorization state information. The state transition module is used to convert the target state into multiple consecutive transition states according to the authorized state information, wherein the amount of change between adjacent transition states satisfies a preset maximum change constraint. The identity binding module is used to maintain the continuous association between the state identity information and the current migration state during the state transition process, so that the multiple consecutive migration states correspond to the same target state entity. The proof generation module is used to generate execution proof information based on the state transition process, and to establish a connection between the previous proof information and the current proof information to form a non-skippable state execution proof chain.

11. The system according to claim 10, characterized in that, Also includes: The hardware execution control module is used to control the target state to enter the physical execution link based on the authorized status information.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 9.

Images