Autonomous awareness driven semantic planning and behavior generation system

Abstract

The invention discloses a semantic planning and behavior generation system driven by autonomous consciousness. According to the system, a DIKWP mesh cognitive model is adopted, an intention / purpose semantic layer is embedded into an AI internal architecture, and autonomous decision making is achieved through five-layer semantic representation of data, information, knowledge, wisdom and intention. The system comprises a perception module, an intention generation module, a semantic planning module, a meta-cognition monitoring module, an execution module and the like. The AI can set a task target by itself and dynamically plan an action scheme based on a semantic state. Under the action of a double-circulation element cognitive mechanism, the AI can monitor the deviation between a behavior and a target in real time and automatically correct a plan. Compared with a traditional black box model, each step of decision of the AI has a traceable semantic basis and interpretation which can be understood by human beings, and the interpretability and controllability of the system are improved. The autonomous planning system can automatically adjust strategies according to environmental changes, ensures that AI behaviors always accord with expected targets and ethical safety constraints, is widely applicable to the fields of service robots, automatic driving, intelligent assistants and the like, and has important application values and commercial prospects.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and cognitive computing, and more particularly to a semantic planning and behavior generation system driven by autonomous consciousness. This system, based on an artificial consciousness model, enables AI to autonomously set goals and perform task planning, self-monitoring, and dynamic correction based on semantic states, belonging to the fields of intelligent control and cognitive robotics. Background Technology

[0002] Currently, large-scale language models and intelligent agents primarily rely on preset goals or human instructions to drive their behavior, with limited adjustments based on environmental feedback, making true autonomous planning difficult. Traditional AI planning is often a data-driven, black-box optimization process; AI lacks a semantic understanding of its own goals and cannot adjust its plans according to environmental changes. This limitation leads to insufficient adaptability and safety in complex and dynamic scenarios—for example, when unforeseen circumstances arise during task execution, AI often lacks the ability to proactively correct its course.

[0003] The DIKWP network cognitive model adds an "intent" layer to the traditional DIKW (pyramid) architecture, enabling bidirectional semantic feedback between cognitive layers through a network structure. This innovation provides AI with an embedded goal representation and scrutiny mechanism, and is considered an important path towards autonomous consciousness. Simultaneously, its "dual-loop" consciousness architecture introduces metacognitive loops, giving AI self-monitoring and self-regulation capabilities, thus potentially solving the black-box uncontrollability problem of current large-scale models.

[0004] International research on Artificial General Intelligence (AGI) has largely focused on scaling up neural networks and data-driven optimization, lacking explicit modeling and semantic interpretation mechanisms for machine self-purpose. Even in leading approaches like OpenAI, the model decision-making process remains highly opaque to humans. In contrast, this invention, based on the DIKWP theory, introduces a self-awareness-driven planning engine, enabling AI to self-set and adjust goals in a white-box, interpretable manner, ensuring it consistently serves human values ​​and safety needs. This differentiated approach possesses significant technological originality and patent value, holding strategic importance for the future evolution of interpretable, safe, and controllable AGI. Summary of the Invention

[0005] This invention aims to address the lack of autonomous goal setting and dynamic planning adjustment mechanisms in existing artificial intelligence systems, thereby improving the adaptability and safety of AI in complex environments. Traditional AI often strictly follows preset instructions when performing tasks, lacking the ability to adjust based on its own objectives. When the environment changes or task goals need to be refined, existing systems cannot semantically understand these changes and autonomously replan, easily leading to task failure or behavioral deviations. This invention provides a planning and behavior generation architecture with artificial autonomous awareness characteristics, enabling AI to autonomously adjust its goals and plans based on internal semantic states and external feedback, thereby improving robustness and value alignment.

[0006] This invention proposes an autonomous planning system architecture based on the DIKWP cognitive model, comprising five semantic processing units (data, information, knowledge, wisdom, and intent) and a dual-loop feedback mechanism. The system embeds an "intent / purpose" generation module, representing the AI's high-level goals semantically and interacting with other cognitive layers. During execution, the metacognitive monitoring module continuously compares the current behavior with the intent layer's goal; if a deviation is detected, planning adjustments are triggered (the consciousness bug theory posits that consciousness originates from "imperfections" in the cognitive process; this invention utilizes this "deviation" signal to stimulate the AI's self-correction function). Specifically:

[0007] The perception layer converts environmental data into semantic state representations (data → information → knowledge extraction).

[0008] The intelligence layer formulates preliminary action plans based on knowledge reasoning.

[0009] The intent layer combines the system's built-in objectives (which can be derived from initial tasks or autonomously) to evaluate the consistency between the proposed plan and long-term goals.

[0010] The metacognitive layer (second loop) monitors the discrepancy between the output and intent of the intelligence layer. When it detects a conflict or anomaly (cognitive "BUG"), it triggers adjustments, including resetting sub-goals, modifying action sequences, or updating intent weights to realign them.

[0011] The adjusted plan is then refined and executed by the intelligent layer, and this cycle continues until the task is completed or the ideal semantic state is achieved.

[0012] This approach elevates traditional linear programming to iterative semantic programming: AI no longer passively executes predetermined instructions but possesses self-driven and reflective abilities similar to humans, enabling it to make semantic interpretations and adaptive adjustments in response to environmental changes. Through this design, each AI decision has a clear semantic basis, and its underlying purpose can be traced.

[0013] The autonomous semantic planning system of the present invention has the following outstanding advantages:

[0014] Enhanced autonomy and adaptability: AI can autonomously set and refine goals, no longer relying entirely on external instructions. It can adjust plans in a timely manner in the face of dynamic environments, avoid rigid execution, and improve task success rate and flexibility.

[0015] Semantic interpretability: By introducing the semantic hierarchy framework of DIKWP, each decision step in the system is accompanied by a semantic label (data → information → knowledge → wisdom → intent), which can generate human-understandable explanations of the decision chain. This improves the transparency of AI behavior and facilitates auditing by developers and regulatory agencies.

[0016] Value Alignment and Security: The intent layer ensures that the AI's autonomous goals always remain within safe boundaries and serve the core values ​​assigned by humans. When a plan is detected to deviate from these values ​​or original intentions (similar to a consciousness bug triggering), the system can correct course immediately, reducing the risk of losing control.

[0017] Innovation and Versatility: This architecture integrates artificial intelligence theory and planning algorithms, representing a pioneering AGI cognitive planning mechanism that fills a gap in existing technologies. Its concepts can be widely applied to autonomous robots, intelligent driving, and intelligent assistants, enabling higher levels of AI autonomy and significantly driving industrial development.

[0018] For clarity, the technical solution of the present invention will be further described in conjunction with the accompanying drawings:

[0019] Appendix Figure 1 This diagram illustrates the overall architecture of the autonomous semantic planning and behavior generation system of the present invention. The system mainly consists of the following components:

[0020] Semantic State Representation Module (10): Receives raw data from the external environment, maps it step-by-step to a conceptual space representation through multimodal perception and a knowledge base, and further refines it into a semantic space state that can be understood by AI. This module corresponds to the data, information, and knowledge layers of the DIKWP model, and outputs a semantic description of the current environment.

[0021] Intent Generation Module (20): Embedded with an autonomous consciousness engine, used to generate and maintain the AI's internal goal representation (intent layer). This module can set the top-level goal according to the initial task requirements and can autonomously derive sub-goals as needed during execution. Intents are stored in symbolic semantic form, representing the state that the AI ​​"wants to achieve".

[0022] Semantic Planning Module (30): Corresponding to the intelligence layer of DIKWP, it calls the reasoning algorithm to formulate an action plan to achieve the goal based on the current semantic state (from module 10) and the intention goal (from module 20). The planning result includes a series of executable action steps, each with a semantic reason for performing the action.

[0023] Metacognitive Monitoring Module (40): Serves as the monitoring link in the "dual-loop" architecture. This module monitors the execution of the plan in real time and compares the actual behavioral results with the intended goals. If the behavior deviates from the goal or produces anomalies (such as the plan failing to achieve the expected semantic effect), it is identified as an "intent deviation" signal, equivalent to the cognitive inconsistency trigger point in the consciousness BUG theory.

[0024] Behavior execution and feedback module (50): Executes the specific actions planned by module 30 (which may include mechanical control, dialogue generation, etc.) and transforms the execution feedback into new environmental data input module 10 to form a closed loop.

[0025] Appendix Figure 2 A timing diagram of the semantic planning double-loop adjustment process in this invention is provided. The specific steps are as follows:

[0026] Initial goal setting (S101): Module 20 generates initial intent goal semantics based on the input task or system autonomous decision.

[0027] State awareness (S102): Module 10 senses the environment and outputs a semantic state representation.

[0028] Task planning (S103): Module 30 infers a preliminary action plan based on semantic state and intentional goal.

[0029] Plan Execution (S104): Module 50 executes one action in the plan and obtains environmental feedback.

[0030] Meta-monitoring and Evaluation (S105): Module 40 compares the degree of conformity between the execution results and the intended goals. If the deviation is within the acceptable range and the plan is still effective, it returns to step S104 to execute the next action; if a significant deviation or conflict occurs, it proceeds to step S106.

[0031] Intent / Plan Correction (S106): In response to detected issues, module 40 triggers adaptive adjustments: it may modify remaining plan steps, update the current sub-goal, or call module 20 to regenerate a new intent. The adjustment is based on ensuring that subsequent actions are aligned with the higher-level goal.

[0032] Loop Iteration (S107): Return to step S102 to perceive the latest environmental state and continue subsequent tasks until the intended goal is achieved or the task is terminated.

[0033] Through the above-described architecture and process, this invention achieves the integration and interaction of the conceptual space and semantic space within AI: AI transforms low-level perceptual data into high-level semantic understanding, and autonomously makes decisions and corrects its behavior at the semantic level, thereby forming a complete cognitive loop. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the architecture of the "Autonomous Consciousness-Driven Semantic Planning and Behavior Generation System" of the present invention.

[0035] Figure 2 This is a flowchart of the semantic planning and self-adjustment process of the system of this invention. Detailed Implementation

[0036] The working process of this invention will be described below with reference to a specific embodiment. Assume this system is deployed on a home service robot whose overall task is to keep the room clean and ensure the safety of family members. Initially, a vague goal is given to the robot by a human: "Keep the living room clean and comfortable." Based on this, the robot will generate its internal top-level intent:

[0037] Example: Autonomous Planning of Home Service Robots

[0038] Intent setting: The robot transforms “keeping the living room clean and comfortable” into an internally expressed semantic goal through the intent generation module (20), including sub-goals such as “cleaning the floor”, “tidying up clutter”, and “ensuring that family members are not disturbed excessively”. In particular, it combines the preset ethical rules and takes “not disturbing family members’ rest excessively” as one of the constraints (belonging to the semantic boundary of value constraints, supported by the mechanism of Invention 2).

[0039] Environmental perception: The semantic state module (10) collects current living room environmental data (camera, sensors, etc.). It identifies: there is trash and debris on the floor (needs to be cleaned), and family members are resting at night (requires a quiet environment).

[0040] Semantic Planning: Module 30 integrates the above information and intent, and plans the following sequence: [Action 1: Clean the floor with the sweeping module in silent mode -> Action 2: Sort and put away clutter -> Action 3: Adjust the lighting to create a comfortable atmosphere]. Each step is accompanied by a reason explanation (e.g., reason for Action 1: "A dirty and messy floor affects comfort and no one is active, so it can be cleaned immediately"; reason for Action 3: "Adjusting the lighting at night can improve comfort without disturbing rest").

[0041] Execution and Monitoring: Module 50 executes Action 1 (Silent Cleaning), while Metacognitive Module 40 monitors the effect. Since silent mode was used and family members were not disturbed, the result meets the "no excessive disturbance" constraint, and the deviation signal is zero. Then, Action 2 continues.

[0042] Unexpected Situations and Self-Adjustment: During Action 2, the robot discovered an important item whose classification was uncertain (e.g., a document with no clear storage area). Module 40 detected a cognitive uncertainty bug: the current plan did not cover how to handle the item, and following the default might result in misplacement, violating the "tidy" intention. Therefore, an adjustment was triggered: Module 20 derived a new sub-objective, "Identify and properly place the unknown item," based on the rules, and Module 30 inserted a new action, "Ask the owner how to handle the item" or "Place the item in the temporary storage area and remind the owner." The adjusted plan realigned with the overall objective.

[0043] Continuing execution: The robot performs and completes the adjusted action 2, followed by action 3 to adjust the lights. The metacognitive module assesses that each step meets the comfort and non-interference constraints, ultimately achieving the intention of "a clean and comfortable living room," and the task ends. The robot records this autonomous adjustment process for future reference and learning in similar scenarios.

[0044] In this embodiment, the robot fully demonstrates the advantages of the system of the present invention: it can autonomously understand the essential semantic goal of "cleanliness and comfort" and decompose it into a sequence of executable actions; when encountering unplanned situations during execution, it does not mechanically ignore or handle errors, but rather "perceives" the problem through internal awareness (metacognitive monitoring) and actively seeks / adjusts solutions, so that the behavioral outcome conforms to the initial purpose and constraints. This semantically driven adaptive planning is impossible to achieve in existing traditional robot systems, reflecting the innovation and practical value of the present invention.

[0045] Extended Applications: Beyond service robots, this invention can be widely applied to fields such as autonomous driving (vehicles adjust their driving strategies based on traffic semantics, such as changing routes in emergencies to balance safety and efficiency) and intelligent customer service systems (AI adjusts its response style and objectives based on the semantic emotion of the conversation to better solve user problems). In complex and variable environments such as military and medical missions, introducing an autonomous semantic planning mechanism can significantly improve the reliability of AI actions and the degree to which they conform to human commands.

Claims

1. An autonomous consciousness-driven semantic planning and behavior generation system, comprising a perception module, an intention generation module, a semantic planning module, a meta-cognition monitoring module, and an execution module, characterized in that: - the perception module converts external environment data into semantic state representations step by step; - the intention generation module generates AI's internal semantic goal representations based on initial task requirements and preset rules, and can dynamically update intentions during execution; - the semantic planning module inferences to generate a behavior plan containing multiple specific actions according to the semantic state and internal semantic goal, with each action attached to a corresponding semantic reason; - the meta-cognition monitoring module monitors the consistency of the behavior results of the execution module with the internal semantic goal in real time, and triggers automatic adjustment of the behavior plan or internal semantic goal if deviations or abnormalities are detected; - the execution module executes actions according to the behavior plan and feeds back execution feedback to the perception module to form a closed loop until the internal semantic goal is reached or the task is completed.

2. The system of claim 1, wherein when the meta-cognition monitoring module detects a deviation, the intention generation module regenerates a corrected internal semantic goal according to the deviation type, and the semantic planning module updates the remaining behavior plan accordingly to ensure that the adjusted action is semantically consistent with the initial goal.

3. The system of claim 1, wherein the intention generation module includes an ethics and safety constraint unit for integrating predefined moral norms, laws and regulations, or user preferences as constraint conditions when generating or adjusting internal semantic goals, ensuring that the behavior plan is semantically consistent with human values and safety requirements.

4. The system of claim 1, wherein the semantic planning module uses a hierarchical knowledge graph and reasoning algorithm to generate plans with DIKWP model data, information, knowledge, and wisdom as intermediate representations, ensuring that each step of action has an explainable semantic basis.

5. A method of autonomous semantic planning and behavior generation for controlling an artificial intelligence agent to autonomously plan and execute tasks according to internal semantic goals, comprising the following steps: (a) obtaining environment data and converting it into semantic state representations; (b) setting the internal semantic goal intention of the artificial intelligence agent; (c) generating a behavior plan based on the semantic state and internal intention; (d) executing the action steps of the plan and monitoring the execution results; (e) comparing the execution results with the internal intention, and automatically adjusting the internal intention and behavior plan if there is a significant deviation; (f) repeating steps (a)-(e) until the internal semantic goal is met or the task is completed.

6. The method of claim 5, wherein the deviation in step (e) includes failure to achieve the expected semantic effect of the task, violation of preset constraint rules, or unexpected changes in the environment, and when a deviation occurs, the internal state of the artificial intelligence agent is adjusted through a meta-cognition feedback loop to promote subsequent behavior to be consistent with the internal semantic goal.

Images