A somatic intelligent cognitive decision-making method and system based on internal physical inertia retention

Abstract

The application discloses a kind of embodied intelligent cognitive decision-making method and system based on internal physical inertia retention, belong to artificial intelligence and cognitive computing technology field.The method includes: obtaining real-time feedback of executor by sensor, current physical motion state is expressed as physical state vector;According to current linear velocity, the corresponding relationship of preset speed interval and cognitive situation type is inquired, and internal inertia situation identification is generated;The internal inertia situation identification is used as the initial condition of cognitive decision-making, and initial energy vector is generated;With the situation type corresponding to the internal inertia situation identification as the path starting point, the situation migration path is planned by situation migration path planning engine;On the basis of the initial energy vector, superimpose external event disturbance energy, drive cognitive energy vector evolution to stable situation;According to situation type, generate execution instruction;After execution, actual physical state is backfilled to cognitive layer, and complete cognitive closed loop is formed.The application enables cognitive system to continuously perceive and keep the physical motion inertia of itself, and when external event arrives, situation migration path planning is carried out with physical inertia as starting point, realizes "perception→mapping→backfilling→initialization→path planning→game→execution→feedback" embodied cognitive closed loop, and can be widely applied to robot, automatic driving and other embodied intelligent scenes.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and cognitive computing technology, specifically relating to an embodied intelligence cognitive decision-making method and system based on the maintenance of internal physical inertia. It can be used in embodied intelligence application scenarios such as robots, autonomous driving, and intelligent agent control, which require deep integration of physical state and high-level cognitive decision-making. Background Technology

[0002] The core characteristic of embodied AI is that the intelligent agent not only possesses cognitive decision-making capabilities but also has a physical body and interacts with the physical world in real time. In recent years, with the development of technologies such as humanoid robots, quadruped robots, and autonomous driving, how to organically integrate the physical motion state of robots with high-level cognitive decision-making has become a key technical issue in the field of embodied AI.

[0003] In existing technologies, there is a significant "cognitive-physical gap" between cognitive decision-making systems and physical execution systems. Specifically, this manifests as follows: (i) The cognitive decision-making level lacks continuous perception of its own physical motion state. Existing cognitive decision-making systems (such as deep learning-based decision models and rule-based state machines) typically rely solely on external environmental perception information (e.g., images, point clouds, and speech) when making decisions, neglecting the agent's own physical inertia. For example, the current physical state of a robot moving at high speed—its momentum, velocity, and acceleration—should constrain its cognitive judgment of "what to do next." However, in current technologies, there is a lack of an effective two-way information channel between the cognitive layer's decision-making process and the execution layer's physical state. The cognitive layer "doesn't know how fast it's moving," potentially leading to decision results that are detached from physical reality.

[0004] While existing technologies include research on "internal disturbances," focusing on improving a robot's adaptability to internal disturbances (such as inertial coupling effects during high-speed multi-joint motion, motor torque fluctuations, and center of gravity shifts caused by load changes), these technologies primarily focus on the stability of the underlying motion control, falling within the scope of execution layer optimization. They do not abstract the physical state and feed it back to the cognitive decision-making layer, nor do they address the impact of the physical state on higher-level situational judgments.

[0005] Existing technologies include robot motion state cognition methods based on rat hippocampal cognitive mechanisms, as well as perceptual speed solution methods based on velocity cells and visual information. However, these methods mainly serve the cognition of motion state and position (i.e., "where am I" and "how fast am I moving"), and do not map physical state to high-level situation types to participate in cognitive decision-making, nor do they involve the inertial maintenance mechanism of physical state for cognitive situation judgment.

[0006] (ii) Lack of a closed-loop feedback mechanism between cognitive decision-making and physical execution In existing embodied intelligence systems, the decision-making results of the cognitive layer flow unidirectionally to the execution layer. After the execution layer completes its execution, its results (actual speed, actual acceleration, actual position, etc.) are not fed back to the cognitive layer as input conditions for the next cognitive decision. This results in the cognitive layer making decisions as if facing the current situation for the "first time," lacking continuous perception of its own execution history, and failing to form a complete cognitive loop of "perception → judgment → action → perception of its own action results → re-judgment."

[0007] Although some existing technologies involve "physical state deduction" or "physical space feedback", their feedback mechanisms mainly serve the reward calculation of motion planning or reinforcement learning, rather than as a basis for situation judgment in cognitive decision-making, and do not involve technical solutions for mapping physical states to high-level cognitive situations.

[0008] (iii) Lack of effective methods for mapping continuous physical states to discrete cognitive situations Cognitive decision-making systems typically operate in a discrete situation space (such as "forward state," "stopped state," "obstacle avoidance state," etc.), while physical execution systems generate continuous motion parameters (velocity values, acceleration values, etc.). Existing technologies lack a method to effectively map continuous physical state quantities to discrete cognitive situations, making it difficult for the physical state to be directly understood and utilized by the cognitive layer.

[0009] (iv) External events can easily completely overshadow the internal physical state, causing decision-making to lose physical continuity. In existing event-driven cognitive systems, whenever a new external event (such as "red light detected" or "obstacle detected") occurs, the system's cognitive state is often completely reset or overwritten, ignoring the system's current physical inertia. For example, if a vehicle traveling at high speed detects a red light, and the cognitive system completely ignores its current speed and jumps directly to the "stop" decision, it cannot plan a smooth transition path of "decelerating first and then stopping," resulting in a decision that is disconnected from the actual physical motion.

[0010] (v) The existing situational migration path planning lacks the input of physical inertia. Prior to the filing date, the applicant's earlier patent applications (such as CN202610506445.X, CN202610601934.3, etc.) have disclosed a state representation method based on a six-dimensional energy vector, a complete situation space, an energy evolution mechanism, and a situation transition path planning method. However, in the aforementioned earlier applications, the situation transition path planning is based only on internal cognitive layer parameters such as event type, uncertainty value, and urgency, and has not yet taken the agent's current physical speed as an input condition for path planning. This leads to the planned path potentially deviating from physical reality—for example, planning a path directly from "high-speed driving" to "stopping," lacking the necessary deceleration intermediate situation.

[0011] In summary, there is an urgent need for a method and system that can continuously maintain the internal physical inertia of a robot and integrate it into the cognitive decision-making process, so that the cognitive layer "knows what it is doing and how fast it is doing it," and uses physical inertia as an important input when planning situational transition paths, and can plan a smooth transition path that conforms to physical reality when external events occur. Summary of the Invention

[0012] (a) Purpose of the invention The purpose of this invention is to provide an embodied intelligent cognitive decision-making method and system based on the maintenance of internal physical inertia, in order to solve the technical problems in the prior art, such as the lack of continuous perception of its own physical motion state by the cognitive decision-making layer, the lack of a closed-loop feedback mechanism between cognitive decision-making and physical execution, the lack of an effective method to map continuous physical states into discrete cognitive states, and the fact that external events can easily completely cover the internal physical state, causing the decision to lose physical continuity.

[0013] Specific objectives include: (1) A method is provided to map the continuous physical state (velocity, acceleration, etc.) of the robot's execution layer to a discrete cognitive state, so that the cognitive layer can "understand" its own physical motion state.

[0014] (2) Provides a closed-loop mechanism for physical state feedback, so that the actual execution results of the execution layer can be continuously fed back to the cognitive layer as the initial condition for the next cognitive decision, forming a complete cognitive closed loop of "perception → judgment → action → perception of one's own action results → re-judgment".

[0015] (3) Provides a cognitive decision-making method for maintaining internal physical inertia, so that when the system receives external events, it can plan a situational migration path that conforms to physical reality through a situational migration path planning engine while maintaining the current physical inertia, rather than being completely covered by external events, thereby achieving physical continuity and smoothness of decision-making.

[0016] (4) Provide a speed-situation mapping method to map different speed ranges to different cognitive situation types, so that the cognitive layer can perceive its own "motion situation" based on the current speed.

[0017] (5) When planning the situational migration path planning engine, the current physical velocity-corresponding situation type can be used as the starting point of the path, so as to ensure that the planned path starts from physical reality and conforms to the laws of kinematics.

[0018] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: 1. System Architecture An embodied intelligent cognitive decision-making system based on internal physical inertia retention includes a processor, a memory, and sensor interfaces and actuator interfaces connected to the processor. The processor is configured to implement the following modules by executing program instructions stored in the memory: (1) Physical state sensing module The physical state perception module is used to acquire real-time motion feedback data of the actuator through the sensor interface and represent the current physical motion state of the agent as a physical state vector.

[0019] The physical state vector includes at least one of the following parameters or a combination thereof: current linear velocity, current angular velocity, current acceleration, current position coordinates, and current orientation angle.

[0020] The physical state sensing module continuously acquires the physical state vector at a preset sampling frequency (preferably 50Hz to 200Hz) and stores the physical state vector in the cache area of ​​the memory for other modules to call.

[0021] (2) Velocity-Situation Mapping Module The velocity-situation mapping module is used to map the velocity parameters in the physical state vector to the corresponding cognitive situation type, and generate an internal inertial situation identifier.

[0022] The speed-situation mapping module has multiple preset correspondences between speed ranges and cognitive situation types. When the current linear velocity falls into the first speed range, it is mapped to the first cognitive situation type; when the current linear velocity falls into the second speed range, it is mapped to the second cognitive situation type; and so on.

[0023] The cognitive situation types are derived from a preset complete situation space. This complete situation space contains a finite and fixed number of basic situation types, preferably sixty-four, corresponding to all combinations of the six-dimensional binary coding space (2). 6 =64), each basic situation type corresponds to a unique six-dimensional binary encoding vector.

[0024] In a preferred embodiment, the correspondence between the speed range and the cognitive situation type is as follows: - When the current linear velocity is higher than the first velocity threshold (preferably 0.8 m / s), it is mapped to the first cognitive state type, preferably the full-speed expansion state, and the six-dimensional binary encoding vector is [1,1,1,1,1,1], which represents the "full-dimensional hyperactivity, maximum expansion" state; - When the current linear velocity is between the second velocity threshold and the first velocity threshold (preferably 0.4 m / s to 0.8 m / s), it is mapped to the second cognitive state type, preferably the high-speed movement state, and the six-dimensional binary encoding vector is [1,1,1,1,0,0], which represents the medium-to-high speed movement state of "rigid movement and strength"; - When the current linear velocity is between the third velocity threshold and the second velocity threshold (preferably 0.1 m / s to 0.4 m / s), it is mapped to the third cognitive state type, preferably the low-speed cautious state, and the six-dimensional binary encoding vector is [1,1,0,1,1,1], which represents the low-speed cautious movement state of "distinguishing up and down and acting cautiously"; - When the current linear velocity is lower than the third velocity threshold (preferably 0.1 m / s), it is mapped to the fourth cognitive situation type, preferably the static locked state, with a six-dimensional binary encoding vector of [0,0,1,0,0,1], representing the stopping state of "stopping at the point of stopping and pausing observation".

[0025] The internal inertial situational awareness generated by the velocity-situation mapping module is transmitted to the cognitive decision-making module as one of the initial conditions for cognitive decision-making.

[0026] (3) Physical state recharge module The physical state feedback module is used to transmit the latest physical state vector obtained by the physical state perception module to the cognitive decision module after each cognitive decision cycle is completed, thereby updating the internal inertial state of the cognitive decision module.

[0027] After each actuator completes its action, the physical state feedback module obtains the actual physical state vector (including the actual velocity, actual acceleration, etc.) and transmits the actual physical state vector and the internal inertial situation label generated by the velocity-situation mapping module to the cognitive decision module.

[0028] The cognitive decision-making module uses the received internal inertial state identifier as the initial energy bias for the next cognitive decision, so that the cognitive state of the system naturally continues along the physical inertial direction when there are no external events, and when an external event is received, the current physical inertial state is used as the starting point for state transition path planning.

[0029] (4) Cognitive Decision Module The cognitive decision module is used to receive external event inputs and the internal inertial situation identifier transmitted by the physical state feedback module, initialize the cognitive energy vector based on the internal inertial situation identifier, and drive the cognitive energy vector to undergo multi-step iterative evolution under the constraint of the potential energy field generated by the external event until it converges to a stable state.

[0030] The cognitive energy vector is a multi-dimensional energy vector, preferably a six-dimensional energy vector e=(e1,e2,e3,e4,e5,e6), where e_i∈[0,1], and the value of each dimension represents the system's tendency intensity in that dimension.

[0031] At the start of each cognitive decision, the cognitive decision-making module generates an initial energy vector based on the six-dimensional encoding vector of the situation type corresponding to the internal inertial situation identifier. Specifically, for the six-dimensional binary encoding vector y∈{0,1} corresponding to the internal inertial situation identifier... 6 The i-th dimension of the initial energy vector takes the following values: 0.7 if y_i=1, and 0.3 if y_i=0.

[0032] The cognitive decision-making module superimposes the disturbance energy generated by external events on the initial energy vector, driving the cognitive energy vector to undergo multi-step iterative evolution according to the preset energy evolution equation under the constraint of the potential energy field, until it converges to a stable state.

[0033] The cognitive decision-making module includes a situation transition path planning engine, which plans a situation transition path from the current situation to the target situation based on the current situation type, the target situation type, the global cognitive certainty U-value, the dynamic urgency, and the event type. The situation transition path planning engine uses the situation type corresponding to the internal inertial situation identifier as the starting point for path planning, ensuring that the planned path originates from the current physical motion state.

[0034] The situation transition path planning engine uses a graph search algorithm (preferably A* search algorithm or Dijkstra algorithm) to search for paths in the six-dimensional situation space, using Hamming distance as the heuristic function and preset state transition weights as edge costs to plan the transition path from the current situation to the target situation.

[0035] The energy evolution equation contains at least seven mechanical components: coupling term, taiji term, sharpening term, awareness term, source term, saturation term, and disturbance term.

[0036] The cognitive decision-making module outputs the converged current situation type, target situation type, situation migration path, and global cognitive certainty.

[0037] (5) Execution Module The execution module receives the situation type and situation transition path output by the cognitive decision module, queries a preset situation-physical action mapping table according to the situation type, generates corresponding actuator control commands (including but not limited to speed commands, steering commands, and braking force commands), and controls the actuator to perform physical actions through the actuator interface.

[0038] The situation-physical action mapping table contains at least the following mapping relationships: - Full-speed expansion state is mapped to full-speed forward motion; - High-speed movement is mapped to powerful forward motion; - Low-speed cautious state is mapped to cautious forward movement; - The static locked state is mapped to an emergency stop action; - A decelerated state is mapped to a decelerated action; - The deceleration state is mapped to a deceleration action; - The deceleration state three-map is a deceleration action; - The standby waiting state is mapped to a waiting action.

[0039] After completing the physical action, the execution module feeds back the actual execution results (actual speed, actual position, etc.) to the physical state sensing module, triggering a new round of physical state feedback.

[0040] 2. Methodology and Flow An embodied intelligent cognitive decision-making method based on the preservation of internal physical inertia includes the following steps: Step S1: Physical State Sensing Real-time motion feedback data from the actuator is acquired via a sensor interface, and the current physical motion state of the agent is represented as a physical state vector. This physical state vector includes at least the current linear velocity.

[0041] The physical state vector is continuously acquired and updated at a preset sampling frequency (preferably 50Hz to 200Hz).

[0042] Step S2: Velocity-Situation Mapping Based on the current linear velocity, the system queries the correspondence between the preset velocity range and the cognitive situation type, determines the cognitive situation type corresponding to the current velocity, and generates an internal inertial situation identifier.

[0043] The cognitive situation type comes from a preset complete situation space, which contains sixty-four basic situation types, each of which corresponds to a unique six-dimensional binary encoding vector.

[0044] The correspondence is as follows: - When the current linear velocity is higher than the first velocity threshold (preferably 0.8 m / s), it is mapped to the first cognitive situation type, preferably the full-speed expansion state, and the six-dimensional binary encoding vector is [1,1,1,1,1,1]; - When the current linear velocity is between the second velocity threshold and the first velocity threshold (preferably 0.4 m / s to 0.8 m / s), it is mapped to the second cognitive situation type, preferably the high-speed travel state, and the six-dimensional binary encoding vector is [1,1,1,1,0,0]; - When the current linear velocity is between the third velocity threshold and the second velocity threshold (preferably 0.1 m / s to 0.4 m / s), it is mapped to the third cognitive situation type, preferably the low-speed cautious state, and the six-dimensional binary encoding vector is [1,1,0,1,1,1]; - When the current linear velocity is lower than the third velocity threshold (preferably 0.1 m / s), it is mapped to the fourth cognitive situation type, preferably the stationary locked state, and the six-dimensional binary encoding vector is [0,0,1,0,0,1].

[0045] Step S3: Physical State Recharge At the beginning of each cognitive decision cycle, the latest physical state vector and the internal inertial situation identifier are transmitted to the cognitive decision module as the initial conditions for this cognitive decision.

[0046] Step S4: Cognitive Energy Initialization An initial energy vector is generated based on the six-dimensional binary encoding vector corresponding to the internal inertial situation identifier. The i-th dimension of the initial energy vector is set to 0.7 if the i-th bit of the encoding vector is 1, and 0.3 if it is 0.

[0047] Step S5: Target Situation Determination and Path Planning Obtain external event input, identify the event relationship type to which the external event belongs according to the preset event relationship syntax, query the preset relationship-situation mapping table, and determine the target situation type.

[0048] Based on the current situation type corresponding to the internal inertial situation identifier, the target situation type, the global cognitive certainty U value, and the dynamic urgency, the situation migration path planning engine plans the situation migration path from the current situation to the target situation.

[0049] The situation transition path consists of one or more intermediate situation types arranged in evolutionary order, with each step in the path changing only one dimension of the six-dimensional binary encoded vector. The situation transition path planning engine uses the situation type corresponding to the internal inertial situation identifier as the starting point of the path, ensuring that the planned path starts from the current physical motion state.

[0050] When there are multiple intermediate situations between the current situation type and the target situation type, the situation transition path planning engine prioritizes the selection of a path that includes a preset deceleration situation type. The deceleration situation type includes at least one of deceleration state one, deceleration state two, and deceleration state three, so as to achieve a smooth deceleration transition that conforms to the laws of physical motion.

[0051] Step S6: Situation Convergence and Game Theory The disturbance energy generated by the external event is superimposed on the initial energy vector, driving the cognitive energy vector to undergo multi-step iterative evolution according to the energy evolution equation under the constraint of the potential energy field, until it converges to a stable state.

[0052] The energy evolution equation is a discrete-time dynamic equation, which includes seven mechanical components: coupling term, Taiji term, sharpening term, awareness term, source term, saturation term, and disturbance term.

[0053] The preferred energy evolution equation is: Δe = α·(J@(e-0.5)) + β·(0.5-e) + γ·sign(e-0.5)·|e-0.5|^1.5 +η·(e-ē) + ζ·(0.5-e) - κ·(e-0.5)³ + d·ξ in: - e is a six-dimensional energy vector; - J is the total coupling matrix, which is formed by superimposing the basic coupling matrix and the specialized coupling matrix activated by external events; - ē is the mean of e; - ξ is a six-dimensional random perturbation vector; - α, β, γ, η, ζ, and κ are preset coefficients; - d represents the disturbance intensity.

[0054] The initial energy vector serves as the initial condition e0 of the energy evolution equation.

[0055] Step S7: Situation Output and Execution Output the current situation type and the situation transition path after convergence. Based on the current situation type, query the preset situation-physical action mapping table to generate actuator control instructions and control the actuator to perform physical actions.

[0056] When the situation transition path includes a deceleration situation type (including at least one of deceleration state one, deceleration state two, and deceleration state three), the execution module sequentially executes the physical actions corresponding to each situation type to achieve a smooth deceleration transition from high speed to low speed.

[0057] Step S8: Execution Feedback and Loopback After the actuator completes the physical action, it obtains the actual physical state vector after execution and returns to step S1, forming a complete cognitive closed loop of "perception → mapping → backfeeding → initialization → path planning → game → execution → feedback".

[0058] 3. Semantic locking of standby waiting state In this invention, the semantics of the standby waiting state (the six-dimensional binary encoded vector is [1,1,1,0,1,0]) is uniquely and immutably locked as "waiting posture", representing a waiting state of "waiting for the right time to act, without being impatient or agitated".

[0059] The standby state does not contain any semantics related to "deceleration". Deceleration is jointly undertaken by deceleration state one, deceleration state two, and deceleration state three. These three situation types are all mapped to deceleration actions in the situation-physical action mapping table.

[0060] The standby state is only transitioned to through the natural evolution of a sequential path when the system is already in a stopped or extremely low-speed state (i.e., the velocity-situation mapping is in a static locked state) and needs to wait for a long time for changes in external conditions.

[0061] 4. Inertia retention in the absence of external events When there is no external event input, the cognitive decision-making module evolves from the initial energy vector corresponding to the internal inertial state identifier. Since there is no interference from an external potential field, the cognitive energy vector remains near the state type corresponding to the internal inertial state, and the system's cognitive state naturally continues along the physical inertial direction.

[0062] The execution module continuously performs physical actions corresponding to the current internal inertial state, and the physical state perception module continuously acquires the actual physical state after execution and feeds it back to the cognitive decision-making module, so that the next cognitive decision is still based on the updated physical inertial state as the initial condition.

[0063] Through this mechanism, when there are no external events, the system's cognitive state naturally continues along the direction of physical inertia, and it will not lose its awareness of its own motion state due to "no new input".

[0064] (III) Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects: (1) The cognitive decision-making layer has achieved continuous perception of its own physical motion state. The present invention maps the continuous velocity parameters of the execution layer to discrete cognitive state types through the velocity-state mapping module, so that the cognitive layer can "understand" what speed it is moving at and what motion state it is in, thus bridging the semantic gap between the cognitive layer and the physical layer in the prior art.

[0065] (2) A complete cognitive closed loop of physical state feedback is realized. Through the physical state feedback module, the present invention enables the actual execution results of the execution layer to be continuously fed back to the cognitive layer as the initial condition for the next cognitive decision, forming a complete cognitive closed loop of "perception → mapping → feedback → initialization → path planning → game → execution → feedback", which enables the system to have a "proprioceptive" ability similar to that of a living organism.

[0066] (3) The cognitive preservation of internal physical inertia is achieved. This invention maps the physical state to a cognitive state and uses it as the initial energy vector, so that the system can naturally continue the cognitive state along the direction of physical inertia when there are no external events. When an external event is received, the system plans the state transition path starting from the current physical inertia state, rather than being completely covered by the external event, thereby ensuring the physical continuity and smoothness of decision-making.

[0067] (4) Makes the situation transition path planning physically realistic. The present invention uses the internal inertial situation identifier as the starting point of the situation transition path planning engine, so that the planned situation transition path starts from the current physical motion state, and achieves a smooth deceleration transition that conforms to the kinematic laws by prioritizing the selection of paths containing deceleration situation types (deceleration state one, deceleration state two, deceleration state three).

[0068] (5) A unified game between external events and internal inertia is realized. This invention makes the cognitive decision-making process no longer a one-way process of "external events covering internal state", but a two-way game process of "internal inertial state" and "external event gravity" in a unified energy field, and the decision results are more in line with physical reality.

[0069] (6) Each situation has a pure semantic meaning and performs its own function. This invention uniquely locks the semantic meaning of the standby waiting state as "waiting", and assigns the deceleration semantic meaning to deceleration state one, deceleration state two, and deceleration state three, avoiding cognitive confusion caused by a single situation type carrying multiple semantic meanings, making the situation language of the cognitive layer clearer and more interpretable.

[0070] (7) Provides a cognitive basis for embodied intelligence through “body perception”. This invention enables the cognitive system to not only “know” what is happening in the external environment, but also to “know” what its own body is doing and at what speed, laying the foundation for higher-level cognition in embodied intelligence (such as self-awareness and body schema). Attached Figure Description

[0071] Figure 1 : Module architecture diagram of the embodied intelligent cognitive decision-making system based on internal physical inertia maintenance described in this invention.

[0072] Figure 2A schematic diagram illustrating the interval division and correspondence of velocity-situation mapping, showing the mapping relationship between velocity thresholds (0.8m / s, 0.4m / s, 0.1m / s) and four types of situations: full-speed expansion state, high-speed movement state, low-speed cautious state, and stationary locked state.

[0073] Figure 3 The flowchart of physical state reinjection and cognitive energy initialization shows the complete data flow from physical state perception to velocity-state mapping to energy initialization.

[0074] Figure 4 A schematic diagram illustrating the energy evolution of the game between internal inertial state and external events, showing the evolution trajectory of the initial energy vector converging to a stable state under the constraint of the potential energy field.

[0075] Figure 5 The overall flowchart of the cognitive decision-making method described in this invention shows the complete process of steps S1 to S8.

[0076] Figure 6 : A schematic diagram of the correspondence between velocity-situation mapping and complete situation space.

[0077] Figure 7 A schematic diagram of the situation transition path planning including deceleration situation types, showing a smooth deceleration path from high-speed travel state through deceleration state to stationary locked state. Detailed Implementation

[0078] It should be understood that the following embodiments are used to illustrate the technical solutions of the present invention and not to limit it. Those skilled in the art can implement the complete technical solutions of the present invention based on the teachings of this specification. The system can be deployed in an edge computing device, which includes an in-vehicle domain controller, a robot main control chip, or an embedded intelligent processing unit.

[0079] Example 1: Velocity-Situation Mapping This embodiment demonstrates how the system maps the robot's current speed to the corresponding cognitive situation type.

[0080] Assume the robot's current linear velocity is 0.6 m / s. The system obtains the current velocity value through the physical state perception module and transmits it to the velocity-situation mapping module.

[0081] The speed-situation mapping module queries the correspondence between preset speed ranges and cognitive situation types: - Velocity > 0.8m / s → Full-speed expansion state, six-dimensional encoding [1,1,1,1,1,1]; - 0.4m / s < speed ≤ 0.8m / s → high-speed travel state, six-dimensional code [1,1,1,1,0,0]; - 0.1m / s < speed ≤ 0.4m / s → low-speed cautious state, six-dimensional encoding [1,1,0,1,1,1]; - Speed ​​≤ 0.1m / s → stationary locked state, six-dimensional code [0,0,1,0,0,1].

[0082] The current speed of 0.6 m / s falls into the second interval (0.4 to 0.8 m / s), which is mapped to a high-speed travel state. The speed-situation mapping module generates an internal inertial situation label "high-speed travel state", which represents that the system is currently in a medium-to-high speed motion state of "rigid and dynamic, strong".

[0083] Example 2: Physical State Reinjection and Cognitive Energy Initialization This embodiment demonstrates how the system feeds back the physical state to the cognitive layer and initializes cognitive energy.

[0084] Following Example 1, the current internal inertial state of the system is high-speed travel, and the six-dimensional code is [1,1,1,1,0,0].

[0085] The physical state feedback module transmits the high-speed travel state identifier to the cognitive decision module. The cognitive decision module generates an initial energy vector based on the six-dimensional encoding of the high-speed travel state. - For dimensions that are 1 in the encoding (dimensions 1, 2, 3, and 4), the value is 0.7; - For dimensions that are 0 in the encoding (the 5th and 6th dimensions), the value is 0.3.

[0086] The initial energy vector is e0 = [0.7, 0.7, 0.7, 0.7, 0.3, 0.3].

[0087] This initial energy vector serves as the starting point for this cognitive decision, representing the system "entering the cognitive decision process with an inertial velocity of 0.6 m / s." If the system receives an external event at this point, the disturbance energy generated by the external event will be superimposed on this initial energy vector, engaging in a struggle with the internal inertia within the energy field, rather than directly overwriting the internal state.

[0088] Example 3: The interplay between internal inertial state and external events, and path planning with deceleration state This embodiment demonstrates how the system engages in a game with external events while maintaining internal physical inertia, and how it plans a smooth transition path including deceleration situation types through a situation transition path planning engine.

[0089] The system is currently traveling at 0.6 m / s, with an internal inertial state of high-speed travel and an initial energy vector e0 = [0.7, 0.7, 0.7, 0.7, 0.3, 0.3].

[0090] The system receives an external event "Red light detected," which triggers an "obstacle-avoidance" causal chain. Based on a pre-defined relationship-situation mapping table, the "obstacle-avoidance" relationship associates a static locked state as the target situation type.

[0091] The situation transition path planning engine plans the situation transition path based on the current situation type (high-speed moving state, determined by the internal inertial situation identifier), the target situation type (stationary locked state), the global cognitive certainty U value (currently 0.12), and the dynamic urgency (currently 2.0).

[0092] The situation transition path planning engine uses the situation type (high-speed travel state) corresponding to the internal inertial situation identifier as the starting point of the path, ensuring that the planned path starts from the current physical motion state. The path planning result is high-speed travel state → deceleration state → stationary locked state.

[0093] The path includes a deceleration state three, which is mapped to a deceleration action in the situation-physical action mapping table. This allows the system to naturally include a deceleration phase during the transition from a high-speed traveling state to a stationary locked state, achieving a smooth deceleration transition that conforms to the laws of physical motion.

[0094] Based on the initial energy vector e0, the cognitive decision-making module performs multi-step iterative evolution according to the energy evolution equation, eventually converging to deceleration state three, and then continuing to evolve to the stationary locked state.

[0095] The execution module executes the physical actions corresponding to each situation type in the path sequentially: - High-speed movement → Powerful forward movement (maintain current speed); - Deceleration state three → Deceleration (smooth deceleration); - Stationary locked state → Emergency stop (complete stop).

[0096] The system smoothly transitions from "high-speed motion" to "complete stop" via a path that includes deceleration status types, rather than jumping directly from "high-speed motion" to "stop". Once the speed drops below 0.1 m / s, the velocity-situation mapping automatically updates the internal inertial status to a stationary locked state.

[0097] If the system needs to wait for a long time after stopping (such as waiting for a red light to turn green), the path planned by the situation transition path planning engine will transition from the static locked state to the standby waiting state. The standby waiting state represents the waiting state of "waiting for the right time to act and waiting for the right opportunity".

[0098] Example 4: Inertial retention in the absence of external events This embodiment demonstrates how the system maintains physical inertia in the absence of external events.

[0099] The system is currently traveling at 0.6 m / s, and its internal inertial state is that of high-speed travel. Without external event input, the cognitive decision-making module evolves from an initial energy vector e0 = [0.7, 0.7, 0.7, 0.7, 0.3, 0.3]. Due to the absence of external potential energy field interference, the energy vector remains near the high-speed travel state, and the system continuously outputs the cognitive judgment of "medium-high speed forward" and the corresponding execution command.

[0100] After the execution module executes the forward command, the physical state perception module obtains the actual speed after execution (assuming it is still 0.6m / s). The physical state feedback module feeds back the updated physical state to the cognitive decision module, so that the next cognitive decision is still based on the high-speed travel state as the initial condition.

[0101] Through this mechanism, when there are no external events, the system's cognitive state naturally continues along the direction of physical inertia, and it will not lose its awareness of its own motion state due to "lack of new input." This gives the system a "body continuity perception" similar to that of a living organism—it knows what speed it is moving at, even with its eyes closed.

[0102] Example 5: Automatic situation update after velocity change This embodiment demonstrates how the system automatically updates its internal inertial state after a change in speed.

[0103] The system initially travels at 0.6 m / s, maintaining a high-speed state due to internal inertia. During execution, the system's speed decreases to 0.3 m / s due to a slight uphill incline.

[0104] The physical state perception module detects a velocity change and transmits the new velocity value of 0.3 m / s to the velocity-situation mapping module. The velocity-situation mapping module queries the mapping table and finds that 0.3 m / s falls into the third interval (0.1 to 0.4 m / s), which is mapped to a low-speed cautious state.

[0105] The physical state feedback module transmits the new internal inertial state "low-speed cautious state" to the cognitive decision module. The cognitive decision module generates a new initial energy vector e0 = [0.7,0.7, 0.3, 0.7, 0.7, 0.7] using the six-dimensional encoding [1,1,0,1,1,1] of the low-speed cautious state.

[0106] The system automatically updated its cognition from a "medium-to-high speed motion state" to a "low-speed, cautious motion state" without the need for external events. This mechanism ensures that the system's cognitive state is always synchronized with its physical motion state, realizing the core idea of ​​embodied intelligence: "body is cognition."

[0107] Example 6: Comparison of mappings across different speed ranges This embodiment compares the mapping results at different speeds to comprehensively demonstrate the speed-situation mapping mechanism: Full speed sprint 1.2m / s Full-speed expansion state [1,1,1,1,1,1] All-dimensional hyperactivity, maximum expansion Normal driving 0.6m / s High-speed travel [1,1,1,1,0,0] Strong and vigorous Slowly 0.3m / s Low-speed cautious state [1,1,0,1,1,1] Distinguish between superiors and inferiors, and act with caution. stop 0.05m / s Static locked state [0,0,1,0,0,1] Stop where you want to stop, suspend observation As the speed gradually decreases from 1.2 m / s to 0.05 m / s, the internal inertial state undergoes a complete evolutionary sequence of full-speed expansion state → high-speed travel state → low-speed cautious state → stationary locked state. The cognitive layer can clearly perceive its own complete motion process from "full speed" to "stop".

[0108] Example 7: Long path planning with multiple deceleration states This embodiment demonstrates how, when the target situation is far away, the situation transition path planning engine plans a long path that includes multiple deceleration situations.

[0109] The system is currently traveling at a high speed of 1.2 m / s, and its internal inertial state is in a state of full-speed expansion. The system has received an external event: "There is a toll station 500 meters ahead; you need to stop."

[0110] The causal chain is identified as "obstacle-avoidance," and the target situation is locked at a stationary state. The situation transition path planning engine plans the situation transition path based on the current full-speed expansion state, the target's stationary state, the U-value (0.08), and the urgency level (1.5): Full-speed expansion state → High-speed travel state → Deceleration state three → Deceleration state two → Stationary locked state.

[0111] The path includes two deceleration states (deceleration state three and deceleration state two), enabling the system to smoothly transition from full speed through medium speed and deceleration stages to a stop, which fully conforms to the laws of real physical motion.

[0112] The execution layer executes sequentially: - Full-speed expansion state → Full-speed forward movement; - High-speed movement → Powerful forward movement; - Deceleration state three → Deceleration; - Deceleration state two → Deceleration; - Stationary locked state → Emergency stop.

[0113] Example 8: Comparison with the prior art This embodiment further illustrates the inventiveness of the present invention by comparing it with the prior art.

[0114] Existing technology 1 (China FAW humanoid robot patent): Improves the robot's adaptability to internal disturbances through a training system. This technology focuses on the stability of low-level motion control—how the control system adjusts joint torques to maintain body balance when the robot's joints experience inertial coupling or center of gravity shift. Its technical path is "physical disturbance → execution layer compensation," which belongs to execution layer optimization. This technology does not involve abstracting the physical state into a cognitive situation and feeding it back to the cognitive decision-making layer; its physical state information is only used within the execution layer and does not participate in high-level cognitive decision-making.

[0115] Existing technology 2 (a robot motion state cognition method based on the cognitive mechanism of the rat brain hippocampus): speed is perceived through velocity cells and visual information, but its goal is to recognize the motion state and position (i.e. "where am I" and "how fast am I moving"), without mapping the physical state to a high-level situation type to participate in cognitive decision-making, and without involving the inertial maintenance mechanism of the physical state for cognitive situation judgment.

[0116] Existing technology three (the applicant's existing patents such as CN202610506445.X): has disclosed a state representation method based on a six-dimensional energy vector, a complete situation space, and a situation transition path planning method. However, the situation transition path planning in these patents is only based on internal cognitive layer parameters such as event type, uncertainty value, and urgency, without taking physical velocity as an input condition for path planning, nor providing a specific method for mapping continuous velocity parameters to discrete situation types.

[0117] The essential difference between this invention and the aforementioned prior art lies in: (1) Different technical paths: Existing technologies are "physical disturbance → execution layer compensation", existing technologies are "physical state → position / velocity cognition", and the present invention is a complete closed loop of "physical state → cognitive situation mapping → cognitive decision → motion control → physical state feedback".

[0118] (2) Different target levels: Existing technologies one and two serve the stability and accuracy of low-level motion control, while this invention serves the physical continuity and situational awareness of high-level cognitive decision-making.

[0119] (3) Different information flow: In the prior art, physical state information is only used for internal execution layer or location cognition. In this invention, physical state information is abstracted into cognitive situation and then fed back to the cognitive layer to participate in high-level decision-making.

[0120] (4) Different core mechanisms: There is no cognitive abstraction layer of "velocity-situation mapping" in the existing technology, nor is there a mechanism that uses physical inertia as the initial condition for cognitive decision-making and the starting point of situational migration path.

[0121] (5) Complementary relationship with existing cognitive patents: Based on the cognitive framework of existing patents (CN202610506445.X, etc.), this invention adds physical state perception, velocity-situation mapping and physical state feedback mechanism, enabling the situation migration path planning engine to take physical inertial situation as the path starting point, filling the "cognitive-physical gap".

Claims

1. An embodied intelligent cognitive decision-making method based on the preservation of internal physical inertia, characterized in that, Includes the following steps: Step S1: Obtain real-time motion feedback data of the actuator through the sensor interface, and represent the current physical motion state of the intelligent agent as a physical state vector, wherein the physical state vector includes at least the current linear velocity; Step S2: Based on the current linear velocity, query the correspondence between the preset velocity range and the cognitive situation type, determine the cognitive situation type corresponding to the current velocity, and generate an internal inertial situation identifier; Step S3: At the beginning of each cognitive decision cycle, the internal inertial state identifier is transmitted to the cognitive decision module as the initial condition for this cognitive decision. Step S4: Generate an initial energy vector based on the encoding vector of the situation type corresponding to the internal inertial situation identifier; Step S5: Obtain external event input, determine the target situation type, and plan a situation migration path from the current situation to the target situation using the situation type corresponding to the internal inertial situation identifier as the starting point through the situation migration path planning engine. Step S6: Convert the external event into perturbation energy, and superimpose the perturbation energy on the initial energy vector to drive the cognitive energy vector to undergo multi-step iterative evolution under the constraint of the potential energy field until it converges to a stable state; Step S7: Output the current situation type after convergence and the situation transition path, generate actuator control instructions based on the current situation type, and control the actuator to perform physical actions; Step S8: After the executor completes the physical action, obtain the actual physical state vector after execution and return to step S1.

2. The method according to claim 1, characterized in that, The cognitive situation type comes from a preset complete situation space, which contains sixty-four basic situation types, each corresponding to a unique six-dimensional binary encoding vector; the correspondence between the velocity range and the cognitive situation type is as follows: - When the current linear velocity is higher than the first velocity threshold, it is mapped to the first cognitive situation type; - When the current linear velocity is between the second velocity threshold and the first velocity threshold, it is mapped to the second cognitive situation type; - When the current linear velocity is between the third velocity threshold and the second velocity threshold, it is mapped to the third cognitive situation type; - When the current linear velocity is lower than the third velocity threshold, it is mapped to the fourth cognitive situation type.

3. The method according to claim 2, characterized in that, The first speed threshold is 0.8 m / s, the second speed threshold is 0.4 m / s, and the third speed threshold is 0.1 m / s; The first cognitive state type is the full-speed expansion state, and the six-dimensional binary encoding vector is [1,1,1,1,1,1]; The second cognitive state type is high-speed movement state, and the six-dimensional binary encoding vector is [1,1,1,1,0,0]; The third cognitive state type is a low-speed cautious state, and the six-dimensional binary encoding vector is [1,1,0,1,1,1]; The fourth cognitive state type is a static locked state, and the six-dimensional binary encoding vector is [0,0,1,0,0,1].

4. The method according to claim 1, characterized in that, In step S4, the initial energy vector is generated as follows: for the six-dimensional binary encoded vector y∈{0,1} corresponding to the internal inertial situation marker. 6 The i-th dimension of the initial energy vector takes the following values: 0.7 if y_i=1, and 0.3 if y_i=0.

5. The method according to claim 1, characterized in that, In step S6, the energy evolution equation is: Δe = α·(J@(e-0.5)) + β·(0.5-e) + γ·sign(e-0.5)·|e-0.5|^1.5 + η·(e-ē) + ζ·(0.5-e) - κ·(e-0.5)³ + d·ξ Where e is a six-dimensional energy vector, J is the total coupling matrix, ē is the mean of e, ξ is a six-dimensional random perturbation vector, α, β, γ, η, ζ, κ are preset coefficients, and d is the perturbation intensity; the initial energy vector serves as the initial condition for the energy evolution equation.

6. The method according to claim 1, characterized in that, In step S5, when the situation transition path includes a deceleration situation type, the deceleration situation type includes at least one of deceleration state one, deceleration state two, and deceleration state three; the situation transition path planning engine preferentially selects the path containing the deceleration situation type to achieve a smooth deceleration transition that conforms to the laws of physical motion.

7. The method according to claim 1, characterized in that, Step S3 further includes: when there is no external event input, the cognitive decision module evolves from the initial energy vector corresponding to the internal inertial state identifier, maintains the internal inertial state, and allows the cognitive state of the system to continue naturally along the physical inertial direction.

8. An embodied intelligent cognitive decision-making system based on internal physical inertia retention, characterized in that, The processor includes a processor, a memory, and sensor and actuator interfaces connected to the processor. The processor is configured to implement the following modules by executing program instructions stored in the memory: The physical state perception module is used to acquire real-time motion feedback data of the actuator through the sensor interface and represent the current physical motion state of the agent as a physical state vector. The velocity-situation mapping module is used to query the correspondence between a preset velocity range and the cognitive situation type based on the current linear velocity in the physical state vector, and generate an internal inertial situation identifier. The physical state feedback module is used to transmit the internal inertial state identifier to the cognitive decision module at the beginning of each cognitive decision cycle; The cognitive decision-making module is used to generate an initial energy vector based on the encoding vector of the situation type corresponding to the internal inertial situation identifier, determine the target situation type based on the external event input, plan the situation migration path with the situation type corresponding to the internal inertial situation identifier as the path starting point through the situation migration path planning engine, and superimpose the disturbance energy generated by the external event on the initial energy vector to drive the cognitive energy vector to undergo multi-step iterative evolution until it converges to a stable situation. The execution module is used to generate actuator control commands based on the situation type and situation migration path output by the cognitive decision module, control the actuator to perform physical actions through the actuator interface, and feed back the actual physical state after execution to the physical state perception module.

9. The system according to claim 8, characterized in that, In the velocity-situation mapping module, the cognitive situation type comes from a preset complete situation space, which contains sixty-four basic situation types, each corresponding to a unique six-dimensional binary encoding vector; the correspondence between the velocity range and the cognitive situation type is as follows: - Velocity > 0.8m / s → Full-speed expansion state, encoded as [1,1,1,1,1,1]; - 0.4m / s < speed ≤ 0.8m / s → high-speed travel state, encoded [1,1,1,1,0,0]; - 0.1m / s < speed ≤ 0.4m / s → low-speed cautious state, encoded as [1,1,0,1,1,1]; - Speed ​​≤ 0.1m / s → stationary locked state, encoded [0,0,1,0,0,1].

10. 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 7.

Images