Cognitive-physical collaborative embodied intelligent control method and system

Abstract

This invention discloses a cognitive-physical collaborative embodied intelligence control method and system, belonging to the interdisciplinary field of embodied intelligence and automatic control. The method includes: a cognitive layer, based on event relation grammar and a complete situation space, emerges a target cognitive situation through the evolution equation of the first seven mechanical components; a physical layer acquires the current physical state and constructs an n-dimensional physical situation vector; according to the semantic features of the target cognitive situation, it is mapped to a neutral steady state where all dimensions of the physical layer take 0.5 through mapping rules; the physical layer uses the second seven mechanical component equation to drive the physical situation vector to dynamically converge towards the 0.5 neutral steady state, and the dynamic priority of the cognitive situation can modulate the coupling matrix and sharpening coefficient of the physical layer in real time; when the control certainty of the physical layer exceeds the limit, it can backflow to the cognitive layer to force convergence to a conservative defensive situation. This invention, for the first time, unifies cognitive decision-making and physical execution under the same set of dynamic laws, achieving true "unity of knowledge and action" and dynamic equilibrium control at the dissipative structure level through a closed-loop structure of "cognition begins at 0.5, physics returns to 0.5," possessing extremely strong cross-domain endogenous security capabilities and can be widely applied to various embodied intelligent agents.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of artificial intelligence, cognitive computing, and automatic control. Specifically, it relates to an embodied intelligent control method and system that unifies cognitive decision-making and physical execution under the same set of seven mechanical component energy evolution laws. This invention can be applied to embodied intelligent systems that require deep coupling between the "brain" and the "body," such as humanoid robots, quadruped robots, and autonomous vehicles. Background Technology

[0002] Existing embodied intelligence systems typically employ a hierarchical architecture of "perception-decision-control". At the decision-making level, large language models, deep neural networks, or rule-based reasoning systems (the brain) are commonly used; at the execution level, traditional low-level control algorithms such as PID control and model predictive control (MPC) are typically employed (the body). This fragmented architecture has a fundamental flaw: (1) Cognition and physical control use two completely different mathematical frameworks, which makes it impossible for the abstract decision-making instructions generated by the "brain" to be accurately and internally understood and executed by the "body", resulting in serious cross-domain information loss.

[0003] (2) The decision-making level lacks awareness of the uncertainty of physical control, while the physical level lacks understanding of the high-dimensional cognitive situation, resulting in the system lacking the global "body-mind unity" metacognitive ability.

[0004] (3) Traditional control algorithms (such as MPC) have a large computational load and are difficult to run in real time on resource-constrained embedded platforms; traditional cognitive decision-making models lack endogenous security mechanisms and rely on externally superimposed rules for security protection, which are easily bypassed.

[0005] The applicant's prior patent application (application number: 202610601934.3) discloses a causal spatiotemporal continuum cognitive model based on event relation grammar, which uses a seven-component energy evolution equation to drive a six-dimensional energy vector to emerge discrete cognitive states in a complete state space; the applicant also filed a prior application for "A system state control method and system based on the energy evolution of seven-component mechanical components", which reveals a mechanism for using the same seven-component mechanical component equation to drive the system state vector to converge to a 0.5 neutral steady state to achieve physical dynamic equilibrium control.

[0006] However, the aforementioned prior applications focus on the independent implementation of "cognitive emergence" and "physical control," respectively, and have not yet provided a complete technical solution for mapping discrete cognitive states across the semantic gap to a 0.5 neutral control objective at the physical layer, and for utilizing the same set of seven mechanical component equations to work collaboratively at both the cognitive and physical levels (especially by dynamically modulating physical layer parameters using cognitive states). This disconnect between cognition and physics prevents the system from achieving true "unity of knowledge and action" and cross-domain intrinsic security. Summary of the Invention

[0007] (a) Purpose of the invention The purpose of this invention is to provide a cognitive-physical collaborative embodied intelligent control method and system to solve technical problems in the prior art, such as the cross-domain disconnect between cognitive decision-making and physical execution, the lack of unified dynamic laws, and the lack of endogenous cross-domain security mechanisms.

[0008] Specific objectives include: (1) A cross-domain bridge method is provided to map discrete cognitive states to a 0.5 physical layer neutral steady state.

[0009] (2) A unified control framework is provided in which cognition and physics share the same set of seven mechanical component energy evolution equations, and the dynamic modulation mechanism of physical layer parameters through cognitive situation is revealed.

[0010] (3) Provides an endogenous security mechanism that links cognitive certainty and physical control certainty in two ways.

[0011] (4) A dynamic equilibrium control method based on the principle of "dissipative structure" is provided so that the system always reverts to the 0.5 neutral steady state when performing complex motion.

[0012] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A cognitive-physical collaborative embodied intelligence control method includes the following steps: Step S1: Cognitive Situation Emergence The input event sequence is obtained, and the first situation vector is driven to converge toward the target cognitive situation through the first seven mechanical component energy evolution equations, based on the preset event relationship syntax and complete situation space, thereby giving rise to the target cognitive situation.

[0013] The energy evolution equation for the first seven mechanical components is as follows: Δe_cog = α_c·J_c·(e_cog-0.5) + β_c·(0.5-e_cog) + γ_c·sign(e_cog-0.5)·|e_cog-0.5|^1.5 + η_c·(e_cog-ē_cog) + ζ_c·(0.5-e_cog) - κ_c·(e_cog-0.5)³ + d_c·ξ_c Where e_cog is the n-dimensional cognitive situation vector of the cognitive layer, J_c is the cognitive layer coupling matrix, α_c, β_c, γ_c, η_c, ζ_c, and κ_c are preset coefficients of the cognitive layer, d_c is the cognitive layer perturbation intensity, and ξ_c is the cognitive layer random perturbation vector; the value of the cognitive layer coupling matrix J_c is dynamically adjusted according to the situation attractor activated by the input event.

[0014] Step S2: Physical State Perception Obtain the current physical state information of the embodied intelligent agent, the physical state information including at least a set of continuous physical quantities characterizing the current motion state or posture state of the embodied intelligent agent; construct an n-dimensional physical situation vector e_phy based on the current physical state information, where n is a positive integer greater than or equal to 2.

[0015] Step S3: Cross-domain dynamic mapping Based on the semantic features of the target cognitive state, the target cognitive state is mapped to a target control vector at the physical layer through a preset state-physics mapping rule. The target control vector is a neutral steady state in which all dimensions are 0.5. The 0.5 is the normalized mapping value when the physical quantity deviation is zero when the physical system is in a dynamic equilibrium state. It is also the fixed mathematical anchor point of the source term ζ·(0.5-e) and the Taiji term β·(0.5-e) in the energy evolution equations of the first and second seven mechanical components.

[0016] Step S4: Physical layer dynamic equilibrium convergence Using the energy evolution equation of the second and seventh mechanical components, the n-dimensional physical state vector is driven to iteratively converge toward the 0.5 neutral steady state based on the difference between the n-dimensional physical state vector and the 0.5 neutral steady state.

[0017] The energy evolution equation for the second and seventh mechanical components is as follows: Δe_phy = α_p·J_p·(e_phy-0.5) + β_p·(0.5-e_phy) + γ_p·sign(e_phy-0.5)·|e_phy-0.5|^1.5 + η_p·(e_phy-ē_phy) + ζ_p·(0.5-e_phy) - κ_p·(e_phy-0.5)³ + d_p·ξ_p Where e_phy is the n-dimensional physical state vector of the physical layer, J_p is the physical layer coupling matrix, α_p, β_p, γ_p, η_p, ζ_p, and κ_p are the preset coefficients of the physical layer, d_p is the physical layer perturbation intensity, and ξ_p is the physical layer random perturbation vector.

[0018] Core coordination mechanism (parameter modulation in step S4): When the target perception situation that emerges in step S1 is an emergency avoidance situation or a conservative defense situation, the cross-dimensional coupling coefficient in the physical layer coupling matrix J_p is increased, and the physical layer sharpening term coefficient γ_p is increased simultaneously, so that the physical situation vector converges quickly to 0.5; when the perception situation is a normal stable situation, the physical layer maintains the standard coupling coefficient to achieve a smooth transition.

[0019] Step S5: Execute control When the difference between the n-dimensional physical state vector and the 0.5 neutral steady state is less than a preset convergence threshold, an actuator control command is generated based on the converged n-dimensional physical state vector to control the actuator action of the embodied intelligent agent.

[0020] Furthermore, the method also includes a cross-layer security backflow mechanism: The control certainty of the physical layer is calculated in real time as U_phy=(1 / n)∑(e_phy_i-ē_phy)², where n is the number of dimensions, e_phy_i is the i-th component of the physical situation vector, and ē_phy is the mean of all components. When U_phy exceeds the preset physical warning threshold, the abnormal signal is fed back to the cognitive layer, which forces the coupling term coefficient α_c of the cognitive layer to decrease and the sharpening term coefficient γ_c to increase, so that the cognitive layer is forced to converge to a conservative defensive posture, thereby realizing the security protection of the cognitive layer by the physical layer.

[0021] (III) Beneficial Effects (1) It realizes the unified law of cognition and physics. For the first time, this invention unifies cognitive decision-making and physical execution in the same set of seven mechanical component energy evolution equations (only the parameter set is different), breaking the traditional system's "brain uses one algorithm and the body uses another algorithm" separation, realizing absolute logical unity and greatly simplifying the system architecture.

[0022] (2) It achieves true "unity of knowledge and action" and cross-domain modulation. The cognitive layer starts from 0.5 and emerges a discrete target situation through information game; the physical layer maps the target to 0.5 and returns to the neutral steady state of 0.5 through dynamic game. At the same time, the cognitive situation can dynamically modulate the physical layer coupling matrix J_p and sharpening coefficient γ_p in real time, forming a complete closed loop of "intelligent control body".

[0023] (3) It has an inherent security mechanism that crosses domains. When the physical layer certainty U_phy exceeds the limit, the present invention can suppress the divergence of the cognitive layer in reverse, forcing the system to make conservative decisions under high uncertainty; this security mechanism is an inherent property of the system dynamics architecture, does not depend on external instructions, and cannot be bypassed.

[0024] (4) Supports the dynamic stability of dissipative structures. Under strong disturbances (such as a robot running or a car making a sharp turn), the physical layer excites state deviation through the disturbance term and pulls it back to 0.5 by the nonlinear force of the source term and the saturation term, so that the system can maintain "continuous operation in a stable state" even when it is in high-speed motion, which truly conforms to the characteristics of dissipative structures in the physical world. Attached Figure Description

[0025] Figure 1 : Overall flowchart of the cognitive-physical collaborative control method described in this invention.

[0026] Figure 2 A schematic diagram of the cross-domain mapping from the cognitive state to the physical 0.5 neutral steady state.

[0027] Figure 3 A schematic diagram of the co-evolutionary trajectory of cognitive layer emergence (starting from 0.5) and physical layer convergence (returning to 0.5).

[0028] Figure 4 Flowchart of the intrinsic security mechanism for cross-layer U-value linkage.

[0029] Figure 5 : Module architecture diagram of the cognitive-physical collaborative control system described in this invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0031] Example 1: Cognitive-Physical Coordinated Balance Control of Robots This embodiment demonstrates a complete implementation of the invention on a robot.

[0032] Step 1 (Cognitive Layer): The robot receives the image event "Icy road ahead" and activates the obstacle-avoidance relationship. The cognitive layer uses the first and seventh mechanical component energy evolution equation to emerge the target cognitive state as "stop observing state" (conservative defensive state) in the complete situation space.

[0033] Step 2 (Physical Sensing): The IMU acquires the robot's attitude angle as 0.35 rad and angular velocity as 0.4 rad / s. Based on the preset normalization parameters, construct a physical situation vector e_phy = [0.80, 0.70, 0.5, 0.65, 0.5, 0.5] with n=6.

[0034] Step 3 (Cross-Domain Mapping): Based on the semantics of "stopped observation state," the system maps the target to a 0.5 neutral steady state in the physical layer. That is, the pitch angle deviation must be zero, the angular velocity must be zero, and the center of gravity projection must be located at the center of the support surface. The physical target vector e_target = [0.5, 0.5, 0.5, 0.5, 0.5, 0.5].

[0035] Step 4 (Physical Convergence and Parameter Modulation): The physical layer uses the 27th mechanical component equation to drive e_phy to converge towards 0.5. Since the cognitive layer exhibits a "conservative defensive posture" (highest priority), the system initiates a parameter modulation mechanism: the pitch-roll coupling coefficient in the physical layer coupling matrix J_p is drastically increased (from 0.2 to 0.5), while the sharpening term coefficient γ_p increases simultaneously. This modulation enables the physical system to converge to near 0.5 within 40 steps, allowing the robot to achieve an emergency stop and attitude stabilization.

[0036] Example 2: Cross-layer safety backflow mechanism This embodiment demonstrates the system's inherent safety protection when performing dangerous actions.

[0037] The robot's current cognitive state is "high-speed running". Due to slippery road conditions, the calculated physical control certainty U_phy rapidly increases to 0.55, exceeding the preset physical warning threshold of 0.40. The cross-layer safety backflow mechanism is immediately triggered: the system feeds this abnormal signal back to the cognitive layer, forcing a change in the parameters of the cognitive layer's first equation—the coupling term coefficient α_c decreases, and the sharpening term coefficient γ_c increases. The cognitive layer is forced to stop divergent exploration and converge to a "conservative defensive stance". The physical layer then reverts the target speed to the safe value corresponding to 0.5, and the robot safely stops.

[0038] Example 3: Cooperative Control of Automated Driving in Vehicles This embodiment demonstrates how the present invention is implemented in an autonomous vehicle.

[0039] The cognitive layer detects the red light ahead and enters a "stop and observe" state in the complete situation space. The physical layer perceives the current speed deviation (80 km / h to a 0 target) and constructs a physical situation vector. The cognitive state is conservative and defensive, modulating the physical layer's J_p and γ_p to ensure rapid braking without slippage or overshoot. After smoothly decelerating to the target speed, the physical situation vector accurately returns to 0.5, achieving perfect coordination between cognition and physics.

[0040] Example 4: General Applicability Description The method described in this invention is not only applicable to robots and autonomous vehicles, but also to all scenarios requiring deep coupling of "cognitive decision-making" and "physical / process control," such as industrial robotic arm trajectory control, environmental temperature control systems, and energy dispatching systems. Those skilled in the art can achieve the technical solution of this invention simply by adjusting the dimension definition of the n-dimensional situation vector and the value of the coupling matrix according to the physical characteristics of the specific system, and by uniformly mapping the control target to 0.5.

Claims

1. A cognitive-physical collaborative embodied intelligent control method, characterized in that, Includes the following steps: Step S1: Obtain the input event sequence, and according to the preset event relationship syntax and complete situation space, drive the first situation vector to converge towards the target cognitive situation through the first seven mechanical component energy evolution equation, so that the target cognitive situation emerges. Step S2: Obtain the current physical state information of the embodied intelligent agent and construct an n-dimensional physical situation vector, where n is a positive integer greater than or equal to 2; Step S3: Based on the semantic features of the target cognitive situation, the target cognitive situation is mapped to a target control vector in the physical layer through a preset situation-physical mapping rule. The target control vector is a neutral steady state in which all dimensions are 0.

5. Step S4: Using the energy evolution equation of the second and seventh mechanical components, drive the n-dimensional physical state vector to converge iteratively towards the 0.5 neutral steady state; Step S5: When the difference between the n-dimensional physical state vector and the 0.5 neutral steady state is less than a preset convergence threshold, an actuator control command is generated to control the actuator action of the embodied intelligent agent.

2. The method according to claim 1, characterized in that, The situation-physical mapping rule in step S3 is as follows: when the target's perceived situation is a stop-observation state or a conservative defensive state, the mapped physical target is an actuator speed deviation of 0, a position deviation of 0, and an attitude angle deviation of 0; when the target's perceived situation is a cautious forward movement state, the mapped physical target is to maintain a preset safe gait phase and attitude balance margin.

3. The method according to claim 1, characterized in that, In step S4, the value of the physical layer coupling matrix J_p in the energy evolution equation of the second seventh mechanical component is dynamically adjusted not only according to the current working mode of the physical system, but also according to the priority of the target cognitive situation that emerges in step S1. When the target cognitive situation is an emergency avoidance situation or a conservative defense situation, the cross-dimensional coupling coefficient in the physical layer coupling matrix J_p increases, and the physical layer sharpening term coefficient γ_p increases to accelerate convergence to 0.

5.

4. The method according to claim 1, characterized in that, It also includes a cross-layer security backflow mechanism: real-time calculation of the control certainty of the physical layer U_phy, U_phy=(1 / n)∑(e_phy_i-ē_phy)², where n is the number of dimensions, e_phy_i is the i-th component of the physical situation vector, and ē_phy is the mean; when U_phy exceeds the preset physical warning threshold, the abnormal signal is fed back to the cognitive layer, forcing the cognitive layer to converge to a conservative defensive posture, thereby realizing the security protection of the cognitive layer by the physical layer.

5. The method according to claim 1, characterized in that, The 0.5 neutral steady state is the normalized mapping value when the deviation of physical quantities is zero when the physical system is in a dynamic equilibrium state; the 0.5 is also the fixed mathematical anchor point of the source term ζ·(0.5-e) and the Taiji term β·(0.5-e) in the energy evolution equation of the first and second seven mechanical components.

6. The method according to claim 1, characterized in that, The complete situation space consists of 2^n basic situation types, each of which corresponds to a unique n-dimensional binary encoding vector; the preset event relationship syntax contains multiple event relationship types, each of which is associated with a specific basic situation type as a situation attractor.

7. The method according to claim 1, characterized in that, The physical layer dynamic equilibrium convergence in step S4 has dissipative structure characteristics: during the physical layer dynamic equilibrium convergence process, the external real-time disturbance force stimulates the physical state vector to deviate from 0.5 through the disturbance term d·ξ. At the same time, the source term, coupling term and saturation term work together to enable the system to maintain the stable operating state corresponding to 0.5 in high dynamic motion (running, sharp turning).

8. The method according to claim 1, characterized in that, In step S2, the construction of the n-dimensional physical state vector is obtained by normalizing and mapping the physical quantity deviation of the current physical state information; the normalization mapping precisely maps the "zero deviation state" to 0.

5.

9. A cognitive-physical collaborative embodied intelligent control system, characterized in that, include: The cognitive decision-making module is used to acquire the input event sequence and, based on the preset event relationship syntax and complete situation space, drive the first situation vector to converge toward the target cognitive situation through the energy evolution equation of the first seven mechanical components. The physical state perception module is used to acquire the current physical state information of the embodied intelligent agent and construct an n-dimensional physical situation vector; The cross-domain mapping module is used to map the target cognitive situation into a target control vector at the physical layer based on the semantic features of the target cognitive situation and through a preset situation-physical mapping rule. The target control vector is a neutral steady state with all dimensions taking 0.

5. The physical convergence module is used to drive the n-dimensional physical state vector to converge iteratively towards the 0.5 neutral steady state using the energy evolution equation of the second and seventh mechanical components. The execution control module is used to generate actuator control commands when the difference between the n-dimensional physical state vector and the 0.5 neutral steady state is less than a preset convergence threshold.

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 8.

Images