Dynamic interaction cognitive and motion control method and system based on topological field, and medium

Abstract

The application discloses a dynamic interaction cognition and motion control method and system based on a topological field and a medium, and belongs to the technical field of autonomous motion control. The application constructs an interaction cognition engine, and unifies environment perception, interaction cognition and motion control into an endogenous closed loop: endogenous construction of a dynamic interaction topological graph with objects as nodes and normalized interaction risks as edge weights by real-time collection of perception data; generation of a global topological potential field by superimposing three types of potential fields of tasks, safety and interaction, wherein the interaction potential field is directly generated by superimposing the topological graph edge weight and the risk point; generation of a motion trajectory along the negative gradient of the potential field, projection combined with dynamic constraints; and realization of the collaborative evolution of the three by reverse correction of the topological graph and the potential field parameters through motion deviation. The decision-making process is endogenous and interpretable, the algorithmic power requirement is low, the obstacle avoidance success rate reaches 99.9%, and the system can safely operate in a complex dynamic environment without external rules, and is suitable for autonomous navigation and obstacle avoidance scenes of various mobile platforms and dynamic interaction systems.

Description

Technical Field

[0001] This invention relates to the fields of autonomous motion control, intelligent robots, and intelligent transportation, specifically to a dynamic interactive cognition and motion control method, system, storage medium, and mobile platform based on topological fields. Background Technology

[0002] Current autonomous motion control in mobile platforms and other dynamic interactive systems primarily employs a linear modular architecture of "perception-decision-control" or an end-to-end black-box architecture. The former separates perception and control, relying on manually preset rules, while the latter suffers from uninterpretable decision-making and poor generalization ability. Both use fixed spatiotemporal parameters, failing to adapt to dynamic environmental changes, resulting in low obstacle avoidance success rates, high response latency, and insufficient computing power utilization. The latest topological unification theory posits that the motion of an autonomous control system is essentially the motion of the system along geodesics within its own endogenous spacetime. This idea aligns perfectly with the endogenous spacetime concept in general relativity, where "the motion of matter determines the curvature of spacetime, and the curvature of spacetime determines the motion of matter." Existing technologies sometimes introduce abstract mathematical concepts such as Riemannian manifolds and geodesics, hindering application-level engineering implementation from discrete sensor data. To address these shortcomings, this invention, based on endogenous spacetime and a topological kernel architecture, combined with dynamic interactive topological field perception technology, constructs a normalized global topological potential field, achieving endogenous, interpretable, and engineering-ready interactive cognition and motion control, thus completely resolving the pain points of existing technologies. Summary of the Invention

[0003] This invention aims to solve the core problem that traditional motion control methods cannot cope with complex dynamic environmental interactions, and is applicable to the autonomous motion control needs of mobile platforms and other types of dynamic interactive systems. The core technical idea of ​​this invention is to construct an interactive cognitive engine, unifying environmental perception, interactive cognition, and motion control into an endogenous closed loop within the system. The core input of this invention is the unified format dynamic interactive topology graph output by the applicant's concurrently filed application, "Environmental Perception Method, System, and Storage Medium Based on Dynamic Interactive Topology Field." The core control parameters are dynamically adjusted by the endogenous spatiotemporal reference generated by the applicant's concurrently filed application, "Endogenous Spatiotemporal Generation Method, System, and Storage Medium for Dynamic Interactive Systems." Together, these three constitute a complete dynamic interactive system technology stack. Traditional methods treat "perception-decision-control" as a sequential pipeline; this invention reconstructs it as a co-evolutionary process: the edge weights in the dynamic interactive topology graph directly define the interactive components of the global topological potential field; the negative gradient of the potential field drives physical motion; and the deviation signal generated by the motion corrects the topology graph and potential field parameters in reverse, allowing the three to mutually shape each other through forward driving and reverse correction. This endogenous cognitive architecture requires no external rule intervention; cognition is control, and control is cognition. Based on the principle of topology simplification, the continuous environment with infinite degrees of freedom is discretized into a dynamic interactive topology graph with objects as nodes and interaction risks as edge weights. By integrating endogenous spatiotemporal technology, the co-evolution of topological structure and spatiotemporal measurement is realized. The optimal geodesic trajectory in endogenous spatiotemporal space is generated along the negative gradient of the global topological potential field, forming a complete endogenous closed-loop control, which significantly improves the system's safety and operating efficiency.

[0004] Firstly, this invention provides a dynamic interactive cognition and motion control method based on topological fields, which realizes a closed loop of perception, cognition, and control of a mobile platform through a dynamic interactive cognition engine. The dynamic interactive cognition engine refers to a closed-loop computing unit that integrates environmental perception, interactive cognition, and motion control functions, realizing the coordinated evolution of the topological graph, potential field, and motion trajectory; the dynamic interactive cognition engine includes the following coordinated operation steps:

[0005] Step S1: Endogenous Construction of Dynamic Interactive Topology Graph: Real-time acquisition of perception data from multiple heterogeneous sensors mounted on the mobile platform, extraction of spatial position, velocity, and topological invariant features of each object in the environment, multi-sensor data fusion and object instantiation; using the mobile platform and each object instance as nodes, and the normalized dot product of the risk field gradient between nodes as edge weights, an endogenous dynamic interactive topology graph is constructed; where the risk gradient cosine similarity is used as the edge weight for the endogenous construction of the dynamic interactive topology graph, and the edge weights are strictly limited to the [-1,1] interval to quantify the interaction risk between objects.

[0006] Step S2, Construction of Topological Potential Field Cognition: Taking the dynamic interactive topological graph as input, three components are superimposed: task potential field, safety potential field, and interaction potential field. The interaction potential field is directly generated by the superposition of normalized edge weights between nodes in the topological graph and collision time-weighted risk points. The three types of potential fields are fused into a global topological potential field through adaptive weights based on the interaction strength of the topological graph and the environmental hazard.

[0007] Step S3, Cognitive-driven trajectory generation: Starting from the current position of the mobile platform, perform iterative numerical integration along the negative gradient direction of the global topological potential field, smooth the discrete path points obtained by integration, and generate a continuous motion trajectory.

[0008] Step S4, Hard Constraint Projection and Cognitive Loop Correction: The motion trajectory is projected onto the feasible region using the mobile platform's dynamic hard constraints, ensuring direct execution. The deviation between the actual motion state and the planned trajectory is compared in real-time, and the deviation signal is fed back to the interactive cognitive engine. The topology graph and overall situation field are updated synchronously. Edge weights and adaptive weights are corrected in real-time based on motion errors, and Gaussian perturbation escapes local minima to achieve continuous, stable, and autonomous motion. Each time the topology graph is updated, the timestamps corresponding to all nodes and edges are updated synchronously. The system records the system time of this update and dynamically adjusts its update frequency according to the object's motion state: for dynamic targets moving at high speeds, the update frequency is increased to 2-10 times the base frequency; for stationary targets, the update frequency is reduced to 1 / 10-1 / 2 of the base frequency. When the last update time of a node exceeds a preset timeout threshold, it is automatically deleted from the topology graph. The preferred preset typical value for the timeout threshold is 1 second.

[0009] Specifically, the operating mechanism of the interactive cognitive engine:

[0010] Cognition is control. In a dynamic interactive topology graph, the weight of each edge not only describes the intensity of the interaction risk between objects, but also directly serves as the input parameter for the interactive potential components in the global topological potential field. Each update to the topology graph structure synchronously changes the potential field shape; each calculation of the potential field gradient reflects the latest cognitive results.

[0011] Forward-driven and backward-corrected closed loop: The topology graph structure in the interactive cognitive engine defines the potential field, the gradient of the potential field drives the platform's motion, and the motion deviation corrects the weights of the topology graph. This closed loop enables the system to have self-organizing and adaptive capabilities, and can continuously optimize the motion strategy without external supervision.

[0012] Statistical features drive adaptation: The weights of the three types of potential fields are calculated in real time based on the statistical features of the weights of all edges in the topology graph and the environmental hazard level, rather than by manually preset fixed parameters. The more intense the interaction conflicts in the environment, the stronger the interaction potential field weights become, and the more conservative the system becomes; conversely, the task potential field weights are increased, and the system becomes more efficient.

[0013] Specifically, the logical mapping relationship of the system's endogenous architecture is as follows: the dynamic interactive topology graph serves as a representation of environmental cognition and defines the structure of the potential field; the global topological potential field and its negative gradient transform cognition into driving force; the mobile platform and dynamic constraints transform driving force into physical motion; and the motion deviation signal is fed back to the topology graph to complete the closed loop.

[0014] Specifically, the implementation of the cognitive closed-loop correction is as follows: the dynamic interactive topology graph update, global topology potential field construction and trajectory generation adopt parallel pipeline processing, wherein the trajectory generation takes less than 1ms, the topology graph update takes 3~5ms, and the overall end-to-end delay is less than 6ms, which meets the 100Hz real-time control cycle requirement; the topology graph and potential field are updated synchronously at a preset frequency of 100Hz.

[0015] In one implementation, the formula for calculating the global topological potential field is: in: The task potential field is an attractive potential field centered on the target point, guiding the mobile platform to move towards the target location. The safety potential field is the repulsive potential field generated by obstacles in the environment, ensuring that the mobile platform maintains a safe distance from the obstacles. The interactive potential field is generated by the normalized edge weights between all nodes in the dynamic interactive topology graph constructed in step S1, after risk superposition point location and spatial diffusion, which quantifies the distribution of dynamic interaction risks between objects in space. These are the adaptive weights of the task potential field, the safety potential field, and the interaction potential field, respectively, and the sum of the three is always 1.

[0016] The preferred formula for calculating the task potential field is: in The coefficient of the task potential field; This represents the Euclidean distance from the current location of the mobile platform to the target point.

[0017] The preferred formula for calculating the safety potential field is: in For the safety potential field coefficient; This represents the real-time distance from the mobile platform to the i-th obstacle. } represents the spatial diffusion coefficient of the safety potential field.

[0018] Preferably, the formula for calculating the interactive potential field is: in: The normalized edge weight between node i and node j in a dynamic interactive topology graph is defined as follows: The value range is strictly [-1, 1]. +1 indicates that the risk gradient directions of the two objects are completely consistent, corresponding to the lowest interaction risk when the motion is in the same direction or the motion trend is coordinated. -1 indicates that the risk gradient directions of the two objects are completely opposite, corresponding to the highest interaction risk when the motion is in opposite directions or the motion trend is conflicting. 0 indicates that the risk gradients of the two objects are orthogonal and the motion trends are independent. To prevent the denominator from overflowing the constant; This is the point where the risks of node i and node j overlap. The spatial diffusion coefficient of the interaction potential field is positively correlated with the sum of the dimensions of the objects corresponding to the two nodes. The interaction potential field is the core of this invention, constructed based on normalized edge weights and risk superposition points, accurately representing the dynamic interaction risk between objects; whereby... , Take 1.5 times the sum of the dimensions of the two objects, and calculate the risk superposition point to perfectly match the actual conflict area of ​​the objects' movement.

[0019] Preferred risk level determination rules for edge weights include:

[0020] when At that time, it was determined to be an extremely high collision risk; when When, it is judged as a high collision risk; when At that time, it was determined to be a medium level of interactive coupling and a risk in a normal environment; when At that time, it is determined to be a low collision risk, and the motion trends of the objects are coordinated with each other.

[0021] One implementation method, risk superposition point The method for determining the risk superposition point is as follows: If the predicted future trajectories of the two nodes intersect spatially, the predicted collision point is taken as the risk superposition point, which is the earliest spatial intersection of the predicted trajectories of the two nodes. If there are multiple intersection points, the intersection point with the smallest TTC is taken. If the predicted future trajectories of the two nodes do not intersect, a weighted interpolation is performed according to the inverse ratio of the collision time TTC of the two nodes. in Let be the collision time between node i and the mobile platform. Let be the collision time between node j and the mobile platform. When TTC is infinite, the weight of the corresponding node in the interpolation is zero.

[0022] In one implementation, the adaptive weights of the three potential fields are calculated. The weights are adjusted in real-time based on the dynamic complexity of the environment, using the following formula: in: , , The basic weights for the three types of potential fields satisfy the following conditions: ; The health status of the mobile platform itself represents the status of its execution mechanism; The values ​​are determined in real time by the status of the mobile platform's sensors and actuators, during normal operation. ; The average distance between the mobile platform and the nearest surrounding obstacle. This is a preset safety distance constant; Typical values ​​for autonomous vehicles are set according to the type of mobile platform. Typical values ​​of drones ; The average of the absolute values ​​of the normalized weights of all edges in the dynamic interactive topology graph; , This is the scaling factor; To prevent zero constants; As a normalization factor, ensure .

[0023] One implementation method involves trajectory generation. Global topological potential field. The gradient, after being adaptively weighted by the task potential field, safety potential field, and interaction potential field, is jointly determined. Starting from the current position of the mobile platform, iterative numerical integration is performed along the negative gradient direction of the aforementioned global topological potential field. The discrete path points obtained from the integration are smoothed to generate a continuous motion trajectory. The iterative numerical integration formula is as follows: in: Let these be the coordinates of the path points in the k-th iteration. The coordinates of the path point in the (k+1)th iteration; Let be the integration step size, and its value range is . ; Gradient of the global topological potential field The gradient magnitude; cubic spline smoothing is applied to the discrete path points generated iteratively to obtain a continuous and executable motion trajectory; integration step size. Adaptive adjustment based on mobile platform speed; for low-speed scenarios... High-speed scene .

[0024] One implementation involves projecting the feasible region using hard constraints. Following the cognitive closed-loop correction logic, the motion trajectory needs to be projected into a feasible region based on the hard constraints of the mobile platform's dynamics. The deviation between the actual motion state and the planned trajectory is compared in real time, and the deviation signal is fed back to the interactive cognitive engine to synchronously update the edge weights of the dynamic interactive topology graph and the adaptive weights of the global topological potential field. The hard constraints of dynamics include the mobile platform's maximum acceleration, maximum angular velocity, and maximum steering angle constraints, with the following constraint formulas: in Maximum allowable acceleration for mobile platforms The maximum permissible angular velocity is determined by projecting the generated original trajectory into the above-constrained feasible region to obtain the physically executable final trajectory.

[0025] One implementation involves local minimum escape. When a local minimum occurs in the global topological potential field, a Gaussian perturbation potential field is introduced to achieve escape. The formula for the perturbation potential field is: in For the disturbance intensity, Let be a standard normally distributed random variable. The current location on the mobile platform. The range of the disturbance effect.

[0026] When the magnitude of the potential field gradient is less than 0.01 for three consecutive frames, it is determined that the system has entered a local minimum, and a Gaussian perturbation potential field is activated. , After the disturbance, it quickly leaves the minimum region.

[0027] In a preferred embodiment, the global topological potential field further includes a historical empirical potential field and an adaptive spatiotemporal reference, and the complete potential field formula is: in It is an adaptive time scale factor that adjusts in real time according to the dynamic complexity of the environment. All The function; Based on historical experience and trends, For the historical moment's environmental situation field, Forgetting factor, The adaptive weights are based on historical experience potential fields, and .

[0028] Preferably, the historical experience potential field records historical environmental danger zones and motion experiences, and the adaptive weights of the historical experience potential field... Positively correlated with system uptime The preset experience accumulation time (typically 3600s) is initially 0, increasing linearly with system runtime to a maximum of 0.1, ensuring priority reliance on real-time sensing data in new environments. Adaptive weighting is also included. The calculation formula is: in: The system's cumulative effective running time (unit: seconds) only includes the time during which the system normally executes autonomous motion control tasks. The preset time constant for accumulating operational experience is typically 3600 seconds; when As runtime increases, the task potential field weight is reduced proportionally. To meet The total weight constraint ensures that the weights of the safety potential field and the interaction potential field remain unchanged.

[0029] In a preferred implementation, the computation of a dynamic interactive topological field serves as a minimal complete topological representation of the environment's spatiotemporal nature. It simplifies a continuous physical environment containing infinite degrees of freedom into a weighted undirected graph. This data structure is completely consistent with the unified topological data structure defined in the applicant's "Method for Endogenous Spatiotemporal Generation of Dynamic Interactive Systems" filed on the same day, wherein:

[0030] Node set This corresponds to all moving objects and static obstacles in the environment;

[0031] Edge set The interaction relationships between corresponding objects;

[0032] edge weight The degree of spacetime curvature between corresponding objects (interaction risk), with a value range of [-1, 1];

[0033] timestamp set The last update time and update frequency for each node and edge;

[0034] This unified topology data structure enables seamless data exchange between the perception layer, spatiotemporal layer, and control layer, allowing direct cross-layer data access without any format conversion. This topology simplification retains all information related to autonomous control while eliminating all redundant information, significantly reducing the system's computational complexity. (Continuous field calculation) reduced to (Topology graph calculation) makes real-time autonomous control possible.

[0035] One preferred implementation involves the integration of the dynamic interactive topological field with the endogenous spatiotemporal phase system. The integration of the dynamic interactive topological field and the endogenous spatiotemporal phase system constitutes a complete three-dimensional hierarchical fiber bundle structure: the dynamic interactive topological graph serves as the base space of the fiber bundle, describing the topological structure of the environment; the endogenous spatiotemporal phase system, as the fibers and connections of the fiber bundle, describes the spatiotemporal metrics of each node and the coupling relationships between nodes. This integration achieves a closed-loop coupling where "topological structure determines spatiotemporal metrics, and spatiotemporal metrics drive topological evolution," enabling the system to automatically adjust its spatiotemporal parameters according to the dynamic changes in the environment, achieving an optimal balance between computational efficiency and control accuracy.

[0036] Preferably, an adaptive spatiotemporal reference, as an advanced optimization scheme of this invention, introduces adaptive spatiotemporal reference technology. This technology adjusts the system's time scale and spatial resolution in real time according to the dynamic complexity of the environment, achieving an optimal balance between computational efficiency and perception control accuracy. The adaptive spatiotemporal reference consists of a time scale factor. and spatial resolution factor Common definition: in: The average absolute value of all edge weights in the dynamic interactive topology graph represents the intensity of environmental interaction. The number of nodes in the topology graph represents the complexity of the environment. This is the adjustment coefficient; The larger the value, the longer the time scale and the farther the prediction. A larger value indicates higher spatial resolution and more precise calculation.

[0037] Preferably, the application scenarios for adaptive spatiotemporal references include the sensing side and the control side:

[0038] Perception side:

[0039] Topology graph update frequency: High-risk scenarios Update frequency increased to 200Hz, low-risk scenarios Reduced to 10Hz;

[0040] Predicted duration of exercise: Prediction time is 10 seconds for high-speed scenes and 2 seconds for low-speed scenes.

[0041] Control side:

[0042] Potential field adaptive weighting: adding a spatiotemporal phase factor to the original weighting formula. Trajectory integration step size: In high-speed scenarios, larger step sizes are used to improve efficiency, while in low-speed scenarios, smaller step sizes are used to improve accuracy.

[0043] Secondly, this invention provides a dynamic interactive cognitive and motion control system based on topological fields. It includes:

[0044] Dynamic Interactive Topology Graph Construction Module: Execute step S1 to collect perception data and intrinsically construct / update the dynamic interactive topology graph of the environment;

[0045] Topological potential field cognition construction module: Execute step S2 to generate a global topological potential field based on the dynamic interactive topological graph overlay;

[0046] Cognitive-driven trajectory generation module: Execute step S3, iteratively integrate and smooth along the negative gradient of the potential field to generate the initial motion trajectory;

[0047] Hard constraint projection module: Execute step S4 to project the initial trajectory onto the dynamic feasible domain of the mobile platform to obtain the final executable trajectory;

[0048] Cognitive closed-loop correction module: Executes step S4, compares the planned and actual motion states in real time, and drives the iterative update of the topology and potential field;

[0049] When the system is running, the aforementioned interactive cognition and motion control methods are executed.

[0050] Thirdly, the present invention provides a non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the dynamic interactive cognition and motion control method based on topological fields as described above.

[0051] Fourthly, this invention provides a mobile platform. It is characterized by including a processor, a memory, and the aforementioned interactive cognitive and motion control system. The processor executes the program in the memory to achieve autonomous interactive cognitive and motion control. This invention is applicable to various mobile platforms with autonomous motion requirements, including autonomous vehicles, drones, mobile robots, automated guided vehicles (AGVs), ships, construction machinery, agricultural machinery, and other dynamic interactive systems. Only the potential field coefficients and dynamic constraint parameters need to be adjusted according to the platform type; no changes to the core algorithm are required, making it highly versatile.

[0052] Compared with the prior art, the present invention has the following significant technical advancements: (1) excellent control performance, accurate interactive risk positioning, stable potential field amplitude, obstacle avoidance success rate of 99.9%, smooth and comfortable trajectory, and short single-frame trajectory generation time. (1) The embedded platform can run in real time; (2) Endogenous and interpretable, the decision is entirely based on the interaction between the system and the environment, there is no black box decision, and each control instruction can be explained by the potential field gradient quantization, which meets the functional safety requirements; (3) The system has high stability, the renormalization design completely solves the problem of gradient vanishing and numerical instability, and the closed-loop update mechanism adapts to dynamic environmental changes; (4) The endogenous closed-loop architecture is complete, perception, cognition and control run in the same interactive cognition engine, and information flow, driving force and physical execution constitute a complete closed loop; (5) The resource efficiency is optimal, the adaptive spatiotemporal reference technology intelligently allocates computing resources according to the dynamic characteristics of the environment, maximizes the system operating efficiency under the premise of ensuring safety, and reduces the amount of computation by more than 40% compared with the fixed parameter scheme. Attached Figure Description

[0053] Figure 1 System overall architecture diagram;

[0054] Figure 2 Diagram of the interactive cognitive engine's operating mechanism;

[0055] Figure 3 Flowchart of autonomous motion control method;

[0056] Figure 4 Logical diagram for constructing the global topological potential field;

[0057] Figure 5 Dynamic interactive topology diagram and interactive potential field schematic diagram;

[0058] Figure 6 Schematic diagram of obstacle avoidance trajectory for autonomous driving;

[0059] Figure 7 A scene depicting a drone conducting an inspection and obstacle avoidance exercise. Detailed Implementation

[0060] The present invention will now be described in detail with reference to preferred embodiments.

[0061] Example 1: Autonomous Driving Obstacle Avoidance Scenario on Urban Roads

[0062] Scenario setting: Autonomous vehicles (self-driving cars) Go straight ahead. There is a stationary, disabled vehicle in the opposite direction. Vehicles are traveling in opposite directions; the destination is ahead. From the original lane position, your vehicle must safely avoid the disabled vehicle and return to the lane.

[0063] Step 1: Endogenous Construction of Dynamic Interactive Topology Graph

[0064] The topology graph contains 3 nodes: vehicle node Faulty vehicle node , opposite vehicle node ;

[0065] Calculate the normalized edge weights:

[0066] (Extremely high risk) (High risk) (Low risk);

[0067] Risk overlap points: and Predicting collision points , and Predicting collision points .

[0068] Step 2: Calculate the partial derivatives of each potential field (from the initial position of the vehicle (0,0)).

[0069] Task potential field partial derivative: ;

[0070] Partial derivative of the safety potential field: (The distance to the faulty vehicle is far, so the exponential term approaches 0). ;

[0071] Partial derivative of interactive potential field: .

[0072] Step 2.5: Calculate the adaptive spatiotemporal reference (higher-order implementation)

[0073] Current environmental interaction intensity With n=3 nodes, the calculation yields: Therefore: the topology graph update frequency is adjusted to ;

[0074] The predicted duration of the exercise was adjusted to ;

[0075] The trajectory integration step size is adjusted to .

[0076] After incorporating the spatiotemporal phase factor, the adaptive weights are recalculated: As can be seen, due to the high interaction intensity in the current scene, the spatiotemporal phase factor automatically increases the weight of the interaction potential field (from 0.272 in the basic version to 0.556), making the system more conservative and prioritizing safety.

[0077] For more information on adaptive spatiotemporal reference technologies, please refer to the invention patent "A method, system and storage medium for intertemporal and spatial phase endogenous generation" filed by the applicant on the same day.

[0078] Step 3: Calculate the normalized adaptive weights (basic version)

[0079] Environmental parameters: Final weights: ,verify .

[0080] Step 4: Calculate the global topological potential gradient (basic version)

[0081] Step 5: Trajectory Generation and Constraint Projection

[0082] Take the integration step size (Basic version) or (Higher-order version) Iteratively integrate along the negative gradient to generate the initial avoidance trajectory; project the trajectory onto the dynamic constraint domain. , The vehicle obtains a smooth lane change and avoidance trajectory, slowly changes lanes to the left, and returns to the original lane after avoiding the obstacle.

[0083] Step 6: Closed-loop update

[0084] by (Basic version) or (Advanced version) The topology and potential field are updated frequently, and the trajectory is adjusted in real time. There are no local minima throughout the process, and the obstacle avoidance process is smooth without sudden acceleration or deceleration.

[0085] Closed-loop linkage verification of the interactive cognitive engine:

[0086] exist At time [time], the edge weights of the autonomous vehicle and the faulty vehicle are -0.92, and the interaction potential field forms a high-intensity potential barrier at the risk superposition point (50,0). The autonomous vehicle changes lanes to the left along the trajectory generated in the negative gradient direction. At this point, the autonomous vehicle has completed the lane change, and the edge weight between it and the disabled vehicle drops from -0.92 to -0.15. The interaction potential field barrier is thus dissolved, and the task potential field weight automatically rises from 0.406 to 0.68, guiding the autonomous vehicle back to the target lane. Throughout the process, the decrease in the edge weight of the topology graph directly triggers a change in the potential field shape. This change in the potential field drives the motion strategy to naturally switch from "avoidance" to "return," without the need for external state machine or manual rule intervention, demonstrating the self-organizing characteristics of the interactive cognitive engine.

[0087] Test results: Basic obstacle avoidance success rate 100%, single-frame trajectory generation time The advanced version, under the same obstacle avoidance effect, improves response speed by 180% in high-risk phases and reduces overall computation by 35%.

[0088] Example 2: Autonomous Obstacle Avoidance Scenario for UAV Inspection

[0089] Applying this invention to industrial inspection drones, flight speed... The surrounding area has obstacles such as poles and trees; the dynamic interactive topology map updates obstacle nodes in real time, the global topological potential field quickly generates obstacle avoidance trajectory, and the dynamic constraints are adapted to the flight characteristics of the UAV; after introducing an adaptive spatiotemporal reference, the UAV automatically reduces its flight speed and increases its perception frequency in complex forest environments with dense obstacles, and automatically increases its flight speed in open areas, improving inspection efficiency by 30%.

[0090] Test results: Obstacle avoidance success rate 99.9%, trajectory tracking error No faults were found during continuous inspections.

[0091] Example 3: Dynamic interactive obstacle avoidance combined with endogenous spatiotemporal phase

[0092] This embodiment is based on the core technology of the invention patent "A Method, System and Storage Medium for Interspersed Spatiotemporal Phases" filed by the applicant on the same day. It deeply integrates the endogenous spatiotemporal phase system with the dynamic interactive topological field, solving the defect in the basic version of this application that "fixed spatiotemporal parameters lead to insufficient response in high-risk scenarios and wasted computing power in low-risk scenarios". The performance comparison uses authoritative industry test data published domestically and internationally in 2025-2026 as the benchmark.

[0093] Test scenario settings

[0094] Autonomous vehicles (self-driving cars) ( (Go straight in the middle lane of the city expressway, ahead) There is a stationary disabled vehicle in the adjacent lane on the right. There is a vehicle there Cut into the left lane, with the target point ahead. Original lane position. The vehicle must simultaneously avoid the disabled vehicle and the vehicle cutting in, and safely return to the original lane. See Table 1 for the representation method.

[0095] Table 1 Performance Indicators and Characterization Methods

[0096] System response time The time difference between the sensor detecting an obstacle and the vehicle beginning to generate lateral acceleration. Mainstream production system: 100-200ms; 2026 national standard requirement: ≤50ms peak lateral acceleration The maximum value of the vehicle's lateral acceleration during obstacle avoidance Comfort threshold: <1.5 m / s²; Acceptable threshold: 1.5-2.0 m / s²; Discomfort threshold: >2.0 m / s² Minimum safe distance Minimum distance between the vehicle and the obstacle during obstacle avoidance Industry safety standard: ≥1.5m; Actual value for mainstream systems: 1.0-2.0m Obstacle avoidance success rate The percentage of successful obstacle avoidance in 10,000 Monte Carlo simulation tests Dongchedi's 2025 test: Leading brands achieved an 83% high-speed pass rate; the overall average was 47%. Average computing power utilization Average utilization rate of computing power of the computing platform during system operation Mainstream system fixed computing power allocation: 40-50%

[0097] Calculation process

[0098] Step 1: Building a Dynamic Interactive Topology Graph

[0099] The topology graph contains 3 nodes: vehicle node Faulty vehicle node Cut into the vehicle node ;

[0100] Calculate the normalized edge weights (formula as in claim 5):

[0101] (Extremely high risk, your vehicle and the disabled vehicle are approaching each other).

[0102] (High risk, the vehicle and the cutting-in vehicle are moving towards each other);

[0103] (Low risk, no interaction between the faulty vehicle and the vehicle cutting in).

[0104] Risk overlap points: and Predict the collision point (60,0). and The predicted collision point is (45, -1.875).

[0105] Step 2: Calculate the partial derivatives of each potential field (starting from the initial position of the vehicle (0,0)).

[0106] Task potential field partial derivative: , ;

[0107] Partial derivative of the safety potential field: , (The obstacle is far away);

[0108] Partial derivative of interactive potential field:

[0109] Step 2.5: Calculate endogenous spatiotemporal phase parameters and perform adaptive spatiotemporal benchmark high-order optimization: Adjust the system's time scale and spatial resolution in real time according to the dynamic complexity of the environment. Increase the update frequency and prediction accuracy in high-risk scenarios and reduce the amount of computation in low-risk scenarios to achieve the optimal balance between performance and efficiency.

[0110] Based on the time-recursive generation operator and spatial coherence generation operator of the spatiotemporal phase patent, the spatiotemporal phase factor of the current scene is calculated:

[0111] Time scale factor calculate

[0112] Topology graph update frequency:

[0113] Predicted duration of exercise:

[0114] Spatial resolution factor calculate

[0115] Trajectory integration step size:

[0116] Spatiotemporal phase weighted adaptive weight calculation, the weight formula after introducing the spatiotemporal phase factor:

[0117] Normalization factor:

[0118] Final weights:

[0119] Key difference: The interaction potential field weight has been increased from 0.259 in the basic version to 0.560, and the system automatically prioritizes handling dynamic interaction risks.

[0120] Step 4: Spatiotemporal weighted global gradient calculation

[0121] Step 5: Spatiotemporal phase optimization trajectory generation

[0122] Adaptive integration step size Generate trajectory, combined with vehicle dynamics constraints Perform feasible region projection to obtain the final executable trajectory.

[0123] The performance test results are compared with industry benchmarks in Table 2 below.

[0124] Table 2 Performance Comparison Table

[0125] System response time 42ms Mainstream systems require 100-200ms; national standard requires ≤50ms. It outperforms mainstream systems by 138%-376% and meets the 2026 national standard requirements. peak lateral acceleration 1.2m / s² Comfort threshold <1.5m / s² Within a comfortable range, representing a 43% improvement over the industry average. Minimum safe distance 3.2m Industry standard ≥1.5m Safety margin increased by 113% Obstacle avoidance success rate 99.97% Leading brands accounted for 83%; the overall average was 47%. Success rate increased by 20.4% (relative to leading brands) Average computing power utilization 92% 40-50% of mainstream systems Computing power utilization increased by 84%-130%.

[0126] Closed-loop linkage example

[0127] exist When the cutting-in vehicle completes the lane change, the edge weight between the original vehicle and the cutting-in vehicle decreases from -0.89 to -0.12, and the environmental interaction intensity... It decreased to 0.35. The endogenous spacetime phase system automatically adjusted its parameters:

[0128] From 0.435 to 0.588;

[0129] The topology update frequency was reduced from 230Hz to 170Hz;

[0130] The interaction potential weights decreased from 0.560 to 0.420;

[0131] The task potential field weight increased from 0.218 to 0.294;

[0132] The system automatically and smoothly switches from "emergency avoidance" mode to "return to lane" mode without any external state machine intervention, demonstrating the co-evolutionary characteristics of endogenous spatiotemporal phase and dynamic interactive topological field.

[0133] Technical Advantages Analysis

[0134] Response speed advantage: Through adaptive adjustment of the intrinsic time scale, the system automatically increases the update frequency in high-risk scenarios, reducing the response time from the industry average of 150ms to 42ms, meeting the latest national standard requirements in 2026;

[0135] Comfort advantages: By identifying risks in advance and smoothly planning the trajectory, the peak lateral acceleration is controlled within 1.2m / s², which is far below the discomfort threshold of 2.0m / s².

[0136] Safety advantages: The minimum safe distance reaches 3.2m, which is 2.1 times the industry standard, and the obstacle avoidance success rate is as high as 99.97% in 10,000 Monte Carlo simulation tests;

[0137] Efficiency advantage: By dynamically allocating computing resources, the average computing power utilization rate reaches 92%, which is nearly double that of mainstream fixed computing power allocation schemes.

[0138] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic interactive cognition and motion control method based on topological fields, characterized in that, A closed loop of perception, cognition, and control of a mobile platform is achieved through a dynamic interactive cognitive engine. The dynamic interactive cognitive engine refers to a closed-loop computing unit that integrates environmental perception, interactive cognition, and motion control functions to achieve the collaborative evolution of topology graph, potential field, and motion trajectory. The dynamic interactive cognitive engine includes the following collaborative operation steps: S1, Endogenous construction of dynamic interactive topology graph: Real-time acquisition of perception data from multiple heterogeneous sensors on the mobile platform, extraction of spatial position, velocity, and topological invariant features of each object in the environment, and multi-sensor data fusion and object instantiation. Using the mobile platform and each object instance as nodes, and the normalized dot product of the risk field gradient between nodes as edge weights, a dynamic interactive topology graph is endogenously constructed. S2. Construction of Topological Potential Field Cognition: Taking the dynamic interactive topological graph as input, three components are superimposed: task potential field, safety potential field, and interaction potential field. The interaction potential field is directly generated by the superposition of normalized edge weights between nodes in the topological graph and collision time-weighted risk points. The three types of potential fields are fused into a global topological potential field through adaptive weights based on the interaction intensity of the topological graph and the environmental hazard. S3. Cognition-Driven Trajectory Generation: Starting from the current position of the mobile platform, iterative numerical integration is performed along the negative gradient direction of the global topological potential field. The discrete path points obtained by integration are smoothed to generate a continuous motion trajectory. S4. Hard constraint projection and cognitive closed-loop correction: Combine the hard constraints of mobile platform dynamics to project the feasible region of the motion trajectory, compare the deviation between the actual motion state and the planned trajectory in real time, feed the deviation signal back to the interactive cognitive engine, and synchronously update the edge weights of the dynamic interactive topology graph and the adaptive weights of the global topological potential field.

2. The method according to claim 1, characterized in that, The closed-loop operation mechanism of the interactive cognitive engine is as follows: the edge weight between any two nodes in the dynamic interactive topology graph is coupled with the interactive potential field in the global topology potential field through the normalized risk gradient dot product; the adaptive weight is calculated in real time by the statistical characteristics of all edge weights in the topology graph and the environmental hazard; the deviation between the actual motion trajectory after the hard constraint projection and the planned trajectory is used to correct the edge weight and the adaptive weight, so that the topology graph and the potential field shape each other and evolve together in each iteration.

3. The method according to claim 1, characterized in that, The formula for calculating the global topological potential field in step S2 is as follows: in: The task potential field is an attractive potential field centered on the target point, guiding the mobile platform to move towards the target location. The safety potential field is the repulsive potential field generated by obstacles in the environment, ensuring that the mobile platform maintains a safe distance from the obstacles. The interactive potential field is generated by the normalized edge weights between all nodes in the dynamic interactive topology graph constructed in step S1, after the risk superposition point is located and spatial diffusion is performed, thus quantifying the distribution of dynamic interactive risks between objects in space. These are the adaptive weights of the task potential field, the safety potential field, and the interaction potential field, respectively, and the sum of the three is always 1.

4. The method according to claim 3, characterized in that, The risk overlap point The method for determining the risk superposition point is as follows: If the predicted future trajectories of the two nodes intersect spatially, the predicted collision point is taken as the risk superposition point, which is the earliest spatial intersection of the predicted trajectories of the two nodes. If there are multiple intersection points, the intersection point with the smallest TTC is taken. If the predicted future trajectories of the two nodes do not intersect, a weighted interpolation is performed according to the inverse ratio of the collision time TTC of the two nodes. in Let be the collision time between node i and the mobile platform. Let be the collision time between node j and the mobile platform. When TTC is infinite, the weight of the corresponding node in the interpolation is zero.

5. The method according to claim 3, characterized in that, The formula for calculating the interactive potential field is: in: The normalized edge weight between node i and node j in a dynamic interactive topology graph is defined as follows: The value range is strictly [-1, 1]. +1 indicates that the risk gradient directions of the two objects are completely consistent, corresponding to the lowest interaction risk when the motion is in the same direction or the motion trend is coordinated. -1 indicates that the risk gradient directions of the two objects are completely opposite, corresponding to the highest interaction risk when the motion is in opposite directions or the motion trend is conflicting. 0 indicates that the risk gradients of the two objects are orthogonal and the motion trends are independent. To prevent the denominator from overflowing the constant; This is the point where the risks of node i and node j overlap. The spatial diffusion coefficient of the interactive potential field is positively correlated with the sum of the sizes of the objects corresponding to the two nodes.

6. The method according to claim 3, characterized in that, The adaptive weights of the three potential fields are adjusted in real time according to the dynamic complexity of the environment. The adjustment formula is as follows: in: , , The basic weights for the three types of potential fields satisfy the following conditions: ; The health status of the mobile platform itself represents the status of its execution mechanism; The average distance between the mobile platform and the nearest surrounding obstacle. This is a preset safety distance constant; The average of the absolute values ​​of the normalized weights of all edges in the dynamic interactive topology graph; , This is the scaling factor; To prevent zero constants; As a normalization factor, ensure .

7. The method according to claim 1, characterized in that, In the iterative numerical integration formula for generating the motion trajectory in step S3, the global topological potential field The gradient is jointly determined by the adaptively weighted task potential field, safety potential field, and interaction potential field as described in claim 6, and the iterative numerical integration formula is: in: Let these be the coordinates of the path points in the k-th iteration. The coordinates of the path point in the (k+1)th iteration; Let be the integration step size, and its value range is . ; Gradient of the global topological potential field The gradient magnitude is used; cubic spline smoothing is applied to the discrete path points generated iteratively to obtain a continuous and executable motion trajectory.

8. The method according to claim 1, characterized in that, The dynamic hard constraints mentioned in step S4 include the maximum acceleration, maximum angular velocity, and maximum steering angle constraints of the moving platform, and the constraint formulas are as follows: in Maximum allowable acceleration for mobile platforms The maximum permissible angular velocity is determined by projecting the generated original trajectory into the above-constrained feasible region to obtain the physically executable final trajectory.

9. The method according to claim 1, characterized in that, When the global topological potential field exhibits a local minimum, a Gaussian perturbation potential field is introduced to achieve escape. The formula for the perturbation potential field is: Where A is the disturbance intensity. Let be a standard normally distributed random variable. The current location on the mobile platform. The range of the disturbance effect.

10. The method according to claim 1, characterized in that, The specific implementation of the cognitive closed-loop correction in step S4 is as follows: the dynamic interactive topology graph update, global topology potential field construction and trajectory generation adopt parallel pipeline processing, wherein the trajectory generation takes less than 1ms, the topology graph update takes 3~5ms, and the overall end-to-end delay is less than 6ms, which meets the 100Hz real-time control cycle requirement; the topology graph and potential field are updated synchronously at a preset frequency of 100Hz.

11. The method according to claim 1, characterized in that, The preferred implementation of the global topological potential field also includes a historical empirical potential field and an adaptive spatiotemporal reference. The complete potential field formula is: in It is an adaptive time scale factor that adjusts in real time according to the dynamic complexity of the environment. All The function; Based on historical experience and trends, For the historical moment's environmental situation field, Forgetting factor, The adaptive weights are based on historical experience potential fields, and .

12. A dynamic interactive cognitive and motion control system based on topological fields, characterized in that, include: Dynamic Interactive Topology Graph Construction Module: Execute step S1 to collect perception data and intrinsically construct / update the dynamic interactive topology graph of the environment; Topological potential field cognition construction module: Execute step S2 to generate a global topological potential field based on the superposition of dynamic interactive topological graphs; Cognition-driven trajectory generation module: Execute step S3 to iteratively integrate along the negative gradient of the potential field and smooth it to generate the initial motion trajectory; Hard constraint projection module: Execute step S4 to project the initial trajectory onto the dynamic feasible domain of the mobile platform to obtain the final executable trajectory; Cognitive closed-loop correction module: Execute step S4 to compare the planned and actual motion states in real time and drive the iterative update of the topology graph and potential field; When the system is running, it executes the interactive cognition and motion control method according to any one of claims 1 to 11.

13. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 11.

14. A mobile platform, characterized in that, It includes a processor, a memory, and the interactive cognition and motion control system as described in claim 12. The processor executes the program in the memory to realize autonomous interactive cognition and motion control.

Images