Intelligent logistics carrying robot based on digital twinning
By generating a three-dimensional terrain mesh and virtual gravity using digital twin technology, the problem of intelligent logistics handling robots slipping on low-traction surfaces was solved, enabling the robots to smoothly escape from complex ground environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU TESTING CENTRE OF CONSTRUCTION QUALITY AND SAFETY CO LTD
- Filing Date
- 2026-06-02
- Publication Date
- 2026-07-03
Smart Images
Figure CN122331561A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automated control technology for intelligent warehousing and logistics equipment, and in particular to an intelligent logistics handling robot based on digital twins. Background Technology
[0002] Currently, intelligent logistics handling robots play a crucial role in material handling in warehousing and industrial manufacturing scenarios. These devices typically travel autonomously along preset routes under the guidance of a central scheduling system, with the underlying controller often employing a speed closed-loop tracking mode to maintain the operating rhythm. This mode relies on real-time feedback data from wheel-end speed sensors, and continuously adjusts the input current of the drive motor through control algorithms to ensure that the actual travel speed of the robot closely follows the expected planned speed.
[0003] In real-world factory or warehouse environments, ground conditions often vary unpredictably. Besides the usual unevenness, some areas may develop low-friction zones due to residual water or accumulated dust and oil. When a handling robot enters such a low-friction surface, the physical contact friction between the drive wheels and the ground decreases, causing the wheels to easily lose traction and slip. At this point, the actual physical displacement of the robot lags behind, while the speed sensor continues to send wheel rotation pulses to the system.
[0004] Faced with slippage caused by sudden changes in physical resistance, the existing speed closed-loop tracking mechanism will experience adjustment mismatch. Since the vehicle displacement fails to reach the predetermined progress, the closed-loop adjustment circuit will automatically increase the output torque of the drive motor to attempt to catch up with the preset travel speed. This control logic, which continuously attempts to follow the speed setting, will further degrade the remaining traction on low-friction surfaces, causing the wheels to spin more freely and even triggering lateral skidding, making it difficult for the transport equipment to escape from the slippage zone. Therefore, the industry needs a solution that can automatically adjust the underlying control strategy to help the equipment smoothly extricate itself from the slippage situation. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent logistics handling robot based on digital twins to solve the technical problem that existing logistics handling robots, when encountering low-traction surfaces, suffer from increased wheel spin and difficulty in leaving the slippage area due to the continuous attempt to catch up with the planned speed while maintaining the conventional speed closed-loop control.
[0006] This invention provides an intelligent logistics handling robot based on digital twins, including a robot body and a digital twin control terminal. The digital twin control terminal communicates with the robot body and performs the following steps: The operational status data of the ontology is mapped to the twin model, and the target path is planned within the twin model; The difference between the actual driving power and the theoretical expected power of the body is extracted as the power deviation value. The power deviation value is converted into an equivalent height and superimposed on the bottom layer of the twin model to generate a three-dimensional terrain mesh. Virtual gravity is applied to the three-dimensional terrain mesh, the tangential component force at the mapped position of the body is calculated, and it is converted into a compensating torque and sent to the body. When the power deviation value is negative, a virtual swamp region with attenuation properties is generated within the twin model; When the target path passes through the virtual swamp area, the velocity closed loop of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body. The virtual tension is calculated based on the tensile displacement of the virtual navigation node from the main body, and then converted into an open-loop traction torque to drive the main body to move.
[0007] Optionally, mapping the runtime status data of the ontology to a twin model and planning the target path within the twin model includes: The sensors deployed on the main body are used to collect real-time multi-axis wheel speed values and spatial point clouds; The real-time multi-axis wheel speed values and the spatial point cloud are injected into the three-dimensional reference frame inside the twin model; In the twin model, avoid physical obstacles and search for a collision-free continuous grid connecting the scheduling start point to the scheduling end point; A guiding trajectory is generated along the geometric center sequence of the collision-free continuous grid, serving as the target path.
[0008] Optionally, before extracting the difference between the actual driving power and the theoretical expected power of the body as the power deviation value, the method further includes: In the twin model, a virtual rigid body mass representing the total mass of the entity under the current load state is established; Based on the set desired driving speed and desired driving acceleration, calculate the standard traction force of the total mass of the virtual rigid body under the set standard state; The reference power value is obtained by multiplying the standard traction force by the desired driving speed, and the reference power value is used as the theoretical expected power.
[0009] Optionally, converting the power deviation value into an equivalent height and superimposing it onto the bottom layer of the twin model to generate a three-dimensional terrain mesh includes: For positive power deviation values, a positive bulge height is generated by multiplying them by a preset height conversion constant; For negative power deviation values, the negative indentation depth is generated by multiplying them by the height conversion constant. The positive bulge height and the negative depression depth are reconstructed by top surface displacement along the normal direction perpendicular to the bottom layer of the twin model to generate the three-dimensional terrain mesh with high and low undulation deformation characteristics.
[0010] Optionally, the step of applying virtual gravity to the three-dimensional terrain mesh, calculating the tangential component of the body's mapped position, and converting it into a compensating torque that is then transmitted to the body includes: Extract the current stationary coordinates of the ontology mapped within the three-dimensional terrain mesh; Calculate the spatial directional derivative of the current stationary coordinates along the forward direction to obtain the local tilt angle; Multiply the set gravity constant by the total mass of the virtual rigid body of the main body, and then multiply it by the sine value of the local tilt angle to obtain the tangential component force; If it is determined that the body is in an uphill state along the three-dimensional terrain grid, a positive gain torque is sent to the body as the compensation torque; If it is determined that the body is in a downhill state along the three-dimensional terrain grid, a reverse braking torque is issued to the body as the compensation torque.
[0011] Optionally, when the power deviation value is negative, generating a virtual swamp region with attenuation properties within the twin model includes: Extract the absolute value of the power deviation value that is negative as the absolute power difference and the corresponding current dwell coordinate; Using the current dwell coordinates as the geometric center, the initial warning boundary is rendered using a size that is positively correlated with the absolute power difference, and the virtual swamp region is established within the twin model; The background evaporation clock is activated, and as physical time passes, the coverage area of the initial warning boundary gradually shrinks inward according to a set ratio until the coverage area of the virtual swamp area is cleared to zero, thereby simulating the natural dissipation process of low-friction pollutants on the real ground surface.
[0012] Optionally, when the target path passes through the virtual swamp area, the velocity closed-loop circuit of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body, including: Determine whether the target path spatially overlaps with the virtual swamp region at the current moment; If an overlap occurs, before the main body reaches the boundary of the virtual swamp area, the proportional-integral-differential algorithm loop on the bottom side of the main body responsible for adjusting the rotational speed deviation will be suspended to block the tracking of the target speed. The current pose state of the ontology is extracted from the twin model and copied to generate the virtual navigation node; The virtual navigation node continues to travel along the target path without being restricted by terrain friction, and a virtual elastic traction line with elastic tensile properties is generated between the virtual navigation node and the main body in a trapped state.
[0013] Optionally, the step of calculating the virtual tension based on the tensile displacement of the virtual navigation node away from the main body and converting it into an open-loop traction torque to drive the main body to move includes: The three-dimensional spatial straight-line distance between the virtual navigation node and the main body is read in real time, and the three-dimensional spatial straight-line distance is used as the stretching displacement; The virtual tension is calculated by multiplying the set virtual spring stiffness coefficient by the tensile displacement; The virtual tension is converted into the open-loop traction torque, and the drive motor on the bottom side of the body is controlled to shield the speed error feedback, and the torque is driven forward only according to the open-loop traction torque; By using the smooth, incremental pure torque-driven driving method, the wheels are prevented from spinning and getting stuck on surfaces with extremely low traction due to the continuous attempt to catch up with the planned speed.
[0014] Optionally, during the period when the drive motor on the bottom side of the control body is shielded from speed error feedback and driven forward only according to the open-loop traction torque, the method further includes: Monitor the angular deviation between the forward direction of the virtual navigation node and the current heading of the vehicle body; The included angle deviation is converted into differential distribution weights applied to the left and right drive wheels; Based on the total open-loop traction torque, asymmetrical yaw compensation torques are superimposed on the left and right drive wheels according to the differential distribution weight.
[0015] Optionally, after driving the body to move, a synchronization recovery step is further included: Track the current coordinates of the entity as it traverses the virtual swamp region; When the current stationary coordinates cross the far boundary of the virtual swamp area and the power deviation value falls back to the stable range, the friction loss condition is determined to have ended. The virtual navigation node and the virtual elastic traction line are deleted from the twin model. Reactivate the velocity closed-loop circuit of the body, allowing the body to resume its normal closed-loop tracking mode for the target path.
[0016] The present invention has achieved the following beneficial effects: This invention extracts the deviation between the robot's actual driving power and the theoretically expected power, converts it into an equivalent height in a digital twin model to generate a three-dimensional terrain mesh, and combines this with virtual gravity calculation of the tangential component for adaptive torque compensation, enabling the chassis to sense and adapt to resistance fluctuations in the micro-terrain. Addressing the issue of handling robots easily slipping and getting stuck on low-traction surfaces, as mentioned in the background art, this invention generates a virtual muddy area within the model when the power deviation is negative. When the device traverses this area, it actively blocks the underlying speed closed-loop control loop, cutting off the feedback path of continuously increasing motor torque due to the continuous attempt to catch up with the target speed. Simultaneously, the system establishes a virtual elastic traction line that follows the laws of physics between the virtual navigation node and the stuck robot, calculating the tensile displacement as a pure torque-driven open-loop traction torque. This gently increasing, flexible pure torque-driven mechanism avoids the worsening of wheel spin-out situations, guiding the logistics handling robot to smoothly escape from slippery areas by gradually accumulating ground thrust, thus improving the device's autonomous passage capability in complex ground physical environments.
[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.
[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A system module composition diagram of an intelligent logistics handling robot based on digital twin provided for an embodiment of the present invention; Figure 2 A general flowchart of the intelligent logistics handling robot control method based on digital twin provided in the embodiments of the present invention; Figure 3 This is a detailed flowchart of the steps for generating a three-dimensional terrain mesh in an embodiment of the present invention; Figure 4 This is a detailed flowchart illustrating the steps involved in calculating the tangential force component and converting it into compensating torque in an embodiment of the present invention. Figure 5 This is a detailed flowchart of the steps for blocking the speed closed loop and establishing a virtual elastic traction line in an embodiment of the present invention; Figure 6 This is a flowchart showing the detailed steps of performing synchronization recovery control in an embodiment of the present invention. Detailed Implementation
[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0021] This application discloses an intelligent logistics handling robot based on digital twins, such as... Figure 1 As shown, it includes the host and the digital twin control terminal.
[0022] In this embodiment, the main body is configured as a mobile electromechanical actuator in an industrial environment. Its hardware architecture includes a load-bearing chassis, a power battery system, servo motor groups mechanically connected to the drive wheels on both sides, and a multimodal sensing sensor group. The digital twin control terminal is deployed in a computing server, and its memory space is loaded with global 3D map structure data matching the operating environment and the rigid body dynamics parameter model of the main body. The digital twin control terminal communicates with the main body, establishing a data interaction communication link. The main body periodically encapsulates and sends the electromechanical parameters and sensor data collected by the chassis controller to the digital twin control terminal; for example... Figure 2 As shown, the digital twin control terminal performs the following control steps and sends control commands to the host.
[0023] Step S1: Map the runtime status data of the ontology to the twin model, and plan the target path within the twin model.
[0024] Specifically, mapping the operational status data of the ontology to a twin model and planning a target path within the twin model includes: collecting real-time multi-axis wheel speed values and spatial point clouds using sensors deployed on the ontology; injecting the real-time multi-axis wheel speed values and spatial point clouds into a three-dimensional reference frame within the twin model; avoiding physical obstacles in the twin model and searching for a collision-free continuous grid connecting the scheduling start point to the scheduling end point; and generating a guiding trajectory along the geometric center sequence of the collision-free continuous grid as the target path.
[0025] Furthermore, the multimodal sensing sensor group includes a photoelectric rotary encoder connected to the end of the rotor of the servo motor group, and a three-dimensional lidar fixed to the main body. The photoelectric rotary encoder outputs pulse electrical signals, and the processor of the main body counts the edge transition events of the pulse electrical signals. Combining this with the set rolling radius of the drive wheels and the transmission ratio of the reducer, the instantaneous tangential velocity of the wheel end of each independent drive wheel is calculated to form the real-time multi-axis wheel speed value. Simultaneously, the three-dimensional lidar emits laser pulses into the surrounding space and receives optical echo signals, calculates the spatial relative coordinates of each reflection feature point, and generates the spatial point cloud.
[0026] It is understood that the three-dimensional reference frame is the global static Cartesian coordinate system constructed by the digital twin model during the initialization phase. The digital twin control terminal inputs the real-time multi-axis wheel speed values into the kinematic model, calculates the lateral yaw angle micro-element and translational displacement micro-element of the body within the time step, and derives the cumulative pose change homogeneous matrix of the body in the local coordinate system. Using this homogeneous matrix, coordinate transformation is performed on the spatial point cloud, registering it from the local coordinate system to the three-dimensional reference frame. To control computational resource consumption, the digital twin control terminal performs downsampling reconstruction processing based on a voxel mesh model on the registered spatial point cloud, extracts the geometric centroid of the point cloud inside the voxel mesh, and completes the mapping and injection of spatial state data.
[0027] Specifically, in the twin model, physical obstacles are avoided, and a collision-free continuous grid connecting the scheduling start point to the scheduling end point is searched. The digital twin control terminal spatially discretizes the three-dimensional reference system, generating a three-dimensional discrete state matrix composed of grids. The number of centroids of point clouds contained in each grid is counted. If the count in a certain grid exceeds the set spatial occupancy density judgment threshold, the grid is marked as occupied to reconstruct the exact boundary of the physical obstacle. The setting of the judgment threshold directly maps to the optical divergence characteristics of the lidar hardware. The system extracts the vertical and horizontal angular resolution of the radar scan line, and combines it with the radial spatial straight-line distance between the current inspected grid and the optical emission center to calculate the physical spot diffusion area of a single laser beam projected to that depth. Then, the minimum rigid body feature cross-sectional area that can cause bottoming interference as defined in the industrial warehouse anti-collision specifications is extracted, divided by the above spot diffusion area, and the standard photon escape loss rate caused by floor dust is deducted. The resulting lower limit of the number of echo particles is established as the spatial occupancy density judgment threshold. The system performs an expansion algorithm on grates marked as occupied, reserving a safety margin for collision avoidance. The size of the grates covered by the expansion algorithm is controlled by the chassis's extreme kinematic sweep envelope. The computation module extracts half the diagonal length of the body's expanded rectangular contour as the basic geometric rotation radius, superimposed with the emergency braking physical sliding distance at the maximum planned speed and the bus communication hysteresis displacement, to synthesize the physical limit value of the expansion radius. This limit value is divided by the geometric side length of a single grates in the three-dimensional discrete state matrix and rounded up to directly generate the final morphological expansion array steps. The system receives the coordinates of the scheduling start point and the scheduling end point, and maps them to the starting and target empty grates in the three-dimensional discrete state matrix. The spatial graph search algorithm is invoked to evaluate the transition cost and heuristic estimated cost of adjacent empty grates in the three-dimensional discrete state matrix, extracting a continuous set of adjacent sequences that avoid all occupied grates, as the collision-free continuous grid. A guiding trajectory is generated along the geometric center sequence of the collision-free continuous grid as the target path. The geometric center coordinates of each grid cell in the collision-free continuous grid are extracted to form a discrete set of coordinate control points. A curve interpolation algorithm is called, using the discrete set of coordinate control points as nodes, and the maximum angular velocity and minimum turning radius parameters of the body are substituted to perform curve fitting, generating a spatial parameter curve with continuous curvature. This spatial parameter curve is established as the guiding trajectory and serves as the target path.
[0028] Step S2: Before extracting the difference between the actual driving power and the theoretical expected power of the body as the power deviation value, establish the total mass of the virtual rigid body representing the body under the current load state in the twin model, calculate the standard traction force and derive the theoretical expected power.
[0029] Specifically, such as Figure 3As shown, before extracting the difference between the actual driving power and the theoretical expected power of the body as the power deviation value, the method further includes: establishing a virtual rigid body total mass representing the body under the current load state in the twin model; calculating the standard traction force of the virtual rigid body total mass under the set standard state by combining the set desired driving speed and desired driving acceleration; multiplying the standard traction force by the desired driving speed to obtain a reference power value, and using the reference power value as the theoretical expected power.
[0030] Furthermore, a virtual rigid body total mass representing the actual body under the current load state is established in the twin model. The vehicle body mass parameters of the actual body in the non-load-bearing state are pre-stored in a system file. The chassis of the actual body is equipped with a weighing sensor. When the actual body carries cargo, the weighing sensor outputs an electrical signal. After the system performs analog-to-digital conversion and digital filtering on the signal, it substitutes it into a linear transformation equation to calculate the real-time physical mass parameters of the carried cargo. The actual body uploads the physical mass parameters to the digital twin control terminal. The system performs an addition operation on the physical mass parameters and the vehicle body mass parameters, and writes the sum into the mass attribute parameter structure of the physics engine within the twin model to establish the virtual rigid body total mass.
[0031] Understandably, the standard traction force of the virtual rigid body's total mass under a set standard state is calculated by combining the set desired driving speed and desired driving acceleration. The digital twin control terminal generates a desired driving speed sequence based on the target path. The desired driving acceleration is derived by performing differential calculation on the desired driving speed. Under the dynamic boundary condition assuming the ground surface is in a state of no abnormal wear, the system calls the vehicle's longitudinal dynamic equation for derivation: multiplying the virtual rigid body's total mass by the desired driving acceleration yields the acceleration drag component used to overcome the system's translational inertia; multiplying the virtual rigid body's total mass, the gravitational acceleration constant, and the set rolling friction coefficient constant yields the basic rolling resistance component used to overcome the basic rolling resistance. The acceleration drag component and the basic rolling resistance component are summed, and the resulting resultant force modulus value is established as the standard traction force.
[0032] Specifically, the reference power value is obtained by multiplying the standard traction force by the desired driving speed, and this reference power value is used as the theoretical expected power. The system extracts the standard traction force parameter and the desired driving speed parameter at the same moment and performs a multiplication operation to obtain the theoretical value of mechanical work power. This theoretical value of mechanical work power is divided by a pre-calibrated comprehensive transmission conversion efficiency attenuation factor to obtain the reference power value corresponding to the input side of the motor inverter. The attenuation factor is calculated by constructing a bidirectional coupling derating matrix of electromechanical thermodynamics. The microprocessor periodically reads the temperature rise data of the thermistor of the reduction gearbox housing through the underlying data bus, substitutes it into the built-in grease viscosity-temperature characteristic curve equation, and calculates the gear mechanical meshing dissipation ratio as the housing temperature drifts. Specifically, in order to reduce the floating-point operation load of the underlying controller, the grease viscosity-temperature characteristic curve equation is specifically instantiated as a one-dimensional discrete mapping table in the industrial field. This mapping table is constructed from the factory measured data points provided by the gearbox supplier and records the calibrated dissipation ratios corresponding to multiple discrete temperature nodes in the range of 0 degrees Celsius to 80 degrees Celsius. The system reads the case temperature in real time and uses a piecewise linear interpolation algorithm to directly address and calculate the exact gear mechanical meshing dissipation ratio in the one-dimensional discrete mapping table. Simultaneously, it extracts the substrate junction temperature parameter of the insulated-gate bipolar transistor inside the servo controller and, combined with the current pulse-width modulation switching frequency, calculates the electrical switching heat dissipation ratio of the semiconductor power device. This reference power value, free from the interference of unknown terrain friction resistance, is assigned by the system and established as the theoretical expected power.
[0033] Step S3: Extract the difference between the actual driving power and the theoretical expected power of the body as the power deviation value, convert the power deviation value into an equivalent height, and superimpose it onto the bottom layer of the twin model to generate a three-dimensional terrain mesh.
[0034] Specifically, the difference between the actual driving power and the theoretical expected power of the motor body is extracted as the power deviation value. The servo motor controller of the motor body collects the inverter stator AC current value and DC side voltage amplitude. The microprocessor calls the coordinate system transformation algorithm to decouple the current in the three-phase stationary coordinate system into direct-axis excitation current and quadrature-axis working current. The quadrature-axis working current component is extracted and multiplied with the motor torque constant to obtain the electromagnetic output driving torque. The microprocessor performs a multiplication operation on the electromagnetic output driving torque and the rotor mechanical angular velocity data to obtain the transient total power consumption. The system calculates the copper loss heat generation based on the phase current and stator winding resistance parameters, and calculates the iron loss based on the alternating frequency. The system extracts the stator winding cold-state DC resistance calibrated on the motor nameplate, and after temperature sensor compensation, multiplies it with the square of the effective value of the phase current to calculate the copper loss heat generation. At the same time, the system retrieves the frequency-iron loss comparison file calibrated on the factory no-load drag test platform for this model of motor, and directly looks up the table using the current alternating frequency to obtain the corresponding iron loss. The actual drive power parameter, representing the net effective mechanical work done at the spindle end, is extracted by subtracting the heat generated by copper losses and the amount of iron losses from the total transient power consumption, and then uploaded to the digital twin control terminal. The system performs a subtraction operation between the actual drive power parameter and the theoretical expected power at the same moment, and the output algebraic difference is established as the power deviation value.
[0035] Further, the step of converting the power deviation value into an equivalent height and superimposing it onto the bottom layer of the twin model to generate a three-dimensional terrain mesh includes: for positive power deviation values, multiplying them by a preset height conversion constant to generate a positive bulge height, representing that the body is subjected to physical friction resistance at the corresponding coordinates; for negative power deviation values, multiplying them by the height conversion constant to generate a negative depression depth, representing that the body is experiencing wheel spin at the corresponding coordinates; and reconstructing the positive bulge height and the negative depression depth along the normal direction perpendicular to the bottom layer of the twin model by top surface displacement to generate the three-dimensional terrain mesh with undulating deformation characteristics.
[0036] Understandably, for a positive power deviation value, a preset height conversion constant is multiplied to generate a positive bulge height, representing the physical friction resistance experienced by the body at the corresponding coordinate. When the bottom tire of the body passes through an area with high friction damping, the contact surface resistance increases. To maintain the planned speed, the closed-loop speed regulation loop increases the quadrature-axis current command output, causing the measured actual drive power value to exceed the theoretical expected power reference, making the power deviation value positive. The system extracts the configured height conversion constant, multiplies it with the positive power deviation value, and calculates the positive bulge height with a spatial normal scalar. This positive bulge height effectively maps the physical friction resistance state represented by the actual increase in dissipation in the digital model.
[0037] The height conversion constant is governed by the kinetic energy equation, which describes the conversion of work done by the chassis against resistance into potential energy. The computational unit extracts a single-step time-varying element from the underlying simulation of the digital twin system, divides it by the product of the total mass of the virtual rigid body at the current moment and the gravitational acceleration constant, and the resulting quotient is established as the dynamic height conversion constant. Relying on this physically equivalent extrapolation link, the additional dissipated energy accumulated by the power deviation within the time-varying element is mapped into the vertical normal displacement of the terrain deformation.
[0038] Specifically, for negative power deviation values, multiplying them by the height conversion constant generates a negative indentation depth, representing wheel spin at the corresponding coordinate. When the vehicle enters a surface with an abnormally low coefficient of friction, the tire's static friction decreases, causing the drive wheel to spin under electromagnetic traction. The speed regulation circuit reduces current output due to a feedback high rotational speed, resulting in actual driving power significantly lower than the theoretically expected power required to maintain movement; the power deviation value is negative within this range. The system multiplies this negative power deviation value by the height conversion constant to calculate the negative indentation depth, which has a negative Z-axis scalar. This parameter maps the lack of ground bearing capacity at the bottom layer of the model, representing the wheel spin condition.
[0039] Furthermore, the positive bulge height and the negative depression depth are reconstructed by top-surface displacement along the normal direction perpendicular to the bottom layer of the digital twin model, generating the three-dimensional terrain mesh with undulating deformation characteristics. The digital twin control terminal maintains a basic two-dimensional discrete raster matrix corresponding to the actual site in memory. Based on the coordinate information attached to the data frame, the corresponding target raster cell is indexed and located in the basic two-dimensional discrete raster matrix. The original reference elevation attribute value of the raster cell is extracted, and the calculated positive bulge height or negative depression depth is used as a normal variable. Algebraic addition and subtraction operations are performed on the reference elevation attribute value to update the node height data. The system extracts the elevation dataset of the updated node and its neighboring raster range, and calls a spatial domain two-dimensional convolutional smoothing algorithm to perform filtering processing. The two-dimensional filtering kernel matrix called by the spatial domain smoothing reconstructs the contact mechanical imprint of the driving tire pressing the ground surface in terms of size dimension and weight distribution. The system extracts the longitudinal physical contact length and lateral tread width of this model of solid rubber tire under standard rated axle load, divides them by the side length resolution of the basic two-dimensional discrete grid, and rounds up to delineate a filter kernel boundary equivalent to the actual tread geometry. The weight distribution of each element within the matrix directly maps to the Gaussian surface equation of elastic contact stress that decreases nonlinearly from the tire center to both sides. The processed discrete elevation matrix data is then used to generate a three-dimensional terrain mesh with a continuously transitioning surface curvature and undulating deformation topology.
[0040] Step S4: Apply virtual gravity to the three-dimensional terrain mesh, calculate the tangential component of the body's mapped position, and convert it into a compensating torque that is sent to the body.
[0041] Specifically, such as Figure 4 As shown, the step of applying virtual gravity in the three-dimensional terrain mesh, calculating the tangential component of the mapped position of the body, and converting it into a compensating torque and sending it to the body includes: extracting the current stationary coordinates mapped by the body in the three-dimensional terrain mesh; calculating the spatial directional derivative of the current stationary coordinates along the forward direction to obtain the local tilt slope angle; multiplying the set gravity constant by the total virtual rigid body mass of the body, and then multiplying it by the sine value of the local tilt slope angle to obtain the tangential component; if it is determined that the body is in an uphill state along the three-dimensional terrain mesh, sending a positive gain torque to the body as the compensating torque; if it is determined that the body is in a downhill state along the three-dimensional terrain mesh, sending a reverse braking torque to the body as the compensating torque.
[0042] Further, the current stationary coordinates mapped by the ontology within the 3D terrain mesh are extracted. Based on the pose matrix data uploaded by the ontology, the digital twin control terminal performs a ray perpendicular projection intersection operation on the reconstructed 3D terrain mesh surface. The centroid of the set of spatial points where the projected ray intersects the mesh surface is obtained and established as the current stationary coordinates. The spatial directional derivative of the current stationary coordinates along the forward direction is calculated to obtain the local tilt slope angle. Using the current stationary coordinates as the origin, the system extracts the 2D direction vector corresponding to the current heading angle of the ontology. Elevation scalar data at two points with fixed step lengths before and after the current stationary coordinates are extracted along this vector. The horizontal spatial fixed step length used to extract the elevation data at both ends is set to half the physical wheelbase constant of the front and rear axles of the chassis. Since the actual pitch attitude of the transport robot is mainly determined by the line connecting the physical contact points between the front and rear wheel sets and the ground, a rigid geometric alignment is achieved between the spatial span of finite difference sampling and the chassis's physical support wheelbase. This allows the calculated first-order spatial directional derivative to characterize the macroscopic slope angle currently adapted to the chassis system. The first-order spatial directional derivative is obtained by calculating the ratio of the elevation difference between two points to the horizontal physical distance span using the finite difference algorithm. The arctangent function is then called to calculate this first-order spatial directional derivative, and the extracted angle variable is established as the local tilt slope angle.
[0043] Understandably, the tangential component force is obtained by multiplying the set gravity constant by the total mass of the virtual rigid body, and then multiplying it by the sine value of the local tilt angle. The system retrieves the gravitational acceleration constant and reads the total mass parameter of the virtual rigid body, and performs a multiplication calculation between the two to obtain the basic gravity vector magnitude. Subsequently, the sine value of the local tilt angle is calculated, and multiplied by the basic gravity vector magnitude to decompose the component force scalar along the tangent direction of the mesh surface, which is defined as the tangential component force.
[0044] Specifically, if the body is determined to be in an uphill state along the three-dimensional terrain grid, a positive gain torque is issued to the body as the compensation torque. When the first-order spatial directional derivative is greater than zero, the body is determined to be in an uphill state along the three-dimensional terrain grid. At this time, the tangential component is the resistance force. The system multiplies the tangential component parameter by the driving wheel rolling radius parameter and divides it by the transmission ratio constant to obtain a torque scalar, sets its polarity to positive, and generates a positive gain torque as the compensation torque. This positive gain torque is superimposed on the current command set in the speed outer loop of the underlying motor controller. If the body is determined to be in a downhill state along the three-dimensional terrain grid, a reverse braking torque is issued to the body as the compensation torque. When the first-order spatial directional derivative is less than zero, the body is determined to be in a downhill state along the three-dimensional terrain grid, and the tangential component is the thrust component. The system executes the same torque conversion logic, sets the polarity to reverse, generates a reverse braking torque as the compensation torque, and issues it to the body. After receiving the torque value command, the underlying driver generates electromagnetic damping braking torque to counteract the gravitational acceleration effect caused by the virtual downhill slope.
[0045] Step S5: When the power deviation value is negative, a virtual swamp region with attenuation properties is generated within the twin model.
[0046] Specifically, when the power deviation value is negative, generating a virtual swamp region with attenuation properties within the twin model includes: extracting the absolute value of the negative power deviation value as the absolute power difference and the corresponding current residence coordinates; using the current residence coordinates as the geometric center, rendering an initial warning boundary using a size positively correlated with the absolute power difference, and establishing the virtual swamp region within the twin model; activating a background evaporation clock, and gradually reducing the coverage area of the initial warning boundary inward according to a set ratio as physical time passes, until the coverage area of the virtual swamp region is cleared to zero, thereby simulating the natural dissipation process of low-friction pollutants on the real ground surface.
[0047] Furthermore, the absolute value of the negative power deviation is extracted as the absolute power difference and the corresponding current dwell coordinates. When the system detects a negative power deviation, it determines that it has entered a slippage condition. The system extracts the absolute value scalar of this negative value as the absolute power difference and reads the vector data containing the X-axis and Y-axis positioning coordinates at the time of the anomaly, establishing them as the corresponding current dwell coordinates. Using the current dwell coordinates as the geometric center, the system renders the initial warning boundary using a size positively correlated with the absolute power difference, establishing the virtual swamp region within the twin model. The system calls the closed polygon region generation function in the two-dimensional global grid map, using the current dwell coordinates as the geometric center coordinates. Based on the set mapping formula, the system calculates a size variable that is positively correlated with the absolute power difference, generating a two-dimensional geometrically enclosing polygon. The mathematical analytical structure of the size variable is generated by summing the basic envelope base and the dynamic diffusion increment. The system directly extracts the physical circumcircle radius of the chassis and assigns it to the base envelope, ensuring that the generated warning area baseline completely encompasses the current geometric space occupied by the vehicle body. The dynamic diffusion increment is calculated by multiplying the absolute power difference by the energy diffusion slope coefficient preloaded in the site environment configuration file. This slope coefficient, measured in meters per kilowatt, represents the increase in the liquid spray radius generated by each kilowatt of ineffective mechanical power dissipated when the drive tire slips on a standard low-adhesion medium. This slope coefficient is obtained through physical slip test before leaving the factory. During the test, the system applies different levels of ineffective slip power to an epoxy floor surface sprinkled with a standard water-oil mixture. The extent of the contamination mark boundary expansion caused by wheel slippage is measured, and the average ratio of multiple tests is taken as the preload constant. Typically, under industrial epoxy flooring conditions, this coefficient is between 0.15 and 0.25 meters per kilowatt. The data attributes of the spatial grid nodes within the polygon boundary are uniformly changed to low-friction feature markers, forming the virtual swamp area. The outer contour of the polygon is marked as the initial warning boundary.
[0048] Understandably, by activating a background evaporation clock, the coverage area of the initial warning boundary gradually shrinks inward according to a set ratio as physical time passes, until the coverage area of the virtual swamp region is reduced to zero, thereby simulating the natural dissipation process of low-friction pollutants on the real ground surface. The system allocates an independent background evaporation clock thread to each generated virtual swamp region. Within the time step when the cycle is activated, this thread extracts a preset decay coefficient constant less than 1 and applies a scaling multiplication operation to the current initial warning boundary coordinate node matrix. Using the current resident coordinates as the scaling reference origin, the reduced region boundary is iteratively calculated. The decay coefficient constant is based on the thermodynamic evaporation time law of fluid media in industrial settings. The calculation thread, based on the real-time temperature and humidity parameters returned from the plant environment terminal, matches the evaporation half-life of liquid pollutants and inversely calculates the decay base that is strictly aligned with the current clock activation step; its value is limited by the underlying system to between 0.95 and 0.999. As the clock-triggered integral accumulates, the total area covered by the initial warning boundary converges in a monotonically decreasing direction. For grid nodes that exit the envelope, the system resets their status attributes to the idle state. When the clock thread detects that the polygon area scalar has shrunk and is below the system's set calculation precision threshold, it triggers a deregistration cleanup command to clear and release the data structure memory occupied by the virtual swamp region.
[0049] Step S6: When the target path passes through the virtual swamp area, the velocity closed loop of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body.
[0050] Specifically, such as Figure 5 As shown, when the target path passes through the virtual swamp area, the velocity closed-loop loop of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body. This includes: determining whether the target path and the virtual swamp area at the current moment spatially overlap; if they overlap, before the main body reaches the boundary of the virtual swamp area, pausing the proportional-integral-differential algorithm loop on the bottom side of the main body responsible for adjusting the rotational speed deviation, thus blocking the tracking of the target speed; extracting the current pose state of the main body in the twin model and replicating it to generate the virtual navigation node; allowing the virtual navigation node to continue moving along the target path without being restricted by terrain friction, and generating the virtual elastic traction line with elastic stretching characteristics between the virtual navigation node and the main body in a trapped state.
[0051] Further, it is determined whether the target path spatially overlaps with the virtual swamp region at the current moment. The system trajectory analysis module extracts the point sequence of the currently executing target path, calls the geometric intersection logic component, and performs an overlap test calculation with the polygon point sequence of the outer boundary of the currently active virtual swamp region. If the calculation output is true, it is determined that geometric spatial overlap has occurred. If overlap occurs, before the main body reaches the boundary of the virtual swamp region, the proportional-integral-differential algorithm loop on the bottom side of the main body responsible for adjusting the rotational speed deviation is suspended, blocking the tracking of the target speed. When the positioning coordinates of the main body shrink to the boundary of the virtual swamp region to a set lead time threshold, the digital twin control terminal sends a control mode switching message to the lower-level machine through the data bus. The lead time threshold set here is intended to compensate for the objective physical lag in the control command during its issuance. The system scheduling calculation kernel extracts the current instantaneous linear velocity vector magnitude of the main body, performs a multiplication calculation with the longest round-trip communication handshake cycle of the vehicle local area network control bus, and obtains the communication lag displacement. Based on this, an additional feedforward safety buffer distance equal to the width of the single-side drive tire is added, and the sum of the two is established as the lead time threshold. Utilizing a space-time look-ahead calculation mechanism based on velocity vectors, the system ensures that at the moment the vehicle's leading edge enters the low-friction mud interface, the integral history error accumulation register responsible for speed correction within the underlying microcontroller is accurately locked and stopped updating. Upon receiving this message, the underlying microcontroller locks the proportional-integral-derivative (PID) algorithm module of its internal speed outer loop, freezes its integral history error accumulation register data, and overwrites the proportional gain parameter to zero, thereby severing the data association path between the speed measurement feedback error signal and the given current input, and suspending the operation of the PID algorithm loop.
[0052] Understandably, the current pose state of the physical entity is extracted from the twin model and copied to generate the virtual navigation node. Simultaneously with executing the velocity closed-loop blocking command, the digital twin control terminal reads the current absolute position coordinate matrix and heading angle data of the physical entity. Using this parameter matrix, a pure kinematic coordinate system node is instantiated in the model and assigned to the virtual navigation node. The virtual navigation node continues to move along the target path without being restricted by terrain friction, and a virtual elastic traction line with elastic stretching characteristics is generated between the virtual navigation node and the physical entity in a trapped state. This virtual navigation node inherits the velocity curve of the original target path and strictly updates its own coordinates along the target path according to the plan within the interpolation period. The system extracts the coordinates of the moving virtual navigation node and the current physical entity's mapped coordinates, constructing a one-dimensional linear mathematical structure between the two sets of coordinate variables. The mechanical characteristics of this structure satisfy the physical Hooke's Law model, that is, the output variable is linearly proportional to the spatial stretching deformation distance of the two endpoints, and this is established as the virtual elastic traction line.
[0053] Step S7: Calculate the virtual tension based on the tensile displacement of the virtual navigation node from the main body, and convert it into an open-loop traction torque to drive the main body to move.
[0054] Specifically, the step of calculating virtual tension based on the tensile displacement of the virtual navigation node deviating from the main body and converting it into open-loop traction torque to drive the main body to move includes: real-time reading of the three-dimensional straight-line distance between the virtual navigation node and the main body, using the three-dimensional straight-line distance as the tensile displacement; multiplying the tensile displacement by a set virtual spring stiffness coefficient to calculate the virtual tension; the tuning process of the virtual spring stiffness coefficient is achieved by inversely solving the road surface limit static boundary under slippage conditions. The control architecture extracts the a priori limit static friction coefficient lower limit of the drive tires for the water film road surface, and combines it with the real-time total vehicle weight fed back by the current on-board weighing sensor to derive the maximum usable effective static friction thrust benchmark; simultaneously, it extracts the maximum geometric tensile physical tolerance distance that the scheduling algorithm allows the vehicle to disengage and follow. The value of this physical tolerance distance is limited by the geometric anti-collision hard boundary of the actual storage channel. The specific calculation method is as follows: Extract the physical span width of the current passage in the warehouse management system, subtract the maximum outer width of the main body, and deduct half of the obtained remainder from the safety buffer anti-scratch allowance (e.g., 0.1 meters) to establish the maximum geometric tensile physical tolerance distance. Divide the above maximum effective static friction thrust benchmark by the maximum tensile tolerance distance, and the resulting quotient is locked as the virtual spring stiffness coefficient. Convert the virtual tension into the open-loop traction torque, control the drive motor on the bottom side of the main body to shield the speed error feedback, and drive forward only according to the open-loop traction torque; with the smooth increase of the pure torque drive forward mode, avoid the wheels from spinning and getting stuck on the ground due to continuously trying to catch up with the planned speed on the road surface with extremely low adhesion.
[0055] Furthermore, the system reads the three-dimensional straight-line distance between the virtual navigation node and the main body in real time, and uses this three-dimensional straight-line distance as the tensile displacement. Within the calculation step, the system extracts the three-dimensional coordinate vector of the virtual navigation node and the actual three-dimensional coordinate vector uploaded by the main body sensor, and performs a vector subtraction operation. The Euclidean distance modulus is calculated for the resulting difference vector, and this scalar distance value is established as the tensile displacement. The virtual tension is calculated by multiplying the tensile displacement by the set virtual spring stiffness coefficient. The arithmetic unit extracts the pre-tuned virtual spring stiffness coefficient constant, multiplies it with the tensile displacement input multiplier, and outputs a mechanical scalar reflecting the virtual elastic tension, which is established as the virtual tension. The virtual tension is converted into the open-loop traction torque, and the drive motor on the bottom side of the main body is controlled to shield the speed error feedback and drive forward only according to the open-loop traction torque. The system substitutes the virtual tension data into the mapping transformation equation containing the gearbox transmission ratio and the effective radius of the drive wheel, solves to obtain the corresponding rotational torque scalar, encapsulates it, and sends it out as the open-loop traction torque. After receiving this data, the underlying servo driver inputs it to the current setting stage. The drive motor inverter pulse width modulation algorithm outputs current solely based on the current amplitude index corresponding to the open-loop traction torque, without responding to the error adjustment of the speed negative feedback closed loop. This smoothly increasing pure torque-driven forward movement avoids wheel spin and getting stuck on surfaces with extremely low traction due to continuous attempts to catch up with the planned speed. The virtual tension generation mechanism relies on the linear accumulation of the stretching displacement distance, resulting in a smoothly increasing drive torque with a gradual slope. This ensures that the driving thrust at the tire contact surface accumulates slowly, reaching the critical value of static friction to obtain forward thrust.
[0056] It is understandable that during the period when the drive motor on the bottom side of the control body shields the speed error feedback and only drives forward according to the open-loop traction torque, the method also includes: monitoring the angular deviation between the forward direction of the virtual navigation node and the current heading of the body; converting the angular deviation into a differential distribution weight applied to the left and right drive wheels; and, based on the total amount of the open-loop traction torque, superimposing asymmetrical yaw compensation torque on the left and right drive wheels respectively according to the differential distribution weight, in order to suppress the lateral sideslip yaw of the body in the absence of speed closed-loop monitoring.
[0057] Specifically, the algorithm monitors the angular deviation between the forward orientation of the virtual navigation node and the current heading of the vehicle body. The tangent direction vector of the virtual navigation node is extracted as the forward orientation reference, and the actual heading direction vector of the vehicle body is extracted simultaneously. The algorithm calculates the absolute physical angle deviation between the two sets of planar vectors using dot product and inverse cosine operations, recording it as the angular deviation. This angular deviation is converted into differential speed distribution weights applied to the left and right drive wheels. This angular deviation is then fed into a first-order proportional module with dead-zone logic. For angle deviations exceeding the dead-zone limit, a product calculation is performed using a set yaw adjustment coefficient to obtain the dimensionless differential speed distribution weight scalar parameter. The dead-zone angle limit range within the first-order proportional module is anchored to the combined physical tolerance of the angle acquisition resolution limit of the photoelectric rotary encoder and the accumulated assembly clearance of the drive wheel reduction gearbox, typically calibrated in hardware to be between ±1.5 and 2 degrees. This physical dead zone is used to absorb high-frequency vehicle body attitude vibrations caused by rough ground textures and to shield the high-frequency differential speed adjustments made by the servo motors on both sides, which exacerbate wear. Once the net deviation exceeds the dead zone, the extracted yaw adjustment coefficient calculation expression is set as follows: half the physical wheelbase span between the centers of the drive wheels on both sides of the chassis is used as the numerator, divided by the actual rolling radius of the currently driven tire under pressure as the denominator. The net angle deviation exceeding the dead zone is multiplied by this yaw adjustment coefficient to obtain the differential distribution weight. Based on the total open-loop traction torque, asymmetrical yaw compensation torques are superimposed on the left and right drive wheels according to the differential distribution weight, in order to suppress the lateral sideslip yaw of the main body in the absence of speed closed-loop monitoring. The system calls the differential distribution weight and the total output open-loop traction torque parameter to perform scalar multiplication calculation to obtain the yaw compensation torque value. Positive and negative direction logical operators are introduced, and the yaw compensation torque value is superimposed and overwritten into the control command queues of the left and right motors controlled by the main body in the form of addition and subtraction instructions, respectively, to generate a lateral correction torque matrix and suppress yaw deviation.
[0058] Step S8: After driving the body to move, perform a synchronization recovery step.
[0059] Specifically, after driving the body to move, a synchronization recovery step is also included, such as... Figure 6 As shown: Track the current stationary coordinates of the body as it passes through the virtual swamp area; when the current stationary coordinates cross the far boundary of the virtual swamp area and the power deviation value falls back to the stable range, it is determined that the friction loss condition has ended; delete the virtual navigation node and the virtual elastic traction line inside the twin model; reactivate the speed closed loop of the body, so that the body resumes the normal closed loop tracking mode for the target path.
[0060] Further, the system tracks the current stationary coordinates of the host as it traverses the virtual swamp region. The digital twin control unit continuously and dynamically tracks and renders the current stationary coordinates of the host's center within a discrete terrain grid coordinate system based on the positioning feedback data. When the current stationary coordinates cross the far boundary of the virtual swamp region, and the power deviation value falls back to a stable range, the friction loss condition is considered to have ended. The system performs dual condition verification: First, it performs an inclusion relationship test between the current stationary coordinates and the polygonal boundary of the virtual swamp region; when the coordinates are outside the boundary area, the spatial condition is satisfied. Second, the system uses a bandpass filter module to monitor the power deviation value stream reported from the underlying layer; when the value stably converges and remains within a set stable threshold range, the energy condition is satisfied. The system retrieves historical power deviation samples recorded during the host's factory initialization phase, when cruising at a constant speed under no-load on a standard high-friction flat surface, and calculates the absolute variance of its steady-state ripple fluctuation; simultaneously, the inherent quantization error width of the analog-to-digital conversion of the servo driver's AC side current sensor is directly superimposed during the calculation. The envelope synthesized from the extreme values of the aforementioned mechanical background resistance fluctuations and the electrical high-frequency quantization noise is solidified into a stable threshold range. When both spatial and energy conditions are met, the friction loss condition is considered to have ended.
[0061] Understandably, the virtual navigation node and the virtual elastic traction line are deleted within the twin model. After determining that the danger has passed, the digital twin control terminal stops the coordinate update interpolation operation of the virtual navigation node and unloads the data object identifier from the object activity table. Simultaneously, it disconnects the coordinate mapping pointer association and releases the computational structure resources occupied by the deleted virtual elastic traction line in the memory linked list. The speed closed-loop circuit of the host is reactivated, allowing the host to resume its normal closed-loop tracking mode for the target path. The digital twin control terminal issues a non-disruptive restart command to the underlying servo controller. The underlying microsystem reads the actual angular velocity and true rotational speed state parameters fed back by the current speed sensor, reverses and overwrites them as the initial state into the speed outer loop given target speed storage register, and clears the historical error data stored in the dormant proportional-integral-derivative controller accumulator space. The write lock permissions for calculation units such as the proportional coefficient are released. After completing the initialization state alignment, the system re-guides the closed-loop circuit to access the input target data in an orderly manner according to the speed planning constraint indicators, restoring the host to execute the normal closed-loop tracking mode for the target path.
[0062] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. An intelligent logistics carrying robot based on digital twinning, comprising a body and a digital twin control end, characterized in that, The digital twin control terminal communicates with the host and performs the following steps: The operational status data of the ontology is mapped to the twin model, and the target path is planned within the twin model; The difference between the actual driving power and the theoretical expected power of the body is extracted as the power deviation value. The power deviation value is converted into an equivalent height and superimposed on the bottom layer of the twin model to generate a three-dimensional terrain mesh. Virtual gravity is applied to the three-dimensional terrain mesh, the tangential component force at the mapped position of the body is calculated, and it is converted into a compensating torque and sent to the body. When the power deviation value is negative, a virtual swamp region with attenuation properties is generated within the twin model; When the target path passes through the virtual swamp area, the velocity closed loop of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body. The virtual tension is calculated based on the tensile displacement of the virtual navigation node from the main body, and then converted into an open-loop traction torque to drive the main body to move. 2.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, The step of mapping the runtime status data of the ontology to a twin model and planning the target path within the twin model includes: The sensors deployed on the main body are used to collect real-time multi-axis wheel speed values and spatial point clouds; The real-time multi-axis wheel speed values and the spatial point cloud are injected into the three-dimensional reference frame inside the twin model; In the twin model, avoid physical obstacles and search for a collision-free continuous grid connecting the scheduling start point to the scheduling end point; A guiding trajectory is generated along the geometric center sequence of the collision-free continuous grid, serving as the target path. 3.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, Before extracting the difference between the actual driving power and the theoretical expected power of the body as the power deviation value, the method further includes: In the twin model, a virtual rigid body mass representing the total mass of the entity under the current load state is established; Based on the set desired driving speed and desired driving acceleration, calculate the standard traction force of the total mass of the virtual rigid body under the set standard state; The reference power value is obtained by multiplying the standard traction force by the desired driving speed, and the reference power value is used as the theoretical expected power. 4.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, The step of converting the power deviation value into an equivalent height and superimposing it onto the bottom layer of the twin model to generate a three-dimensional terrain mesh includes: For positive power deviation values, a positive bulge height is generated by multiplying them by a preset height conversion constant; For negative power deviation values, the negative indentation depth is generated by multiplying them by the height conversion constant. The positive bulge height and the negative depression depth are reconstructed by top surface displacement along the normal direction perpendicular to the bottom layer of the twin model to generate the three-dimensional terrain mesh with high and low undulation deformation characteristics. 5.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, The process of applying virtual gravity to the three-dimensional terrain mesh, calculating the tangential component of the force at the mapped position of the body, and converting it into a compensating torque that is then transmitted to the body includes: Extract the current stationary coordinates of the ontology mapped within the three-dimensional terrain mesh; Calculate the spatial directional derivative of the current stationary coordinates along the forward direction to obtain the local tilt angle; Multiply the set gravity constant by the total mass of the virtual rigid body of the main body, and then multiply it by the sine value of the local tilt angle to obtain the tangential component force; If it is determined that the body is in an uphill state along the three-dimensional terrain grid, a positive gain torque is sent to the body as the compensation torque; If it is determined that the body is in a downhill state along the three-dimensional terrain grid, a reverse braking torque is issued to the body as the compensation torque. 6.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, When the power deviation value is negative, generating a virtual swamp region with attenuation properties within the twin model includes: Extract the absolute value of the power deviation value that is negative as the absolute power difference and the corresponding current dwell coordinate; Using the current dwell coordinates as the geometric center, the initial warning boundary is rendered using a size that is positively correlated with the absolute power difference, and the virtual swamp region is established within the twin model; The background evaporation clock is activated, and as physical time passes, the coverage area of the initial warning boundary gradually shrinks inward according to a set ratio until the coverage area of the virtual swamp area is cleared to zero, thereby simulating the natural dissipation process of low-friction pollutants on the real ground surface. 7.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, When the target path passes through the virtual swamp area, the velocity closed-loop circuit of the main body is blocked, and a virtual elastic traction line is established between the virtual navigation node in the twin model and the main body, including: Determine whether the target path spatially overlaps with the virtual swamp region at the current moment; If an overlap occurs, before the main body reaches the boundary of the virtual swamp area, the proportional-integral-differential algorithm loop on the bottom side of the main body responsible for adjusting the rotational speed deviation will be suspended to block the tracking of the target speed. The current pose state of the ontology is extracted from the twin model and copied to generate the virtual navigation node; The virtual navigation node continues to travel along the target path without being restricted by terrain friction, and a virtual elastic traction line with elastic tensile properties is generated between the virtual navigation node and the main body in a trapped state. 8.The intelligent logistics carrying robot based on digital twinning of claim 1, wherein, The step of calculating virtual tension based on the tensile displacement of the virtual navigation node from the main body and converting it into open-loop traction torque to drive the main body to move includes: The three-dimensional spatial straight-line distance between the virtual navigation node and the main body is read in real time, and the three-dimensional spatial straight-line distance is used as the stretching displacement; The virtual tension is calculated by multiplying the set virtual spring stiffness coefficient by the tensile displacement; The virtual tension is converted into the open-loop traction torque, and the drive motor on the bottom side of the body is controlled to shield the speed error feedback, so that it is driven forward by torque only according to the open-loop traction torque. 9.The intelligent logistics handling robot based on digital twinning of claim 8, wherein, During the period when the drive motor on the bottom side of the control body shields the speed error feedback and only drives forward according to the open-loop traction torque, the following is also included: Monitor the angular deviation between the forward direction of the virtual navigation node and the current heading of the vehicle body; The included angle deviation is converted into differential distribution weights applied to the left and right drive wheels; Based on the total open-loop traction torque, asymmetrical yaw compensation torques are superimposed on the left and right drive wheels according to the differential distribution weight. 10.The intelligent logistics carrying robot based on digital twinning according to claim 1, wherein, After driving the body to move, a synchronization recovery step is also included: Track the current coordinates of the entity as it traverses the virtual swamp region; When the current stationary coordinates cross the far boundary of the virtual swamp area and the power deviation value falls back to the stable range, the friction loss condition is determined to have ended. The virtual navigation node and the virtual elastic traction line are deleted from the twin model. Reactivate the velocity closed-loop circuit of the body, allowing the body to resume its normal closed-loop tracking mode for the target path.