Embedded distributed device coordinated motion control method

Abstract

The invention relates to the technical field of multi-device coordination control, in particular to an embedded distributed device coordination motion control method. The method comprises the following steps: carrying out force sensing data acquisition and processing on distributed equipment through an embedded actuator to obtain a force component vector; mapping the contact state of the equipment and the object according to the force component vector to obtain a contact characteristic matrix; acquiring material parameters and geometric parameters of the object; carrying out contact point stress distribution analysis based on the contact characteristic matrix and the geometric parameters to obtain a stress distribution diagram; performing moment balance calculation on the contact point of the object according to the stress distribution diagram to obtain a moment balance state; calculating a dynamic stability coefficient based on the moment balance state; and performing slip risk assessment according to the dynamic stability coefficient to obtain a slip risk map. By integrating real-time contact state mapping and a dynamic mechanical stability evaluation technology, the problems of accuracy and safety of operation of an object prone to deformation are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-device coordinated control, and in particular to an embedded distributed device coordinated motion control method. Background Art

[0002] When processing easily deformable objects such as large glass panels, hoses, and wiring harnesses, existing technologies rely too much on preset fixtures and fixed trajectories, and lack the ability to perceive and adapt to the real-time status of objects, resulting in low operational accuracy and easily causing deformation or damage to objects; traditional control systems have obvious shortcomings in multi-actuator collaboration, especially in resource-constrained embedded environments. It is difficult to achieve real-time force feedback information exchange and coordinated control between actuators, and often relies on high-bandwidth centralized architectures, with high communication overhead and delayed response; complex mechanical analysis and stability assessment algorithms usually require powerful computing resource support, and are difficult to execute in real time on embedded systems with strictly limited processor capabilities, memory capacity, and power consumption, resulting in the inability to implement advanced control strategies on edge devices.

[0003] In summary, existing technologies have problems such as difficulty in effectively processing deformable objects, low efficiency of multi-actuator collaboration, and inability to execute complex algorithms in real time on embedded systems, which need to be urgently addressed. Summary of the Invention

[0004] Based on this, it is necessary to provide an embedded distributed device coordinated motion control method to solve at least one of the above technical problems.

[0005] To achieve the above object, an embedded distributed device coordinated motion control method includes the following steps:

[0006] Step S1: collecting and processing force sensing data of the distributed device through the embedded actuator to obtain a force component vector; mapping the contact state between the device and the object based on the force component vector to obtain a contact feature matrix;

[0007] Step S2: Obtain the material parameters and geometric parameters of the object; perform force distribution analysis on the contact points based on the contact characteristic matrix and geometric parameters to obtain a force distribution diagram; perform torque balance calculation on the contact points of the object based on the force distribution diagram to obtain a torque balance state; calculate the dynamic stability coefficient based on the torque balance state; perform slip risk assessment based on the dynamic stability coefficient to obtain a slip risk diagram; perform material stress analysis on each area of ​​the object based on the force distribution diagram and material parameters to obtain stress distribution data; perform grasping stability assessment based on the stress distribution data to obtain a stability index map;

[0008] Step S3: Global force distribution is performed on multiple actuators through the embedded control network to obtain a global force distribution plan; local feedback control is performed based on the global force distribution plan to obtain a feedback control parameter set; communication link configuration parameters are established for the multiple actuators, and then a force balance configuration vector is generated based on the feedback control parameter set;

[0009] Step S4: perform trajectory planning and execution according to the force balance configuration vector, including generating a trajectory point sequence and performing distributed device coordinated motion control to obtain a collaborative control instruction flow.

[0010] The present application can realize real-time and accurate perception of the physical interaction state of each actuator and object by performing high-frequency and optimized force sensing data acquisition and processing on each distributed device embedded actuator. By converting the original sensor signal into standardized force components, accurate contact point position, local slip rate and material deformation estimation, a comprehensive contact feature matrix is generated. This distributed and real-time physical interaction information perception capability overcomes the limitations of traditional centralized systems relying on indirect global information or visual perception, providing direct, reliable and high-bandwidth physical feedback for subsequent distributed control decisions, enabling the system to capture subtle state changes of deformable objects under local stress, and is the basis for realizing distributed control based on physical interaction. By implementing an optimized mechanical stability evaluation algorithm on the embedded system, the mechanical state and potential risks of the current gripping configuration can be evaluated in real time and comprehensively. By analyzing the contact force distribution, torque balance, dynamic inertia influence, local slip risk and material stress, especially using a dynamic force closure algorithm optimized for embedded resource constraints (such as geometric approximation, principal direction decomposition, lookup table compensation), real-time stability calculation on low-performance processors is achieved. The generated stability index map intuitively displays the stability score, risk point position and maximum allowable force threshold of each region of the object. This enables the system to predict and identify potential instability and damage risks during operation, changing from "after-the-fact reaction" to "pre-emptive prevention", significantly improving the safety and success rate of operations on fragile or deformable objects. By building a distributed force regulation coordination mechanism based on local feedback and limited communication, the problem of real-time coordination of multiple actuators is effectively solved, especially in embedded environments. By determining the ideal force target through a global force distribution scheme and configuring adaptive local feedback control parameters for each actuator, each actuator can maintain a certain autonomy while sharing and adjusting force information in real time through lightweight and hierarchical communication between actuators (using differential encoding, bitmap marking and other optimization methods to reduce communication volume). This distributed architecture avoids the bottleneck of high-bandwidth centralized control, reduces communication overhead, improves system response speed and robustness, and even if part of the communication link is interrupted, other actuators can maintain basic stability, achieving efficient, flexible and reliable force coordination in the physical interaction process of a multi-actuator system. By converting high-level motion targets into embedded-optimized collaborative control instruction streams, precise synchronization of distributed devices and real-time adaptation are achieved. By planning and discretizing the overall path of the object and mapping it to each actuator to generate its path set, critical states during motion are identified and local adjustment rules responsive to real-time feedback are designed. Synchronous control schedules are used to ensure coordinated and consistent action of multiple actuators.Most importantly, by adopting a series of innovative techniques such as compact binary encoding, relative addressing, incremental encoding, context compression, and hierarchical adaptive caching, the size of the instruction stream is greatly compressed, enabling efficient storage and processing in the memory of resource-constrained embedded controllers, and transmission over limited-bandwidth communication networks. This enables the distributed actuators to accurately track the planned trajectory and, based on the real-time physical interaction feedback of step S1, make rapid and autonomous local behavior adjustments using the local adjustment rules of step S46 and the feedback correction parameters of step S48, while maintaining global coordination, ensuring the stability and accuracy of the operation of the deformable object.

[0011] Therefore, the present application provides an embedded distributed device coordinated motion control method, which integrates real-time contact state mapping and dynamic stability evaluation techniques to enable the system to accurately perceive the force state changes of the object; adopts a distributed force regulation coordination mechanism based on local feedback and limited communication to maintain the autonomy of the actuators while achieving global coordination; and through algorithm optimization such as geometric approximation, principal direction decomposition, and compact binary instruction encoding, reduces the computational complexity from O(n ) to O(n), enabling real-time execution of dynamic stability evaluation and coordination control algorithms on low-performance embedded processors. This series of innovations effectively solves the precision and safety problems of deformable object operation, overcomes the technical bottlenecks of multi-actuator real-time coordination, and breaks through the difficulties of implementing complex algorithms on resource-limited embedded systems. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 Figure 1 is a schematic diagram of the step flow of an embedded distributed device coordinated motion control method. DETAILED DESCRIPTION

[0013] The implementation of the present application, functional characteristics and advantages will be further described with reference to the embodiments and the accompanying drawings.

[0014] The technical method of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0016] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0017] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart of the steps of the embedded distributed device coordinated motion control method of the present invention. In this example, the embedded distributed device coordinated motion control method includes the following steps:

[0018] Step S1: collecting and processing force sensing data of the distributed device through the embedded actuator to obtain a force component vector; mapping the contact state between the device and the object based on the force component vector to obtain a contact feature matrix;

[0019] The embodiment of the present invention focuses on high-frequency, real-time sensor data acquisition and preliminary processing on resource-constrained embedded actuators, converting raw force / torque signals into standardized force components. In implementation, the embedded controller on each actuator synchronously acquires raw data from the six-dimensional force / torque sensor and its own position and posture at a frequency of 500 Hz through a high-speed interface (such as SPI). These raw data are subjected to optimized low-pass filtering (fixed-length sliding window, O(1) complexity), simplified outlier detection (based on MAD threshold), and incremental drift compensation to generate filtered force data. The filtered force data is then transformed from the sensor local coordinate system to a unified object reference coordinate system and decomposed into normal and tangential components to form a force component vector. Combining the actuator position and known end geometry, the precise contact point position is estimated through geometric relationships or the concept of force center to form a contact point space mapping. Based on the tangential force change and small position change, the local slip rate is calculated using a threshold method to obtain slip feature data. Using a simplified elastic model, the material deformation degree of the contact area is estimated based on the contact force to obtain a deformation parameter set. Finally, the contact point positions, force components, slip characteristics, and deformation parameters are integrated to construct a contact feature matrix containing the physical interaction characteristics of each contact point.

[0020] Step S2: Obtain the material parameters and geometric parameters of the object; perform force distribution analysis on the contact points based on the contact characteristic matrix and geometric parameters to obtain a force distribution diagram; perform torque balance calculation on the contact points of the object based on the force distribution diagram to obtain a torque balance state; calculate the dynamic stability coefficient based on the torque balance state; perform slip risk assessment based on the dynamic stability coefficient to obtain a slip risk diagram; perform material stress analysis on each area of ​​the object based on the force distribution diagram and material parameters to obtain stress distribution data; perform grasping stability assessment based on the stress distribution data to obtain a stability index map;

[0021] An embodiment of the present invention implements real-time mechanical stability assessment of a grasping configuration on an embedded system, including overall moment balance, dynamic impact, local slip risk, and material stress risk, ultimately generating a stability index map to guide control decisions. During implementation, parameters such as the object's preset mass, inertia tensor, and material elastic modulus are first acquired. Based on the forces and positions in the contact feature matrix, the force distribution on each surface region of the object is calculated to generate a force distribution map. The current resultant force and torque are then calculated using the object's center of mass as a reference to determine the moment balance state. The expected inertial force / torque is calculated based on the object's mass and the acceleration / angular acceleration of the expected motion trajectory. The contact feature matrix is ​​clustered based on spatial and mechanical similarity to generate simplified contact data. PCA is applied to the simplified contact data to identify the principal constraint directions and decompose the six-dimensional force / torque space into three two-dimensional planar constraints. Within each plane, a geometric approximation method (not convex hull calculation) is used to rapidly calculate the stability boundary. The expected inertial force / torque is projected onto each plane, and the distance to the boundary is evaluated to obtain the stability margin. Geometric approximation and linear decomposition errors are compensated using lookup tables and empirical models to obtain a compensated stability value. The compensated stability value and stability margin are combined to generate a dynamic stability coefficient ranging from 0 to 100. Based on the dynamic stability coefficient, the force ratio, slip ratio, local pressure, and consistency between the tangential force and the motion direction from the contact characteristic matrix, a slip risk score is calculated for each contact point through weighted fusion to generate a slip risk map. Using the force distribution map and material parameters, stresses and strains at the contact points and key parts of the object are estimated using simplified models (Hertzian contact, beam bending, and stretch estimation) and interpolation to generate stress distribution data. High-risk point sets are identified by integrating the moment equilibrium state, slip risk map, and stress distribution data. Based on the risk point set and the dynamic stability coefficient, the maximum force threshold to prevent instability or damage is calculated for each contact point, generating a force threshold map. The dynamic stability coefficient and stress distribution data are combined to generate a stability score matrix for each region of the object. Finally, the stability score matrix, risk point set, and force threshold map are integrated to construct a stability index map containing region scores, risk point labels, and force thresholds.

[0022] Step S3: Global force distribution is performed on multiple actuators through the embedded control network to obtain a global force distribution plan; local feedback control is performed based on the global force distribution plan to obtain a feedback control parameter set; communication link configuration parameters are established for the multiple actuators, and then a force balance configuration vector is generated based on the feedback control parameter set;

[0023] The embodiment of the application is to realize force regulation coordination among distributed actuators, achieve global force target while maintaining local autonomy, and optimize embedded communication. In the implementation, first, the current position, actual force and other state information of all actuators are initialized according to the stability index atlas and contact feature matrix to obtain an actuator state table. According to the risk points and stability scores of the stability index atlas, in combination with the actuator state table, the key contact points that need to be controlled preferentially are determined to obtain a control point priority list. Based on the stability index atlas and the priority list, the ideal target force of each control point that satisfies the overall force moment balance and optimizes stability is calculated through simplified optimization (such as weighted least squares) to obtain a global force distribution scheme. The contact point target force is mapped to the associated actuator in the actuator state table to obtain the actuator force target value. According to the risk information and target force of the stability index atlas, the order and rate of force adjustment of each actuator are calculated to obtain a force adjustment timing table. According to the stability of the associated region of each actuator in the stability index atlas, the parameters of the local force feedback control (PID) of each actuator are set to obtain a feedback control parameter set. Based on the priority list and the actuator state table, the coordination needs among actuators are analyzed to construct a functional proximity matrix. The differential encoding rules and bitmap marking mechanism are defined for data that need to be frequently exchanged. According to the functional proximity matrix and the bitmap template, the high, medium and low three-level communication frequency and content are configured to obtain a hierarchical communication scheme. According to the hierarchical communication scheme, the communication link configuration parameters including the communication objects, frequency, content and encoding format of each actuator are generated. Finally, the actuator force target value, force adjustment timing table, feedback control parameter set and communication link configuration parameters are integrated to form a local force balance configuration vector of each actuator, which guides the actuator how to coordinate and adjust the force.

[0024] Step S4: Trajectory planning and execution according to the force balance configuration vector, including generating a trajectory point sequence and performing distributed device coordinated motion control to obtain a collaborative control instruction stream.

[0025] An embodiment of the present invention transforms high-level motion goals into low-level instruction streams that can be precisely and synchronously executed by distributed embedded actuators, supporting real-time adaptation. During implementation, an external operation task description (starting / target poses, operation type) is first obtained. Combined with the grasping configuration information implicit in the force balance configuration vector, the overall motion target parameters (displacement, rotation, and time constraints) of the object are analyzed and determined. Based on the motion target parameters and environmental constraints (obstacle avoidance), a smooth main trajectory curve of the object's center of mass in the object reference coordinate system is planned using an improved Bezier curve. Based on the actuator response characteristics and force adjustment timing of the force balance configuration vector, the main trajectory curve is discretized and sampled to generate a sequence of object trajectory points with timestamps. This sequence of object trajectory points is then converted into the position and posture paths that each actuator end point must follow in the global coordinate system, resulting in an actuator path set. The actuator path set is analyzed for velocity / acceleration changes, associations with risk areas, and force adjustment timing to identify critical state points during the motion process and generate a state transition point table. Based on the feedback parameters and communication configuration in the force balance configuration vector, local position / force adjustment rules are designed for each actuator to respond to real-time contact state changes such as slippage and overload, resulting in local adjustment rules. The path set, state transition point table, and local rule triggering conditions are integrated to generate a time-accurate multi-actuator synchronization control schedule. Based on the local rules, synchronization schedule, and configuration, position / force feedback correction parameters (gain, threshold) are set for each actuator's local control loop to obtain feedback correction parameters. Finally, the actuator path points, state transition information, synchronization time, local rules (or their indices), and feedback parameters are encoded in a compact binary format designed in the substeps, with relative addressing, incremental encoding, and context compression. These are then organized into a hierarchical structure suitable for embedded caching, generating the final coordinated control instruction stream. This instruction stream is distributed via the embedded network, driving the actuators to execute motion synchronously and adaptively adjust based on real-time feedback using local rules.

[0026] Preferably, step S1 includes:

[0027] Step S11: while the embedded actuator controls the distributed device, force sensor data of the distributed device is collected to obtain an original force data set;

[0028] Step S12: performing signal preprocessing on the original force data set to obtain filtered force data;

[0029] Step S13: performing force vector decomposition on the filtered force data to obtain force component vectors;

[0030] Step S14: acquiring position data of the embedded controller, locating the contact point in combination with the force component vector, and obtaining a spatial mapping of the contact point;

[0031] Step S15: performing slip detection calculation based on the contact point spatial mapping and the force component vector to obtain slip feature data;

[0032] Step S16: Estimating material deformation based on the contact point spatial mapping and the force component vector to obtain a deformation parameter set;

[0033] Step S17: Generate a contact feature matrix based on the sliding feature data and the deformation parameter set.

[0034] In this embodiment of the present invention, a six-dimensional force / torque sensor (e.g., an ATI Nano sensor model FT300-S or similar performance sensor) is installed on each embedded actuator of a distributed device and connected to the actuator's embedded controller (e.g., an ARM Cortex-M4F-based microcontroller operating at 160 MHz) via an SPI interface. The embedded controller is configured with a hardware timer that triggers sensor data reading at a fixed frequency of 500 Hz. In the timer interrupt service routine, the controller reads the sensor registers via the SPI interface to obtain raw force data (Fx, Fy, Fz) and torque data (Tx, Ty, Tz) (typically as 16-bit signed integers). Simultaneously, the controller obtains the current position data of the actuator's end-of-line tool (e.g., Cartesian coordinates X, Y, Z and posture data Rx, Ry, Rz, represented as 32-bit floating-point numbers or high-precision fixed-point numbers) from its internal state or encoder interface. This raw force / torque and position data are encapsulated into a data packet along with the current system timestamp (e.g., a millisecond counter since system startup). The sensor raw data packet format is defined as: [Timestamp (32 bits), Fx (16 bits), Fy (16 bits), Fz (16 bits), Tx (16 bits), Ty (16 bits), Tz (16 bits), PosX (32 bits), PosY (32 bits), PosZ (32 bits), OrientRx (32 bits), OrientRy (32 bits), OrientRz ​​(32 bits)]. These data packets are stored in the embedded controller's internal RAM buffer to form a raw force dataset, awaiting subsequent processing.

[0035] The raw force dataset is processed in real time to remove noise and outliers. Low-pass filtering is first performed: a second-order Butterworth digital filter with a cutoff frequency of 50 Hz is used to filter out mechanical vibration and electrical noise. The filter is implemented using a direct form II structure, fixed-point arithmetic on an embedded controller, and state variables (past input / output samples) are stored in small arrays. The filtering process is performed in a sliding window manner. When each new data point arrives, the filter state is updated and the filtered force / torque value is output. The computational complexity is O(1) per sample. Outlier detection is then performed using a simplified method based on the median absolute deviation (MAD). A sliding window containing the most recent N samples (e.g., N=50) is maintained. The median of the data in the window is calculated, and the median of each data point is then calculated as the absolute deviation of the median from the median to obtain the MAD. If the absolute deviation of the current data point from the median exceeds a threshold of 3×MAD, it is marked as an outlier. Outliers are not discarded directly, but are replaced with the previous valid filtered data point. Finally, sensor drift compensation is performed: Before operation or periodically, the sensor zero point is calibrated, and the no-load sensor reading is recorded as the zero offset value. In subsequent processing, the corresponding zero offset value is subtracted from each raw reading. For slowly changing drift, an incremental compensation method is used, adjusting the zero offset value in small steps based on the average zero point change trend over a long period of time, reducing the need for historical data storage. The processed Fx, Fy, Fz, Tx, Ty, and Tz data are converted into filtered force data and stored in another RAM area.

[0036] The obtained filtered force data (Fx, Fy, Fz, Tx, Ty, Tz) is transformed from the sensor local coordinate system to the unified object reference coordinate system. First, based on the current position and attitude of the actuator (from the acquired Pos and Orient data), the 3×3 rotation matrix R and 3×1 translation vector t between the sensor local coordinate system and the object reference coordinate system are calculated. This transformation matrix R and t can be determined in advance based on the initial installation position of the actuator in the object reference system, or updated in real time according to the actuator kinematic model during the motion process. The filtered force vector F_sensor=[Fx,Fy,Fz] and torque vector τ_sensor=[Tx,Ty,Tz] Perform the coordinate transformation: F_object=R·F_sensor, τ_object=R·τ_sensor +r×F_object, where r is the vector from the origin of the object reference frame to the point where the sensor force is applied (usually the end of the actuator). Since this step focuses on force vector decomposition, we mainly use F_object. Next, we need to determine the surface normal vector at the contact point. For objects with known geometric shapes (such as planes and spheres), the surface normal vector n̂ can be calculated from the contact point position or the end-of-actuator posture. For unknown or complex shapes, The main direction of the contact force F_object can be used for preliminary estimation or can be determined with the help of tactile sensor data. n and the tangential component F t :F n = (F_object· )· , F t =F_object-F n The dot product here represents the vector dot product, and the dot product outside represents the scalar multiplied by the vector. is the unit normal vector. The decomposed F n and F t (including their respective magnitudes and directions) and the original F_object constitute the force component vector.

[0037] The precise position and attitude data of the end effector in the object reference frame are obtained, combined with the calculated force component vectors, to determine the contact point position of the force acting on the object surface. Assume that the end effector tool is a known geometry, such as a spherical gripper with a radius of r. If the force F_object measured by the sensor acts on the center of the sphere P_actuator, and the contact occurs on the surface of the sphere, then the contact point P_contact_estimated is located between P_actuator and the direction of the normal force ( , an offset from step S13). If F n is the compressive stress (i.e., F_object and in the opposite direction), the contact point is estimated to be P_actuator-r· . If the sensor is mounted on the gripping fingertip and the fingertip geometry is known (e.g., a plane pair, a cylindrical surface), the offset of the resultant force point relative to the sensor origin can be calculated based on the force vector F_object and torque vector τ_object (from step S13) measured by the sensor and the geometric model of the gripping fingertip. This offset point p_offset is converted from the sensor local coordinate system to the object reference coordinate system (using R and t from step S13), and then added to the actuator end position P_actuator to obtain a more accurate contact point position P_contact=P_actuator+R·p_offset. For irregularly shaped objects, the contact point position can be estimated using the concept of Center of Pressure by combining data from multiple sensors. Ultimately, each actuator determines one or more contact points associated with it, and the set of three-dimensional coordinates of these points constitutes a contact point space mapping.

[0038] Using the determined contact point position and the calculated force component vectors (normal force F n and tangential force F t ), detect whether the contact point slips and calculate the slip rate in real time. First, calculate the ratio of the local tangential force to the normal force: . If this ratio exceeds the preset local static friction coefficient threshold μ_static_threshold (the threshold can be set according to material experiments or experience, such as 0.5), the contact point is considered to have a slip tendency. Then, by comparing the contact point position change ΔP_contact=P_contact(t)-P_contact(t-Δt) between the current time step (t) and the previous time step (t-Δt) with the expected motion change caused by the actuator instruction. If the actuator instruction does not have expected movement in the tangential direction, but a significant ΔP_contact is detected (exceeding a small threshold, such as 0.1 mm), and the Ratio exceeds μ_static_threshold, it is determined that slip has occurred. The local slip rate is calculated as the modulus of the projection of the contact point position change vector ΔP_contact in the tangential direction when slip is determined to have occurred divided by the time interval Δt: SlipRate=|proj_ t (ΔP_contact)| / Δt. If no slip occurs, the slip ratio is 0. The slip feature data contains the slip status of each contact point (Boolean value: true / false) and the corresponding slip ratio value.

[0039] Based on the determined contact point position, the calculated force component vector and the known material properties of the objects (e.g. elastic modulus E, Poisson's ratio ν), the material deformation degree of the contact area is estimated. For simple compressive contact, a simplified linear elastic model can be used: normal deformation δ≈|Fn | / k_normal, where k_normal is the normal stiffness of the contact area. k_normal can be simplified for a specific geometry based on Hertzian contact theory (for example, the stiffness of a ball and a plane is related to E, ν, the contact force, and the contact geometry), or for flexible objects, k_normal can be directly regarded as an empirical value (for example, the radial stiffness of a hose). For deformation caused by tangential force, the tangential displacement δ can be estimated t ≈|F t | / k_tangential, where k_tangential is the tangential stiffness. For flexible objects such as hoses and harnesses, bending and stretching deformations also need to be considered. Based on the change in distance between contact points and the tangential force / torque, the local curvature change or the amount of stretch / compression between the contact points can be estimated. For example, if the distance between two adjacent contact points increases significantly, the tensile strain between them can be estimated. The deformation parameter set can include the normal compression amount for each contact point, an estimate of the tangential displacement, and, for flexible objects, an estimate of the local curvature or strain in the area near the contact point. These estimates are typically expressed as scalars or small-dimensional vectors.

[0040] The data related to each contact point is integrated to construct the final contact feature matrix. This matrix is ​​a two-dimensional data structure, with each row representing an independent contact point (usually indexed by the corresponding actuator ID or internally maintained contact point ID). The columns of the matrix contain the characteristics of the contact point: Columns 1-3: The three-dimensional spatial position coordinates of the contact point (Px, Py, Pz), which come from the contact point spatial mapping in step S14. Column 4: The normal force magnitude (|F n |), force component vector calculation from step S13. Columns 5-7: Tangential force vector components (F tx ,F ty ,F t2 ), from the force component vector calculation in step S13. Column 8: Local slip rate value, from the slip feature data in step S15. Column 9: Material deformation estimate (e.g., normal compression δ, or a scalar representing local strain), from the deformation parameter set in step S16. The contact feature matrix is ​​stored in a unified floating-point or high-precision fixed-point format and updated periodically (e.g., at the same frequency of 500 Hz as in step S11). This matrix is ​​a direct input data source for the mechanical stability assessment in step S2 and the force control coordination in subsequent steps, providing a comprehensive and standardized description of the current physical interaction state.

[0041] Preferably, calculating the dynamic stability coefficient based on the moment equilibrium state in step S2 includes:

[0042] Obtaining the mass parameters of the object; calculating the expected inertial force based on the mass parameters of the object and the preset expected motion trajectory;

[0043] Calculate the compensation torque value generated when the object moves based on the expected inertial force and geometric parameters;

[0044] Simplifying the contact data of the contact characteristic matrix to obtain simplified contact data;

[0045] The main direction is identified by simplifying the contact data to obtain the main constraint direction;

[0046] Decompose the constraint space of the main constraint direction and simplified contact data to obtain decomposed plane constraints;

[0047] Perform geometric approximation calculations based on the decomposed plane constraints to obtain the plane stability boundary;

[0048] Perform dynamic margin evaluation on expected inertia forces and plane stability boundaries to obtain stability margin;

[0049] Compensate for nonlinear effects based on stability margin and simplified contact data to obtain compensated stability values;

[0050] A dynamic stability coefficient is generated based on the compensated stability value and the stability margin.

[0051] In this embodiment of the present invention, during the system initialization phase, the mass M of the object to be manipulated (e.g., a large glass panel has a mass of 15.0 kg) and the position of its center of mass C (e.g., the vector [0.5, 0.8, 0.0] meters relative to the origin of the object's reference coordinate system) are stored as preset parameters in the embedded controller's non-volatile memory. During the motion planning phase (step S4), the object's temporal position, velocity, and acceleration along the desired motion trajectory are generated. At the current time step, the system reads the expected acceleration vector a_expected of the object's center of mass in the object's reference coordinate system from the preset desired motion trajectory data (e.g., [0.2, 0.1, 0.0] meters per second squared). Based on Newton's second law, the expected inertial force F_inertial = M × a_expected generated by this acceleration is calculated. For example, the inertial force is 15.0 kg × [0.2, 0.1, 0.0] meters per second squared = [3.0, 1.5, 0.0] Newtons. This three-dimensional vector F_inertial represents the equivalent external force on the center of mass of the object under ideal motion conditions and needs to be considered in the stability assessment.

[0052] In addition to linear inertial forces, an object's angular acceleration also generates a moment of inertia. During initialization, the system stores the object's inertia tensor, I_object (a 3×3 symmetric matrix describing the object's moment of inertia about its center of mass, for example, the inertia tensor of a thin plate). At the current time step, the system reads the object's expected angular acceleration vector, α_expected, in the object's reference coordinate system (e.g., [0.0, 0.0, 0.1] radians / second²) from the preset expected motion trajectory data. According to the Euler equation, the moment of inertia, τ_inertial, = I_object·α_expected + ω_expected×(I_object·ω_expected), where ω_expected is the expected angular velocity vector. To simplify embedded computations, for most operations primarily involving translations accompanied by small rotations, the latter term, ω_expected×(I_object·ω_expected), can be ignored or even approximated using a table lookup. The primary calculation is τ_inertial ≈ I_object·α_expected. In addition, if the inertial force F_inertial is not expected to act through the center of mass (for example, in some simplified models), it will also produce an additional torque about the center of mass. The compensation torque value τ_compensation is the sum of τ_inertial and the additional torque caused by F_inertial (if considered), representing the impact of dynamic motion on the moment balance of the object.

[0053] To reduce subsequent computational complexity, the contact point data in the generated contact feature matrix is ​​simplified. A clustering method based on spatial proximity and mechanical similarity is employed: All contact points in the contact feature matrix are traversed. Contact points whose spatial distance is less than a preset threshold d_merge (e.g., 5 mm) and whose normal force direction and magnitude differences, as well as their tangential force direction differences, are less than a preset threshold (e.g., normal force difference less than 10%, tangential force angle less than 15 degrees) are merged into an equivalent contact region. The position of the merged equivalent contact region is the centroid of the included contact points, and its equivalent normal and tangential forces are the sum of the normal and tangential forces corresponding to the included contact points. Isolated critical contact points, such as those with particularly large normal forces or high slip risk, are retained. The simplified contact data includes the reduced number of contact points / regions, as well as the position, equivalent normal force, and tangential force vectors of each simplified point / region. This simplification reduces the number of contact points from dozens to a few or a dozen, significantly reducing the amount of subsequent computation.

[0054] Analyze the normal and tangential force vectors of all equivalent contact points in the simplified contact data to identify the main directions of force constraints on the object. Combine the normal and tangential force vectors of all contact points (in the object reference coordinate system) into a vector set. Apply the principal component analysis (PCA) method to this vector set: construct the covariance matrix of the data and calculate its eigenvalues ​​and eigenvectors. The eigenvectors indicate the main direction of the data distribution, and the size of the eigenvalues ​​reflects the data variance along the corresponding direction. Select the three eigenvectors corresponding to the largest, second largest, and third largest eigenvalues ​​as the main constraint directions v ,v ,v These directions represent the three orthogonal directions in which the object is currently subjected to the strongest, second strongest, and weakest forces. For example, if the object is primarily subjected to a vertical upward gripping force, one principal direction will be approximately vertically upward; if it is also subjected to a horizontal gripping force, the other two principal directions will be approximately horizontal.

[0055] Based on the three identified main constraint directions v1, v2, and v3, the six-dimensional force / torque constraint space is decomposed into three two-dimensional plane problems. Each plane is spanned by two main constraint directions, for example, plane 1 is spanned by v1 and v2, plane 2 is spanned by v1 and v3, and plane 3 is spanned by v2 and v3. The equivalent normal force and tangential force vectors of each simplified contact point / area and their position vector relative to the center of mass of the object are projected onto these three planes. For each plane, the projected force vector of each contact point in the plane and the projected torque generated by the projected position of the projected force vector relative to the projected position of the center of mass of the object in the plane are calculated. For example, for the plane spanned by v i and v j Zhang Cheng, the projected force at contact point k is F_proj,k=proj(F_k,Plane(v i ,v j )), projection position r_proj,k=proj(r_k,Plane(v i ,v j Each decomposed plane constraint contains the projected force vectors and projected torque information for all contact points / regions within that plane. This decomposition transforms the complex six-dimensional force / torque balance problem into three independent two-dimensional problems, significantly reducing computational complexity.

[0056] Based on the projected force / torque information obtained for each decomposed plane, a geometric approximation method is used to calculate the stability boundary within that plane. The stability boundary is defined as the limit of the external force and torque combination that the object can withstand within that plane. Traditional precise calculations require constructing the convex hull of all contact points in the force-torque space, which is computationally intensive. This method uses geometric approximation: for each decomposed plane, the projected force vectors of all contact points within that plane are considered. The stability boundary is approximated as a polygon whose vertices are determined by a few "critical" contact points, such as those that provide maximum support force in a particular direction. Through fast scanning or direction-based search algorithms, the limiting combinations of points within the plane that can resist the resultant forces and moments applied in all directions are identified. For example, the combination of contact points that provides the maximum upward support force, the maximum leftward support force, the maximum clockwise moment, and so on are found within the plane. The approximate polygon defined by these key points serves as the plane's stability boundary. This approximation is significantly less computationally complex than convex hull calculations and can be performed in real time on embedded systems.

[0057] The calculated total force / torque (representing the dynamic load) of the expected inertial force F_inertial and the compensation torque τ_compensation is projected onto each decomposed plane. For example, the projection of the dynamic load onto plane 1 is L1_dynamic. Calculate the shortest distance d1 from this dynamic load point L1_dynamic to the stability boundary of plane 1. This distance d1 represents the "surplus" ability of the current contact configuration to resist the dynamic load. Normalize the shortest distance d1 by the characteristic dimension of the plane's stability boundary, L1_boundary (for example, the maximum diagonal length or average radius of the boundary polygon), to obtain the stability margin for the plane: Margin1 = d1 / L1_boundary. If the dynamic load point is within the boundary, the margin is positive; if it is outside the boundary, the margin is negative (indicating instability). The stability margin is the normalized shortest distance value for each of the three decomposed planes and reflects the system's ability to resist dynamic perturbations in each principal constraint direction.

[0058] Geometric approximation and linear decomposition can introduce errors, particularly for objects with large deformations or complex contact situations. To improve accuracy, a lookup table approach is used to compensate the resulting stability margin. During offline system calibration or simulation, the errors between geometric approximation and linear decomposition and more accurate methods (such as full six-dimensional convex hull calculation) are pre-calculated for different object types, typical contact configurations, and varying dynamic load levels. The errors are then stored in a multidimensional lookup table. The lookup table can be indexed by the stability margin value (e.g., by range), certain characteristics of the simplified contact data (e.g., number of contact points, compactness of distribution), and the object's flexibility parameters. During real-time operation, the lookup table is queried based on the current stability margin and simplified contact data characteristics to obtain the corresponding correction factor or compensation value. This compensation value is added to the original stability margin to obtain the compensated stability value. For example, compensated stability value 1 = margin 1 + LookupTable(margin 1, simplified characteristics). This compensated value more accurately reflects the actual stability after accounting for nonlinear effects.

[0059] The final dynamic stability coefficient is generated by combining the compensated stability values ​​(Compensated Stability Value 1, Compensated Stability Value 2, Compensated Stability Value 3) of the three decomposed planes with the original stability margins. A weighted minimum approach is used: the dynamic stability coefficient is first determined by the weakest compensated stability value among the three planes, for example, taking the min (Compensated Stability Value 1, Compensated Stability Value 2, Compensated Stability Value 3). Fine-tuning can then be performed based on the overall performance of the original stability margins. For example, if all three original margins are very large, overall stability will be better even if the smallest compensated value is not particularly high. The final dynamic stability coefficient is a normalized score ranging from 0 to 100. For example, the compensated stability values ​​can be mapped to a range of 0 to 90, and then a reward adjustment between 90 and 100 is performed based on the average or minimum value of the original margins. A higher score indicates better dynamic stability during the intended motion and greater resistance to external perturbations. This coefficient is a key output of step S2 and is used to guide subsequent risk assessment and force control.

[0060] Preferably, performing slip risk assessment according to the dynamic stability coefficient in step S2 includes:

[0061] Calculate the contact point force ratios on the contact characteristic matrix to obtain a force ratio table;

[0062] Estimate the friction coefficient of the contact point based on the contact characteristic matrix;

[0063] Calculate the initial value of slip risk based on the force ratio table and the contact point friction coefficient to obtain the initial risk value table;

[0064] Calculate the slip rate influencing factors on the contact characteristic matrix and obtain the slip rate factor table;

[0065] According to the contact characteristic matrix and dynamic stability coefficient, the motion direction consistency analysis is carried out to obtain the direction consistency table;

[0066] Calculate the local pressure influence based on the contact characteristic matrix and obtain the pressure influence table;

[0067] Calculate the stability coupling factor based on the dynamic stability coefficient and the initial risk value table;

[0068] Generate a slip risk map based on the slip rate factor table, directional consistency table, pressure influence table and stability coupling factor.

[0069] In the embodiment of the present invention, each contact point in the generated contact feature matrix is ​​traversed. For each contact point k, its normal force magnitude |F is extracted from the matrix. n | k and the magnitude of the tangential force |F t | k (The magnitude of the tangential force is the modulus of its three-dimensional tangential force vector.) Calculate the force ratio of the contact point: k =|F t | k / |F n | k This ratio reflects the proportion of the local tangential force to the normal constraint force and is a direct indicator of the potential slip tendency. n | k Close to zero, to avoid division by zero, we can set a very small threshold ε (for example, 1e-3 Newtons), when |F n | k <ε, Ratio k Treated as infinity or a very large preset value (such as 100). The force ratio values ​​of all contact points are stored in a list to form a force ratio table. This table corresponds to the row index of the contact feature matrix, and each element represents the force ratio value of a contact point.

[0070] Estimate the local friction coefficient at each contact point in the generated contact feature matrix. Due to the inhomogeneity of the object material (especially flexible objects) or surface state, different contact points have different friction coefficients. During the initialization phase, the system stores the object's average static friction coefficient μ_static_avg and kinetic friction coefficient μ_kinetic_avg. In real-time operation, dynamic estimation is performed based on the slip detection results: if a contact point does not slip (the slip rate is 0), its static friction coefficient μ_static_k can be conservatively estimated as the current force ratio Ratio k, with an upper limit of μ_static_avg. If a contact point is slipping (slip rate>0), its dynamic friction coefficient μ_kinetic_k can be estimated as the current force ratio Ratio k , with an upper limit of μ_kinetic_avg. More complex estimation methods can be used to make corrections based on historical data or local pressure. The estimated static friction coefficient μ_static_k and kinetic friction coefficient μ_kinetic_k for each contact point are stored in a table to form a contact point friction coefficient table.

[0071] Using the force ratio table and the contact point friction coefficient table, calculate the initial slip risk value of each contact point. For each contact point k, if no slip occurs, its initial risk value Risk_{0,k}=Ratio k / μ_static_k. If slip has occurred, then Risk_{0,k}=Ratio k / μ_kinetic_k, or simply set it to a high risk value (e.g., 1.5). A ratio greater than 1 indicates that the theoretical friction limit has been exceeded. The calculated initial risk values ​​are stored in a table, forming an initial risk value table. This table reflects the slip probability based solely on the local forces and friction coefficient.

[0072] Using slip feature data (including the slip rate of each contact point), calculate the impact factor of slip rate on risk. If a contact point has already slipped (slip rate > 0), it directly indicates that the risk has become a reality and its risk score should be significantly increased. Define a slip rate impact factor SlipFactor k :If the slip rate SlipRate of contact point k k >0, SlipFactor k Set the value of SlipRate to a larger multiplier (such as 1.5 or 2.0); if k =0, SlipFactor k Set to 1.0. Store the SlipFactor of all contact points in a table to form a slip factor table.

[0073] Analyze the relationship between the tangential force vector F_{t,k} and the object's calculated expected motion direction (determined by the expected acceleration a_expected or the expected velocity direction). If the tangential force direction at contact point k is consistent with the object's expected motion direction or potential instability direction (e.g., the direction of the weakest constraint identified), this indicates that the tangential force is pushing the object in the expected or unintended direction, increasing the risk of slippage. Calculate the angle θ between F_{t,k} and the expected motion direction vector d_motion k. Define a directional consistency factor DirFactor k :DirFactor k =(cos(θ k )+1) / 2. When the angle is 0 degrees (the directions are completely consistent), DirFactor k =1; when the angle is 180 degrees (completely opposite directions), DirFactor k =0; when the angle is 90 degrees, DirFactor k = 0.5. The DirFactor values ​​for all contact points are stored in a table to form a directional consistency table. Furthermore, the overall dynamic stability coefficient, DynamicStability (score of 0-100), is used as a global influencing factor. Low DynamicStability values ​​indirectly increase the risk of all points through the stability coupling factor.

[0074] Using the normal force magnitude |F n | k and the estimated contact point deformation δ k (or estimated contact area), estimate the local pressure at contact point k k ≈|F n | k / ContactArea k .ContactArea k According to δ k The contact geometry model is used for estimation. For example, for a ball in contact with a plane, the contact area is related to the normal force^(2 / 3) and the material's elastic modulus. Local pressure affects the friction coefficient and the material's load-bearing capacity. Low-pressure areas result in insufficient friction, while high-pressure areas lead to material damage. Define a pressure factor PressureFactor k :PressureFactor k Nonlinear relationship with local pressure, for example, at low pressure PressureFactor k >1 (increased risk) at moderate pressure PressureFactor k =1, when high pressure approaches the material limit PressureFactor k >1 (increased injury risk). The specific mapping relationship can be implemented using a lookup table or piecewise function. The PressureFactor of all contact points is stored in a table to form a pressure impact table.

[0075] Calculate the coupling influence factor of the overall dynamic stability on the local slip risk. The dynamic stability coefficient DynamicStability reflects the ability of the object as a whole to resist instability. When the overall DynamicStability is low, even if the local force ratio is not high, small external disturbances or internal force adjustment deviations are more likely to induce local slip. Define a stability coupling factor StabilityCoupling: StabilityCoupling is inversely proportional to DynamicStability, for example, StabilityCoupling=1+C / DynamicStability, where C is an empirical constant (for example, 50). When DynamicStability is 100, StabilityCoupling is close to 1; when DynamicStability is low, StabilityCoupling is significantly greater than 1. This factor will be applied to the risk calculation of all contact points as a weight for the overall stable state.

[0076] The obtained local risk factors and global stability coupling factors are weighted and integrated to calculate the final slip risk score of each contact point. k A fusion formula is: Risk k =(Risk_{0,k}×SlipFactor k ×DirFactor k ×PressureFactor k )^α×StabilityCoupling^β. α and β are weight coefficients used to adjust the influence of different factors (for example, α=1.0, β=0.5). Risk_{0,k} is the initial risk value, SlipFactor k DirFactor is the slip rate influencing factor, k is the directional consistency factor, PressureFactor k is the local pressure influence factor, and StabilityCoupling is the stability coupling factor. k The values ​​are normalized and mapped to a risk score on a scale of 0-100. The final slip risk score for each contact point is stored in a data structure and associated with its position on the surface to form a slip risk map. This map, often presented as a heat map, visually displays the slip risk of various regions on the surface, guiding subsequent force control and trajectory planning.

[0077] Preferably, performing material stress analysis on various regions of the object according to the force distribution diagram and material parameters in step S2 includes:

[0078] Estimate the contact area data of each contact point based on the contact characteristic matrix and material parameters;

[0079] Calculate the compressive stress value of each contact point based on the contact characteristic matrix and contact area data;

[0080] Estimate the bending stress value of the key parts of the object based on geometric parameters and force distribution diagram;

[0081] Calculate the tensile stress and shear stress values ​​borne by the object based on geometric parameters and force distribution diagram;

[0082] Estimate material deformation data based on material parameters, compressive stress values, bending stress values, tensile stress values, and shear stress values;

[0083] The stress distribution data is obtained by interpolating the stress distribution according to the material deformation data and geometric parameters.

[0084] In the embodiment of the present invention, the normal force magnitude |F in the generated contact feature matrix is ​​used. n | k And the material properties of the object used (for example, elastic modulus E, Poisson's ratio ν), estimate the actual contact area Area at each contact point k k For simple contact geometries (e.g., sphere and plane, cylinder and plane), the simplified Hertzian contact theory formula can be used for estimation. For example, for a sphere with a gripping fingertip approximated to have a radius R in contact with a flat plate, the contact radius a≈(3×|F n | k ×R / (4×E*))^(1 / 3), where E* is the equivalent elastic modulus, which is related to E and ν of the gripper and object materials. Contact area k ≈π×a For flexible objects such as hoses, the contact area is proportional to |F n | k The estimated contact area data of each contact point is stored in a list to form a contact area data table.

[0085] Using the normal force magnitude |F in the generated contact feature matrix n | k and the estimated contact area k , calculate the average compressive stress σ_{c,k} at each contact point k. The average compressive stress is defined as the normal force divided by the contact area: σ_{c,k}=|F n | k / Area kThis value represents the average pressure experienced by the contact area. In actual contact, compressive stress reaches its maximum value at the center of the contact area (Hertzian contact theory indicates that the maximum compressive stress is approximately 1.5 times the average compressive stress), but the average value provides an initial indication of risk. The calculated average compressive stress value for each contact point is stored in a table to form a compressive stress value table.

[0086] The generated force distribution map (showing the forces acting on various regions of the object's surface) and the object's geometric parameters (such as length, width, thickness, and cross-sectional shape) are used to estimate bending stresses at key locations of the object. Critical locations are typically areas with concentrated forces, large spans, or dramatic geometric changes. The object is approximated as a simple beam or plate structure, and the bending moment M_bend at the critical cross-section is calculated based on the force distribution. For example, for a long, rectangular object supported by two actuators, the central section experiences a bending moment. The bending stress σ_b ≈ M_bend × c / I, where c is the distance from the neutral axis to the outermost layer of the cross-section, and I is the moment of inertia of the cross-section. These geometric parameters (c, I) must be known in advance or estimated based on the object's current posture. For flexible objects such as hoses, the bending stress is related to the local curvature variation: σ_b ∝ E × κ, where κ is the local curvature. This local curvature can be estimated by analyzing the changes in contact point position or the posture of the actuator end effector. The estimated bending stress values ​​at key locations of the object are stored in a table, forming a bending stress table.

[0087] Using the object's geometric parameters and force distribution diagram, calculate the tensile stress σ_t and shear stress τ experienced globally or locally. Tensile stress is primarily caused by the axial tensile force F_tension acting on the object, where σ_t ≈ F_tension / A_cross-section, where A_cross-section is the cross-sectional area of ​​the force-bearing region. Shear stress is primarily caused by the shear force F_shear acting on the object, where τ ≈ F_shear / A_shear-area, where A_shear-area is the area subjected to the shear force. For flexible objects, tensile stress occurs in the portion being stretched (for example, a hose segment between two contact points), while shear stress occurs where torsional or lateral forces are applied. These forces and areas can be estimated based on the force distribution diagram and the object's geometry. The calculated tensile and shear stress values ​​are stored in corresponding tables.

[0088] The material parameters (E, ν, yield strength σ_y, shear modulus G, etc.) and calculated stress values ​​are used to estimate the material deformation in each region of the object. According to linear elasticity theory, strain is proportional to stress (Hooke's law). For example, compressive strain ε_c ≈ σ_c / E, tensile strain ε_t ≈ σ_t / E, and shear strain γ ≈ τ / G. Deformation can be estimated by multiplying the strain by the corresponding geometric dimension, for example, normal compression δ ≈ ε_c × thickness, and tensile deformation ΔL ≈ ε_t × original length. For complex stress states, equivalent stresses (such as Von Mises stress) can be used to determine whether the yield limit has been reached. For flexible objects, deformation is more directly characterized by local geometric changes (such as curvature and changes in contact point spacing). The estimated material deformation data for each region (for example, normal compression at each contact point, bending deformation angles at key locations, and elongation in the tensile section) are stored in a data structure to form material deformation data.

[0089] Stress / deformation estimates are primarily focused on contact points and a few critical locations. To obtain the global stress state on the object's surface, these local estimates need to be extended to the entire surface mesh. The stress distribution is interpolated using material deformation data (or stress estimates), along with the object's geometric parameters and surface mesh model. Distance-based interpolation methods (such as inverse distance weighted methods) or simplified physical models can be used. For example, stresses can be assumed to be higher near contact points and decay outward, or bending stresses can be assumed to have a specific distribution along the beam. The interpolation result is a stress risk index or stress estimate mapped onto the object's surface mesh. For example, the estimated stress value can be compared with the material's yield strength or a safe stress threshold to generate a stress risk score ranging from 0 to 100. These scores are stored in a matrix or list corresponding to the object's surface mesh, forming stress distribution data that represents the stress level and potential risk experienced by various regions of the object.

[0090] Preferably, performing grasping stability assessment based on stress distribution data in step S2 includes:

[0091] Identify the instability risk points based on the moment equilibrium state, slip risk map and stress distribution data to obtain a risk point set;

[0092] The maximum force threshold of the contact feature matrix is ​​calculated based on the risk point set and the dynamic stability coefficient to obtain the force threshold mapping;

[0093] Generate a stability scoring matrix based on dynamic stability coefficients and stress distribution data;

[0094] A stability index map is constructed based on the stability scoring matrix, risk point set and force threshold mapping.

[0095] In embodiments of the present application, the torque balance state (judging whether the whole is close to static or dynamic imbalance), the slip risk map (showing the possibility of slip of each contact point), and the stress distribution data (indicating the area of material damage) are comprehensively analyzed to identify the key areas or points on the object that have potential instability risks. A multi-dimensional risk threshold is set: for example, if a certain contact point meets any of the following conditions, it is marked as an instability risk point: 1) the slip risk score exceeds a preset high risk threshold (e.g. 80 / 100); 2) the stress risk score of the point or its nearby area in the stress distribution data exceeds a preset material damage threshold (e.g. 90 / 100); 3) the normal force provided by the actuator corresponding to the point is too low or too high, significantly deviating from the ideal force distribution, and the torque balance state shows that the whole is close to the instability boundary. The system traverses all contact points and object surface grid points to screen out points that meet the risk conditions to form a risk point set containing the positions (three-dimensional coordinates) of these points and the corresponding risk types / levels (e.g. slip risk, stress risk, torque imbalance risk). This set is the focus of attention for subsequent adjustment strategies.

[0096] Based on the instability risk point set and the overall dynamic stability coefficient DynamicStability, the maximum force threshold allowed to be applied is calculated for each contact point in the contact feature matrix. For non-risk points, the maximum force threshold can be set to a basic value according to the overall strength of the object and the gripper capacity. For areas marked as risk points, especially areas with high stress risk, the maximum normal force allowed to be applied needs to be significantly reduced to avoid material damage; for areas with high slip risk, the maximum value of the tangential force needs to be limited, and the allowed range of the normal force is adjusted according to DynamicStability to increase friction. The maximum force threshold calculation takes into account the local material strength limit (according to the stress analysis result) and the stability of the overall grasping configuration (according to DynamicStability). For example, for areas with a stress risk score exceeding a certain threshold, their maximum normal force threshold is set to a certain proportion of the material safe bearing capacity. When DynamicStability is low, the overall allowed force range will be tightened. The force threshold mapping is a list corresponding to the row index of the contact feature matrix, and each element specifies the maximum normal force and the maximum tangential force allowed to be applied to the contact point.

[0097] The overall dynamic stability coefficient DynamicStability and stress distribution data (stress risk score) are combined to generate a normalized stability score for each area on the surface of the object. The stability score reflects the contribution of the area to the overall grasping stability and the risk of the area itself bearing stress. A high DynamicStability value indicates that the overall grasping is stable and can be maintained even with slight fluctuations in local force; a low DynamicStability value is the opposite. A high stress risk score indicates that the material in the area is on the verge of damage and is a weak link in stability. A weighted fusion method is used: for example, the stability score StabilityScore of area k is k =w1×DynamicStability+w2×(100-StressRiskScore k ). Where w1 and w2 are weight coefficients (for example, w1=0.6, w2=0.4), StressRiskScore k is the stress risk score (0-100) for that region in the stress distribution data. The calculation results for all regions (corresponding to the surface mesh or simplified contact areas) are stored in a matrix to form a stability score matrix. Scores are typically normalized to a range of 0-100, with higher scores indicating greater stability or lower risk.

[0098] The stability score matrix, risk point set, and force threshold map are integrated to construct a final stability index map. This map provides a comprehensive visualization and data representation of the mechanical stability of the object in its current grasping state. The map is based on the object's surface mesh, with each grid cell (or corresponding to a simplified contact area) containing: 1) a normalized stability score; 2) if the grid cell contains a risk point, its location and risk type / level; and 3) if the grid cell corresponds to a contact point, the maximum normal force and tangential force thresholds allowed for that contact point. The stability index map is represented in a data structure that is easy to understand and use in subsequent control steps, providing critical physical interaction state and stability reference information for the distributed force control coordination in step S3.

[0099] Preferably, step S3 includes the following steps:

[0100] Step S31: Initialize the actuator state according to the stability index map and the contact characteristic matrix to obtain an actuator state table;

[0101] Step S32: determining key control points according to the actuator state table and obtaining a control point priority list;

[0102] Step S33: determining a global force distribution scheme according to the stability index map and the control point priority list;

[0103] Step S34: mapping the target force value of each control point in the global force allocation scheme to the specific actuator to obtain the actuator force target value;

[0104] Step S35: performing time sequence coordination calculation on the actuator force target value according to the stability index atlas to obtain a force adjustment time sequence table;

[0105] Step S36: setting a local feedback control parameter set for the actuator force target value according to the stability index atlas;

[0106] Step S37: constructing a communication link configuration parameter based on the control point priority list and the actuator state table;

[0107] Step S38: integrating the force adjustment time sequence table, the feedback control parameter set and the communication link configuration parameter to obtain a force balance configuration vector.

[0108] In the embodiment of the present application, when entering the force regulation stage, the system needs to know the current state of each actuator and its interaction with the object. All embedded actuators of the distributed devices are traversed. For each actuator, the current ID (for example, a unique integer identifier), the position of the end tool in the object reference coordinate system (for example, from the position of the actuator end), the actual applied force vector currently measured by the force sensor (the F object in the force component vector of step S13) and the current control mode (for example, position control, force control, impedance control) are read from the internal state register. At the same time, according to the contact feature matrix of step S17, the position of the contact point corresponding to the actuator on the object is determined, and the association of the actuator ID and the contact point ID (or position) is established in the actuator state table. The actuator state table is a data structure, each row corresponding to an actuator, containing: [actuator ID, current position (X, Y, Z), current actual applied force (Fx, Fy, Fz), current control mode, associated contact point ID]. This table provides a "snapshot" of the current physical interaction of the distributed system.

[0109] Using the actuator state table and stability index map, critical control points requiring focused attention and prioritized force regulation are identified. All marked risk points in the stability index map are traversed. For each risk point, the actuator state table is searched to determine which actuator is associated with the risk point (i.e., its contact points are located within the risk region). These contact points associated with the risk point are marked as high-priority control points. Furthermore, contact points corresponding to regions in the stability scoring matrix with scores below a preset threshold (e.g., 40 / 100) are also marked as high-priority. All contact points are sorted by their associated risk level (high-risk points > low-stability regions > other points) to generate a control point priority list. Each element in the list consists of: [contact point ID, priority level]. The highest-priority points are prioritized in subsequent force allocation and timing coordination.

[0110] Based on the stability index map and the control point priority list, the ideal target force vector to be applied by each key control point is calculated. The goal is to ensure that the resultant force and torque applied by all contact points meet the torque balance requirements for the object's current motion phase (taking into account expected inertia forces / torques) and optimize overall stability. This is a multi-objective optimization problem with the following objectives: 1) the resultant force and torque are equal to the desired values; 2) the force vector at each contact point does not exceed a maximum force threshold; 3) force adjustments at high-priority control points effectively reduce risk (e.g., increasing the normal force at slip-risk points and reducing the normal force at stress-risk points); and 4) force distribution is as uniform as possible to avoid local overloads. A simplified optimization algorithm based on weighted least squares or quadratic programming is employed: a system of equations is constructed with constraints including resultant force / torque balance and force thresholds. The objective function minimizes the difference between the current force and the target force, with greater weight given to differences at high-priority control points. Solving this optimization problem yields the ideal target force vector (magnitude and direction) to be applied by each contact point. This set of equations forms the global force distribution solution.

[0111] Assign the ideal target force vector for each contact point in the global force distribution scheme to the specific actuator associated with that contact point in the actuator state table in step S31. If an actuator has multiple contact points with the object through its end-tool, its total target force is the sum of the target forces at these contact points. The actuator force target value is a list, with each element corresponding to an actuator, consisting of: [actuator ID, target applied force vector (Fx, Fy, Fz)]. These target force vectors are defined in the object reference coordinate system.

[0112] To smoothly adjust the current actual forces to the target force values, and avoid inducing transient instability during the adjustment, the optimal timing of force adjustment for each actuator needs to be calculated. Based on the risk point locations and types in the stability index map, and the actuator force target values, the order and rate of force adjustment are determined. For example, if a region has high slip risk, and the target force distribution requires an increase in normal force to enhance the grip, the actuator associated with that region should increase normal force first, and at a faster rate. If a region has high stress risk, and the target force requires a decrease in normal force, the actuator should decrease normal force first, and at a slower rate. Timing coordination takes into account the spatial relationship between actuators: the force adjustments of neighboring actuators should be synchronized or performed in a coordinated manner as much as possible, to avoid unnecessary shear or twist on the object. The force adjustment timing table is a data structure containing timestamps and corresponding actuator force adjustment instructions, guiding each actuator when and how to start and complete force adjustment.

[0113] The parameters of local force feedback controllers (e.g. Kp, Ki, Kd gains of PID controllers) are set for the actuator force target values of step S34. These parameters determine how each actuator responds to the deviation between its measured actual force and target force. The parameter settings are adaptive, based on the stability scores and risk information of the actuator-associated regions in the stability index map. For example, for an actuator associated to a low stability region or a high risk point, its force feedback control proportional gain Kp needs to be set higher to respond to force deviation faster, but at the same time needs to limit the integral gain Ki and derivative gain Kd to avoid oscillation or noise sensitivity. For an actuator associated to a high stability region, a lower Kp or more gentle control parameters can be used. The feedback control parameter set is a list, each element corresponding to an actuator, containing: [actuator ID, force control Kp, force control Ki, force control Kd, allowed force error tolerance, response sensitivity]. These parameters are stored locally at each actuator, and used to drive its internal force control loop.

[0114] Based on the priority list of key control points and the actuator state table, the communication topology and parameters between actuators in the distributed embedded control network are constructed. The goal is to support local feedback and timing coordination while minimizing communication overhead. High-frequency communication links are prioritized between actuators associated with high-priority control points. These actuators need to frequently exchange their current actual applied forces, positions, slip states, and local stability assessment results for rapid local coordination. Direct communication links are also established between physically adjacent actuators, even if their associated control points are not high priority, because they support the same part of the object. Communication link configuration parameters include: [actuator ID_A, actuator ID_B, communication content type (e.g., force data, position data, stability score, local command deviation), communication frequency (e.g., 100 Hz, 50 Hz, 10 Hz), data encoding format]. Detailed examples of this step will be provided in subsequent steps.

[0115] The actuator force target values, force adjustment schedule, feedback control parameter set, and communication link configuration parameters are integrated to form the final force balance configuration vector. This force balance configuration vector is a complex distributed control instruction set, with each actuator receiving its corresponding portion. This vector is a direct input to the adaptive trajectory generation and execution in step S4. It instructs each actuator on how to coordinate with other actuators through force control and limited communication while maintaining local autonomy, achieving uniform force distribution, stable grasping, and preparing for subsequent motion.

[0116] Preferably, step S37 includes:

[0117] Perform actuator relationship analysis based on the control point priority list and actuator state table to obtain the functional proximity matrix;

[0118] Perform differential coding processing according to the functional proximity matrix to obtain differential coding rules;

[0119] A bitmap marking mechanism is implemented based on differential coding rules to obtain a bitmap compression template;

[0120] According to the bitmap compression template and the functional proximity matrix, a hierarchical communication strategy is configured to obtain a hierarchical communication scheme;

[0121] Generate communication link configuration parameters according to the hierarchical communication scheme.

[0122] In the embodiment of the present application, the critical control point priority list of the analysis step S32 and the executor state table of the step S31 are used to quantify the "functional proximity" or "cooperation demand intensity" between the executors. The executors with high functional proximity need closer communication coordination. A N x N functional proximity matrix M_neighbor is defined, where N is the total number of executors. The matrix element M_neighbor[i][j] represents the functional proximity between the executor i and the executor j. The calculation method is as follows: 1) if the contact points associated with the executor i and the executor j belong to the same high-risk area in the risk point set, M_neighbor[i][j] is set to a high value (for example, 10); 2) if the contact points associated with the executor i and the executor j belong to the same low stability area in the stability score matrix, M_neighbor[i][j] is set to a medium value (for example, 5); 3) if the executor i and the executor j are physically adjacent (according to the location data, the distance is less than a threshold d_physical_neighbor, for example, 10 centimeters), even if the priority of the associated contact points is not high, M_neighbor[i][j] is set to a basic value (for example, 2), because physical coupling needs local coordination; 4) for other cases, M_neighbor[i][j] is set to 0. The matrix is symmetric, M_neighbor[i][j]=M_neighbor[j][i]. This matrix quantifies which executors most need information exchange to support distributed cooperation.

[0123] To reduce the amount of data communicated, differential encoding is applied to data that the functional proximity matrix indicates requires frequent exchange (e.g., force vectors, positions, and stability scores). Differential encoding doesn't transmit the absolute value of the data, but rather the difference between the current value and the previously transmitted value. If the difference is small, it can be represented using fewer bits or even not transmitted at all if it's zero. Differential encoding rules are defined for different types of data (force, position, and score) based on the data type and expected range of variation: 1) Force vector differential: Calculate the difference ΔF = F_current - F_last between the current force vector F_current and the last transmitted force vector F_last. If |ΔF| is less than a threshold F_diff_threshold (e.g., 0.1 Newton), the force data is not transmitted or only a "no change" flag is transmitted. If the change is significant, ΔFx, ΔFy, and ΔFz are transmitted. These differences are typically smaller than the original force value and can be compressed using fixed-point or variable-length encoding. 2) Position differential: Calculate the difference ΔP = P_current - P_last between the current position P_current and the last transmitted position P_last. Similar processing is used. 3) Stability Score Difference: Calculate the difference between the current score (Score_current) and the last transmitted score (Score_last): ΔScore = Score_current - Score_last. This difference can be represented as a small integer. The differential encoding rules include the difference threshold for each data type, the encoding format, and how to represent "no change."

[0124] Based on differential encoding, a bitmap marking mechanism is introduced to further compress instruction packets. A fixed-length communication packet template is designed, which contains a bitmap field. Each bit in the bitmap corresponds to a potential data item in the packet (e.g., the X component of the force vector, the Y component of the force vector, the Z component of the position, the slip status flag, and the stability score). When constructing the actual data packet to be sent, the differential encoding rules are used to determine which data items have changed significantly (exceeding the differential threshold). Only the differences of those changed data items are included in the packet. The bitmap field uses a 1 or 0 to mark whether the corresponding data item is included in the current packet. The receiver parses the data packet based on the bitmap, reads only the fields indicated by the bitmap, and adds the difference to the last known value of the data item stored locally to recover the current data. The bitmap compression template defines the meaning of each bit in the bitmap, as well as the order and format of the fields in the packet. For example, an 8-bit bitmap can mark the presence of eight different data items.

[0125] A hierarchical communication strategy is configured by combining the functional proximity matrix and bitmap compression template. The frequency and content of communication between actuators i and j are determined by the value of the element M_neighbor[i][j] in the functional proximity matrix. 1) High-Frequency Communication (100-200Hz): Actuator pairs with high M_neighbor[i][j] values ​​(e.g., associated with the same high-risk area) are configured for high-frequency communication. Communication content primarily includes filtered force vector differentials and key state flags (such as slip status). Bitmap markers are used to transmit only the changes. Data packet size is strictly controlled within the CAN bus frame length (e.g., an 8-byte payload). 2) Medium-Frequency Communication (20-50Hz): Actuator pairs with medium M_neighbor[i][j] values ​​(e.g., associated with the same low-stability area or physically adjacent) are configured for medium-frequency communication. Communication content can include position differentials, local stability score differentials, and other information, using bitmap markers and differential encoding. 3) Low-frequency communication (1-10 Hz): Actuator pairs with low M_neighbor[i][j] values ​​are configured for low-frequency communication, primarily for synchronization or exchanging non-critical information. The hierarchical communication scheme specifies the communication frequency and the packet template (bitmap marking and differential encoding rules) used between each pair of actuators that need to communicate.

[0126] Based on a hierarchical communication scheme, each actuator's embedded controller generates its own local communication link configuration parameters. For each actuator i, the configuration parameters include a list of all other actuators j with which it needs to communicate. This configuration parameter set is distributed and stored in each actuator's local non-volatile memory. During runtime, each actuator actively sends and receives specified data to and from other actuators based on this configuration parameter set, enabling distributed information exchange within limited bandwidth and supporting local feedback and timing coordination.

[0127] Preferably, step S4 includes the following steps:

[0128] Step S41: Acquire operation task description data, perform target motion analysis in combination with the force balance configuration vector, and obtain motion target parameters;

[0129] Step S42: performing global path planning based on the motion target parameters to obtain a main trajectory curve;

[0130] Step S43: performing trajectory discretization processing on the main trajectory curve according to the force balance configuration vector to obtain a trajectory point sequence;

[0131] Step S44: performing actuator trajectory mapping on the trajectory point sequence to obtain an actuator path set;

[0132] Step S45: performing critical state identification on the actuator path set to obtain a state transition point table;

[0133] Step S46: performing local path adjustment on the actuator path set according to the force balance configuration vector to obtain a local adjustment rule;

[0134] Step S47: Generate a synchronization control schedule based on the actuator path set, the state transition point table and the local adjustment rules;

[0135] Step S48: performing feedback correction processing according to the local adjustment rule, the synchronous control schedule and the force balance configuration vector to obtain feedback correction parameters;

[0136] Step S49: Integrate the actuator path set, the state transition point table, the synchronous control time table, the local adjustment rules and the feedback correction parameters to generate a collaborative control instruction flow.

[0137] In an embodiment of the present invention, the system retrieves operational task description data from an external interface (e.g., a simplified task instruction received via a UART or network interface). The task description data includes the object's starting position (position and posture), target position, and high-level operation type (e.g., "grasp and lift," "move to position X," "rotate 90 degrees"). Combined with the force balance configuration vector, which contains the force / position relationships of each actuator in the current grasping configuration, the system analyzes how to modify these relationships to achieve the target motion. For example, if the task is to "lift the object," the system analyzes the force balance configuration vector to determine the goal of requiring all actuators to simultaneously apply upward forces while maintaining their relative positions. If the task is to "rotate the object," some actuators will need to apply tangential forces to generate rotational torque. The target motion analysis calculates the required displacement vector, rotation angle, and time or speed constraints for the entire object to achieve. These calculations constitute the motion target parameters.

[0138] Based on the motion target parameters (starting position, target position, time / velocity constraints) calculated in step S41, as well as pre-stored or sensor-perceived environmental constraint information (e.g., three-dimensional boundaries of obstacles, no-entry zones), the motion path of the object's center of mass is planned within the object's reference coordinate system. A modified Bezier curve (or B-spline curve) approach is employed: given a starting point, a target point, and intermediate control points (taking obstacle avoidance into account), a smooth three-dimensional path curve is generated that ensures position and velocity continuity. This improvement is achieved by considering the stability of the grasping configuration implied by the force balance configuration vector in step S38: The planned path avoids passing over or near regions marked as extremely low stability in the stability index map. The resulting main trajectory curve describes the ideal position and posture of the object's center of mass as it changes over time from the starting point to the end point.

[0139] The continuous master trajectory curve is discretized into a series of discrete time steps and corresponding object pose points. The sampling interval of the discretization is determined according to the actuator response characteristics and communication frequency specified in the force balance configuration vector. For example, if the control update frequency of the actuator is 500Hz and the system instruction sending frequency is 100Hz, the sampling interval is set to 10 milliseconds. At each sampling moment, the ideal position and posture of the object are sampled from the master trajectory curve. In addition, the discretization process also takes into account the force adjustment timing defined in the force balance configuration vector in step S38: the sampling density is increased near the time point where significant force adjustment is required to more accurately describe the force / pose coordinated change process. The trajectory point sequence is an ordered list.

[0140] The object's trajectory point sequence is converted into the specific motion path that the embedded actuator end-point of each distributed device must follow. For each timestamp t in the trajectory point sequence, the object's ideal pose P_object(t) at that moment is obtained. Based on the relative position r_contact,i of each actuator i's contact point with the object in the object's reference coordinate system stored in the force balance configuration vector (this relative position typically remains unchanged after the grasp is stabilized), the ideal position of the end-point of actuator i in the global coordinate system is calculated as P_actuator,i(t) = P_object(t) + r_contact,i. The addition here is a pose transformation, which needs to account for the rotational effect of the object's pose on the relative position vector. The actuator path set is a data structure containing N lists, where N is the number of actuators. Each list corresponds to an actuator and contains the ideal position and pose sequence for that actuator at each timestamp.

[0141] Analyze the actuator path set to identify critical states during motion execution that require a change in control strategy or special attention. Critical states include: 1) points where the direction of motion or velocity / acceleration changes significantly (e.g., starting, stopping, and turning); 2) object poses corresponding to risk points identified in the stability index map; 3) points where it is necessary to switch from position control to force control or adjust the target force value (according to the force adjustment timing table in step S35); and 4) points where the gripper approaches an environmental obstacle. By analyzing the velocity / acceleration change rate in the path point sequence, its relative position to the risk area, and the preset control mode switching conditions, these time points or path points are identified and the required actions (e.g., switching control mode, issuing specific instructions, activating anomaly detection) are determined. This information is stored in a state transition point table, for example: [timestamp / path point ID, trigger condition (e.g., reaching position X, detecting force F), transition action (e.g., switching to force control, adjusting gain), and post-transition parameter index].

[0142] Based on the local feedback control parameters and communication link configuration contained in the force balance configuration vector, rules are designed for each actuator to perform local adaptive adjustments during path execution. These rules are used to respond to detected real-time contact state changes (such as slippage and unexpected contact forces). For example: 1) Slip response rule: If the detected slip rate exceeds a threshold, the associated actuator, as configured in step S38, locally increases the normal force (within the maximum force threshold) and fine-tunes the tangential force or position to reduce slippage. The magnitude of this adjustment is limited by the feedback control parameters and communicated to adjacent actuators via the communication link. 2) Overload response rule: If the measured force exceeds the maximum force threshold, the actuator, as configured in step S38, locally reduces the applied force and fine-tunes the position to relieve stress. 3) Unexpected contact rule: If a force in an unexpected direction is detected, the actuator performs a small yield or increases stiffness according to the rules. Local adjustment rules define the local position or force fine-tuning actions that the actuator should perform under specific real-time feedback conditions, as well as their magnitude and rate limits. These rules are stored locally on each actuator and can be activated or parameterized via the instruction stream in step S49.

[0143] The actuator path set, state transition point table, and local adjustment rules are combined to generate a detailed, multi-actuator synchronous control schedule. This schedule specifies the actions that each actuator should perform at key moments, relative to the global system time. The schedule first contains all timestamps from the path set in step S44, specifying the desired position and posture (motion instructions) that each actuator should achieve at that moment. Then, the time points from the state transition point table in step S45 are added, along with the state transition instructions for the associated actuators (for example, at time T_switch, actuator i switches from position control to force control). Finally, the triggering conditions for the local adjustment rules in step S46 are associated with the schedule. While specific local adjustments are made in response to real-time feedback, the schedule can also include instructions to activate or adjust certain local rule parameters within specific time windows. The synchronous control schedule is a precise timeline that ensures that the issuance of instructions and the execution of key actions are coordinated and synchronized across all actuators during their motion.

[0144] Based on local adjustment rules, a synchronization control schedule, and a force balance configuration vector, feedback correction parameters are set for each actuator's local control loop. These parameters are used to address deviations between actual execution and the instructions specified in the schedule. The feedback correction mechanism adds position feedback and more complex error handling on top of force feedback. For example: 1) Position error PN control: If the actuator's actual position deviates from the target position in the schedule in step S47 by more than a threshold, a position PN controller is activated to adjust the output force or velocity command based on the deviation to bring it back on track. The PN gain is adjusted based on the local feedback control parameters configured in step S38 and the stability score. 2) Force deviation threshold: A force deviation threshold is set. When the deviation between the actual measured force and the target force in step S34 exceeds this threshold, the local adjustment rule in step S46 is triggered. 3) Correction gain: The gain coefficients for position and force correction are set, which determines the strength of the response to deviations. The feedback correction parameter set is stored locally on each actuator, guiding it on how to use sensor feedback to adjust its execution behavior in real time to track the ideal trajectory and maintain stability.

[0145] The actuator path set, state transition point table, synchronization control schedule, local adjustment rules, and feedback correction parameters are ultimately integrated to generate a coordinated control instruction stream for each actuator's embedded controller. The instruction stream is a sequence organized according to the synchronization control schedule. Each element in the stream is a command packet for a specific actuator at a specific time point or state trigger. The instruction packet content is encoded in a compact binary format: it includes the actuator ID, timestamp (or relative time), instruction type (e.g., set target pose, set target force, switch control mode, activate local rule), encoded parameter values ​​(position increment, force difference, gain index, etc.), and a checksum. The local adjustment rules themselves or their indexes, as well as the feedback correction parameter set, can be sent to the actuators as part of the instruction stream before motion begins or at state transition points. The instruction stream undergoes compression optimization (differential encoding, bitmaps, and hierarchical caching) to form an efficient and compact data stream. This data stream is distributed to the actuators via the embedded network, driving their synchronized and coordinated motion and supporting local adaptive adjustments based on real-time feedback.

[0146] Preferably, step S49 includes:

[0147] Design a basic instruction template based on the actuator path set, state transition point table and synchronization control time table;

[0148] Perform relative addressing conversion according to the basic instruction template and the executor path set to obtain a relative addressing instruction set;

[0149] Implement incremental encoding on the relative addressing instruction set to obtain an incremental encoding instruction stream;

[0150] Performing context-aware instruction compression on the incrementally encoded instruction stream according to a local adjustment rule to obtain compressed instruction data;

[0151] Hierarchical adaptive caching is performed based on compressed instruction data and feedback correction parameters to obtain a coordinated control instruction flow.

[0152] In this embodiment of the present invention, one or more basic binary instruction templates are designed based on the actuator path set (including sequential position / attitude and force targets), the state transition point table (including trigger conditions and transition actions), and the synchronization control timetable (including timestamp, actuator ID, and instruction type). Each template defines the data structure and field meanings of a specific instruction type. For example, a "motion instruction" template includes: [actuator ID (8 bits), timestamp differential (16 bits), instruction type (4 bits, e.g., 0001 for position control), parameter bitmap (8 bits), position X differential (16 bits), position Y differential (16 bits), position Z differential (16 bits), attitude quaternion increment (16 bits x 3)]. A "force control instruction" template includes: [actuator ID (8 bits), timestamp differential (16 bits), instruction type (4 bits, e.g., 0010 for force control), parameter bitmap (8 bits), force Fx differential (16 bits), force Fy differential (16 bits), force Fz differential (16 bits)]. A "state switch instruction" template contains: [actuator ID (8 bits), timestamp differential (16 bits), instruction type (4 bits, for example, 0011 indicates state switch), new control mode (4 bits), and associated parameter index (8 bits). These templates use bit field reuse and a variable structure instruction header design to ensure that the basic size of each template is compact.

[0153] Using the basic instruction template, the absolute path points of each actuator in the actuator path set are converted into an instruction sequence based on relative addressing. For each actuator i in the actuator path set, starting from its starting point, the position and posture of each subsequent path point are no longer expressed in absolute values. Instead, the difference relative to the target pose of the previous instruction point is calculated. If the previous instruction was a motion instruction, the difference ΔP between the current target pose and the previous target pose is calculated. If the previous instruction was a force control instruction, the relative addressing is based on the preset force control reference point. The relative addressing conversion replaces the absolute coordinates in all motion instructions with the difference relative to the previous instruction target. These relative addressing difference data are filled into the basic instruction template to form a relative addressing instruction set.

[0154] On the basis of the relative addressing instruction set, the incremental encoding technique is applied. For the difference data (such as ΔP, ΔF) in the relative addressing instruction set, variable-length encoding or fixed-point number encoding is used for further compression according to the numerical value. For example, if a position difference ΔX is usually less than 1 mm, a 16-bit fixed-point number can be used, in which the lower 10 bits represent the decimal part and the upper 6 bits represent the integer part, with a maximum representation range of about ±32 mm and an accuracy of about 0.001 mm. If the difference ΔX is very small (less than a certain small threshold, for example, 0.01 mm), it is encoded as zero or a specific "small change" code, without occupying additional bits. For the same type of instructions sent continuously, if the parameter difference (such as ΔP) is zero or very small for multiple time steps, a "keep current state" instruction can be sent or a counter can be used to represent the number of repetitions, instead of repeatedly sending zero difference instructions. Incremental encoding realizes the conversion of the relative addressing instruction set into a more compact binary data stream, forming an incremental encoding instruction stream.

[0155] The local adjustment rules are combined with the incremental encoding instruction stream to perform context-aware instruction compression. The local adjustment rules themselves (such as "if slip rate > threshold, then increase normal force K N") can be issued to the actuator as configuration data before the motion starts and stored locally on the actuator. The instruction stream does not need to include the complete rule logic, but rather instructions to activate, select, or parameterize these rules. For example, a "state switching instruction" or a dedicated "rule parameter instruction" can include a rule index and some parameters (such as new threshold, gain coefficient). In addition, based on the current motion stage and critical state, the instruction stream can dynamically adjust the frequency and content of the transmitted data. For example, in the high-speed moving stage, the position instructions are emphasized; in the grasping or releasing stage, the force or impedance control instructions are emphasized, and the communication frequency of related state (such as slip, stress) data is increased (according to the hierarchical communication scheme). Context-aware compression optimizes the content and structure of the instruction stream according to the current operation stage and local adjustment requirements, forming compressed instruction data.

[0156] Compressed command data, combined with feedback correction parameters, is organized into a hierarchical, adaptive cache format that fits within the memory structure of each actuator's embedded controller. The command stream is divided into multiple logical segments, such as time windows or operational phases. Important command segments (e.g., those about to be executed or related to critical states) are marked as high priority and should be loaded into the actuator's internal fast RAM cache. Non-critical or future command segments can be stored in lower-speed memory (e.g., off-chip SPIFlash) and loaded when needed. Feedback correction parameters (such as PN gains and thresholds) are stored as configuration data in specific registers or memory areas within the actuator for real-time use by the local control loop. The command stream can be organized into one or more files / data blocks, containing header information (e.g., total length, segment information, checksum) and encoded command data. The final coordinated control command stream is this organized and optimized data set, ready for distribution to the actuators via the embedded network. Each actuator receives and interprets its corresponding command segment, utilizing locally stored feedback correction parameters and local adjustment rules to drive its actuators for synchronized and coordinated motion, adaptively adjusting based on real-time sensor feedback.

[0157] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0158] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. An embedded distributed device coordinated motion control method, characterized in that: The following steps are involved: Step S1: collecting and processing force sensing data of the distributed device through the embedded actuator to obtain a force component vector; mapping the contact state between the device and the object based on the force component vector to obtain a contact feature matrix; Step S2: Obtaining material parameters and geometric parameters of the object; performing force distribution analysis on the contact point based on the contact characteristic matrix and geometric parameters to obtain a force distribution diagram; performing moment balance calculation on the contact point of the object based on the force distribution diagram to obtain a moment balance state; Calculate the dynamic stability coefficient based on the moment equilibrium state; Perform slip risk assessment based on the dynamic stability coefficient to obtain a slip risk map; perform material stress analysis on each area of ​​the object based on the force distribution map and material parameters to obtain stress distribution data; The grasping stability is evaluated based on the stress distribution data to obtain a stability index map; Step S3: performing global force distribution on multiple actuators through the embedded control network to obtain a global force distribution solution; Perform local feedback control based on the global force distribution scheme to obtain a feedback control parameter set; construct communication link configuration parameters for multiple actuators and then generate a force balance configuration vector based on the feedback control parameter set; Step S4: perform trajectory planning and execution according to the force balance configuration vector, including generating a trajectory point sequence and performing distributed device coordinated motion control to obtain a collaborative control instruction flow.

2. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: while the embedded actuator controls the distributed device, force sensor data of the distributed device is collected to obtain an original force data set; Step S12: performing signal preprocessing on the original force data set to obtain filtered force data; Step S13: performing force vector decomposition on the filtered force data to obtain force component vectors; Step S14: acquiring position data of the embedded controller, locating the contact point in combination with the force component vector, and obtaining a spatial mapping of the contact point; Step S15: performing slip detection calculation based on the contact point spatial mapping and the force component vector to obtain slip feature data; Step S16: Estimating material deformation based on the contact point spatial mapping and the force component vector to obtain a deformation parameter set; Step S17: Generate a contact feature matrix based on the sliding feature data and the deformation parameter set.

3. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: Calculating the dynamic stability coefficient based on the moment equilibrium state in step S2 includes: Obtaining the mass parameters of the object; calculating the expected inertial force based on the mass parameters of the object and the preset expected motion trajectory; Calculate the compensation torque value generated when the object moves based on the expected inertial force and geometric parameters; Simplifying the contact data of the contact characteristic matrix to obtain simplified contact data; The main direction is identified by simplifying the contact data to obtain the main constraint direction; Decompose the constraint space of the main constraint direction and simplified contact data to obtain decomposed plane constraints; Perform geometric approximation calculations based on the decomposed plane constraints to obtain the plane stability boundary; Perform dynamic margin evaluation on expected inertia forces and plane stability boundaries to obtain stability margin; Compensate for nonlinear effects based on stability margin and simplified contact data to obtain compensated stability values; A dynamic stability coefficient is generated based on the compensated stability value and the stability margin.

4. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: The slip risk assessment according to the dynamic stability coefficient in step S2 includes: Calculate the contact point force ratios on the contact characteristic matrix to obtain a force ratio table; Estimate the friction coefficient of the contact point based on the contact characteristic matrix; Calculate the initial value of slip risk based on the force ratio table and the contact point friction coefficient to obtain the initial risk value table; Calculate the slip rate influencing factors on the contact characteristic matrix and obtain the slip rate factor table; Conduct motion direction consistency analysis based on contact characteristic matrix and dynamic stability coefficient to obtain direction consistency table; Calculate the local pressure influence based on the contact characteristic matrix and obtain the pressure influence table; Calculate the stability coupling factor based on the dynamic stability coefficient and the initial risk value table; Generate a slip risk map based on the slip rate factor table, directional consistency table, pressure influence table and stability coupling factor.

5. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: In step S2, performing material stress analysis on each region of the object according to the force distribution diagram and material parameters includes: Estimate the contact area data of each contact point based on the contact characteristic matrix and material parameters; Calculate the compressive stress value of each contact point based on the contact characteristic matrix and contact area data; Estimate the bending stress value of the key parts of the object based on geometric parameters and force distribution diagram; Calculate the tensile stress and shear stress values ​​borne by the object based on geometric parameters and force distribution diagram; Estimate material deformation data based on material parameters, compressive stress values, bending stress values, tensile stress values, and shear stress values; The stress distribution data is obtained by interpolating the stress distribution according to the material deformation data and geometric parameters.

6. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: The grasping stability evaluation based on the stress distribution data in step S2 includes: Identify the instability risk points based on the moment equilibrium state, slip risk map and stress distribution data to obtain a risk point set; The maximum force threshold of the contact feature matrix is ​​calculated based on the risk point set and the dynamic stability coefficient to obtain the force threshold mapping; Generate a stability scoring matrix based on dynamic stability coefficients and stress distribution data; A stability index map is constructed based on the stability scoring matrix, risk point set and force threshold mapping.

7. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: Initialize the actuator state according to the stability index map and the contact characteristic matrix to obtain an actuator state table; Step S32: determining key control points according to the actuator state table and obtaining a control point priority list; Step S33: determining a global force distribution scheme according to the stability index map and the control point priority list; Step S34: mapping the target force value of each control point in the global force distribution scheme to the specific actuator to obtain the actuator force target value; Step S35: performing a time-series coordination calculation on the actuator force target value according to the stability index map to obtain a force adjustment time sequence table; Step S36: performing local feedback control parameter setting on the actuator force target value according to the stability index map to obtain a feedback control parameter set; Step S37: constructing communication link configuration parameters based on the control point priority list and the actuator status table; Step S38: Integrate the force adjustment timing table, the feedback control parameter set and the communication link configuration parameters to obtain a force balance configuration vector.

8. The embedded distributed device coordinated motion control method according to claim 7, characterized in that: Step S37 includes: Perform actuator relationship analysis based on the control point priority list and actuator state table to obtain the functional proximity matrix; Perform differential coding processing according to the functional proximity matrix to obtain differential coding rules; A bitmap marking mechanism is implemented based on differential encoding rules to obtain a bitmap compression template; According to the bitmap compression template and the functional proximity matrix, a hierarchical communication strategy is configured to obtain a hierarchical communication scheme; Generate communication link configuration parameters according to the hierarchical communication scheme.

9. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Acquire operation task description data, perform target motion analysis in combination with the force balance configuration vector, and obtain motion target parameters; Step S42: performing global path planning based on the motion target parameters to obtain a main trajectory curve; Step S43: performing trajectory discretization processing on the main trajectory curve according to the force balance configuration vector to obtain a trajectory point sequence; Step S44: performing actuator trajectory mapping on the trajectory point sequence to obtain an actuator path set; Step S45: performing critical state identification on the actuator path set to obtain a state transition point table; Step S46: performing local path adjustment on the actuator path set according to the force balance configuration vector to obtain a local adjustment rule; Step S47: Generate a synchronization control schedule based on the actuator path set, the state transition point table and the local adjustment rules; Step S48: performing feedback correction processing according to the local adjustment rule, the synchronous control schedule and the force balance configuration vector to obtain feedback correction parameters; Step S49: Integrate the actuator path set, the state transition point table, the synchronous control time table, the local adjustment rules and the feedback correction parameters to generate a collaborative control instruction flow.

10. The embedded distributed device coordinated motion control method according to claim 1, characterized in that: Step S49 includes: Design a basic instruction template based on the actuator path set, state transition point table and synchronization control time table; Perform relative addressing conversion according to the basic instruction template and the executor path set to obtain a relative addressing instruction set; Implement incremental encoding on the relative addressing instruction set to obtain an incremental encoding instruction stream; Performing context-aware instruction compression on the incrementally encoded instruction stream according to a local adjustment rule to obtain compressed instruction data; Hierarchical adaptive caching is performed based on compressed instruction data and feedback correction parameters to obtain a coordinated control instruction flow.

Images