Intelligent control method based on Internet of Things

Abstract

The invention relates to the technical field of soft robots and intelligent control, in particular to an intelligent control method based on the Internet of Things. The method comprises the following steps: carrying out multi-modal data acquisition and preprocessing on a contact process of a gripper and an object to obtain a time synchronization data set; constructing a tactile feature tensor including pressure, friction force and strain force according to the time synchronization data set; performing stress gradient calculation and mapping according to the tactile feature tensor to obtain a stress gradient field and a region threshold mapping table; performing stress gradient threshold adaptive adjustment according to the regional threshold mapping table and the stress gradient field to obtain an adaptive threshold distribution map; and performing stress risk prediction according to the adaptive threshold distribution map to obtain a stress risk map. According to the method, the potential damage risk is predicted by sensing the contact state of the gripper and stress concentration, the adaptive capacity, robustness and success rate of the grabbing process are remarkably improved, and the method is particularly suitable for grabbing tasks of fragile and irregular objects which are difficult to prejudge.

Description

Technical Field

[0001] The present invention relates to the field of soft robots and intelligent control technology, and in particular to an intelligent control method based on the Internet of Things. Background Art

[0002] Existing intelligent control methods for soft robots based on the Internet of Things face serious deficiencies in force feedback accuracy and safety when handling fragile and irregular objects. Traditional soft grasping devices usually use a single pressure sensor or a simple tactile array, which cannot accurately perceive the local stress distribution. This can easily lead to excessive squeezing during the grasping process, causing irreversible damage to fragile objects. At the same time, there is an irreconcilable contradiction between global stability and local safety. Traditional methods either overemphasize grasping stability and ignore local stress safety, or over-focus on safety, resulting in unstable grasping. Lacking an effective balance mechanism, they cannot meet these two key requirements simultaneously. In addition, current soft grasping controls generally lack real-time adaptive adjustment capabilities. Mainstream technologies often rely on preset grasping modes or fixed pressure threshold control strategies. Faced with complex and changing environments and changes in object characteristics, they are unable to make dynamic adjustments, resulting in high grasping failure rates and poor adaptability.

[0003] In summary, existing technologies have problems such as lack of fine perception in grasping fragile / irregular objects, difficulty in balancing stability and safety, and insufficient real-time adaptive capabilities that need to be addressed urgently. Summary of the Invention

[0004] Based on this, it is necessary to provide an intelligent control method based on the Internet of Things to solve at least one of the above technical problems.

[0005] To achieve the above objectives, an intelligent control method based on the Internet of Things includes the following steps:

[0006] Step S1: Perform multimodal data acquisition and preprocessing on the contact process between the soft gripper and the object to obtain a time-synchronized dataset; construct a tactile feature tensor containing pressure, friction, and strain based on the time-synchronized dataset;

[0007] Step S2: Calculate and map the stress gradient based on the tactile feature tensor to obtain a stress gradient field and a regional threshold mapping table; adaptively adjust the stress gradient threshold based on the regional threshold mapping table and the stress gradient field to obtain an adaptive threshold distribution map; predict stress risk based on the adaptive threshold distribution map to obtain a stress risk map;

[0008] Step S3: Analyze the grasping contact points based on the stress risk map to obtain a contact point feature table; calculate the anti-slip force based on the contact point feature table to obtain an anti-slip force demand map; perform local-global constraint fusion based on the stress risk map and the anti-slip force demand map to obtain an initial force distribution scheme; perform local-global force balance iterative optimization on the initial force distribution scheme to obtain a multi-constraint drive instruction set;

[0009] Step S4: progressively drive and execute the multi-constraint drive instruction set to obtain drive execution status data; evaluate the deformation effect data of the gripper in real time; monitor the contact slip risk based on the tactile feature tensor to obtain the contact status monitoring result; generate fine-tuning compensation instructions based on the deformation effect data and the contact status monitoring result; drive the feedback loop in real time based on the drive execution status data and the fine-tuning compensation instructions to realize intelligent control of the gripper.

[0010] The present invention uses high-density, multimodal tactile sensors to synchronously collect pressure, friction, and strain data, obtaining refined and comprehensive information on the internal and external mechanical states of the gripper that far exceeds that of traditional gripping devices. These raw signals are precisely filtered, denoised, and time-synchronized to ensure data accuracy and availability. The resulting tactile feature tensor integrates the scattered sensor data into a unified, high-dimensional representation, providing a rich, pre-processed, and reliable data foundation for subsequent stress analysis, risk assessment, and control decisions. This allows the system to perceive the complex and subtle interactions between objects and the gripper, a key prerequisite for safely grasping fragile and irregular objects.

[0011] By calculating the stress gradient and identifying stress concentration areas, it is possible to accurately locate potential risk points where the gripper or object is subjected to excessive stress. The introduction of historical data association and regional differentiated threshold calculation means that the safety threshold setting is no longer a fixed universal value, but is personalized according to the characteristics of different areas of the gripper and the grasping experience of similar objects, which improves the rationality and pertinence of the setting. Furthermore, the threshold is adaptively adjusted based on the real-time stress change rate, so that the system can dynamically respond to uncertainties in the grasping process, be more vigilant about rapidly changing risk areas, and avoid being overly conservative in stable areas, thereby maximizing grasping efficiency while ensuring safety. The final generated stress risk map intuitively and comprehensively displays the current stress risk level, stress change direction and future trend forecast of each area of ​​the gripper, providing important safety guidance information for subsequent force planning and effectively preventing damage to objects during the grasping process.

[0012] When planning the grasping force, the limitations of traditional methods that only consider global stability are overcome. By accurately calculating the anti-slip force requirement, the overall firmness of the grasping process is ensured. More importantly, the local safety constraints in the stress risk map (to prevent object damage) and the global stability constraints in the anti-slip force requirement map (to prevent slip instability) are creatively integrated into a unified optimization framework. By adopting a local-global force balance iterative optimization method and alternatingly optimizing the force distribution in different regions, the two often conflicting goals of global grasping stability and local stress safety are effectively balanced, and an optimal force distribution scheme is found that can achieve stable grasping without damaging the object. The final multi-constraint drive instruction set is a direct reflection of this optimization result, providing a scientific and safe mechanical foundation for subsequent precise control.

[0013] This system achieves refined and dynamic control of the soft gripper. Dividing the gripper into multiple independent control zones and implementing progressive drive execution avoids local overloads or underloads caused by traditional overall drive. By coordinating inter-zone change constraints and implementing risk-sequential execution, the system ensures a smooth and stable drive process. Real-time monitoring of gripper deformation and contact slip risk monitoring using multimodal tactile data, particularly sensitive detection of micro-slip and pressure anomalies, provides critical state feedback during the grasping process. Based on these real-time monitoring results, the system can promptly detect deviations between the actual and expected state, as well as potential instability or damage risks, and quickly calculate and generate fine-tuning compensation instructions. The resulting real-time drive feedback loop enables the system to dynamically adjust drive instructions based on the real-time grasping state, forming a closed-loop control system that significantly improves the adaptability, robustness, and success rate of the grasping process. This system is particularly suitable for grasping fragile, irregular, and other difficult-to-predict objects.

[0014] Therefore, the present invention provides an intelligent control method based on the Internet of Things, which realizes fine perception of the contact state between the object and the gripper through multimodal tactile perception, and predicts potential damage risks in combination with stress concentration warning analysis; and adopts local-global force balance iterative optimization to take into account both grasping stability and local stress safety in the force planning stage; finally, dynamic adaptive adjustment of the grasping process is achieved through partitioned differentiated drive control and real-time feedback loop, thereby effectively solving the above-mentioned drawbacks faced by existing technologies in grasping fragile and irregular objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a process flow of an intelligent control method based on the Internet of Things;

[0016] Figure 2 Schematic diagram of the detailed implementation steps of step S1 in the present invention.

[0017] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0018] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0019] 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.

[0020] 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.

[0021] In an embodiment of the present invention, reference 2 is shown as a flow chart of the steps of the intelligent control method based on the Internet of Things of the present invention. In this example, the intelligent control method based on the Internet of Things includes the following steps:

[0022] Step S1: Perform multimodal data acquisition and preprocessing on the contact process between the soft gripper and the object to obtain a time-synchronized dataset; construct a tactile feature tensor containing pressure, friction, and strain based on the time-synchronized dataset;

[0023] In an embodiment of the present invention, an 8×8 grid of flexible pressure sensors and accompanying friction sensors are embedded on the inner surface of the soft gripper, with 12 stretchable strain sensors distributed at key locations within the gripper. During the grasping process, these sensors synchronously acquire resistance or voltage signals at a frequency of 200Hz, forming a raw sensor signal dataset. The pressure signal is low-pass filtered with a 20Hz cutoff frequency, the friction signal is median filtered to remove spikes, and the strain signal is Kalman filtered to eliminate drift, resulting in a filtered signal dataset. Based on sensor timestamps and response delay compensation (5ms for pressure, 8ms for friction, and 12ms for strain), the filtered data is resampled to a unified 5ms time series to form a time-synchronized dataset. Using the time-synchronized dataset, the 8×8 pressure data is bilinearly interpolated to construct a 64×64 contact pressure distribution map. The friction data is used to construct a two-dimensional friction vector field and calculate the local friction coefficient. The 12 strain node data are then used to reconstruct the three-dimensional strain field within the soft gripper via radial basis function interpolation. The pressure distribution (two-dimensional), friction vector field (two-dimensional), and strain field (three-dimensional) are organized into a three-dimensional structure according to the spatial position correspondence, and then the time series are stacked to finally construct a four-dimensional tactile feature tensor [space_x×space_y×space_z×time]. Each spatial point contains pressure, friction vector, and strain tensor components, which fully characterize the interaction between the contact interface between the gripper and the object and the deformation state inside the gripper.

[0024] Step S2: Calculate and map the stress gradient based on the tactile feature tensor to obtain a stress gradient field and a regional threshold mapping table; adaptively adjust the stress gradient threshold based on the regional threshold mapping table and the stress gradient field to obtain an adaptive threshold distribution map; predict stress risk based on the adaptive threshold distribution map to obtain a stress risk map;

[0025] In this embodiment of the present invention, pressure, friction, and strain tensors are extracted from the current tactile feature tensor slice, and equivalent stress is calculated to form a tactile feature map. This map identifies areas where stress exceeds 0.3 MPa. Three-dimensional central difference is applied to the tactile feature map to calculate the partial derivatives of stress in the x, y, and z directions, synthesizing the stress gradient vector field. A 3×3×3 adaptive convolution is applied to the stress gradient field to detect gradient abrupt changes. Regions with gradients exceeding 0.5 MPa / mm are marked as stress concentration areas, generating a stress concentration area marker map. Based on the stress concentration area marker map, a historical database is searched for similar objects and grasping records, historical case similarities are calculated, and association rules between stress concentration and damage are extracted to generate a historical association feature set. Combining the historical association feature set with the gripper's regional sensitivity classification (high, medium, and low sensitivity zones) and material properties, personalized stress thresholds are calculated for each gripper region to form a regional threshold map. The stress or stress gradient change rate is monitored in real time (for example, based on data from the last five time points) to identify areas with rapid changes exceeding 0.05 MPa / ms. Based on the regional threshold mapping table, the threshold for the rapidly changing zone is dynamically lowered according to the formula "new threshold = baseline threshold × (1-α × change rate_normalized)", where α is determined based on regional sensitivity. The adjusted threshold is subjected to three-dimensional Gaussian smoothing to ensure spatial continuity, generating an adaptive threshold distribution map. Based on the adaptive threshold distribution map, the ratio of each stress value to the adaptive threshold is calculated as a stress exceedance risk index (0-1). Based on the time series of the tactile feature tensor, a linear prediction algorithm is used to predict the stress change rate and peak time for the next 100 ms to generate a stress evolution prediction map. The stress exceedance risk index distribution, stress gradient field, and stress evolution prediction map are integrated to construct a three-dimensional stress risk map. Each point contains the risk index, gradient direction, and predicted change rate, and is divided into green, yellow, and red zones according to the risk index.

[0026] Step S3: Analyze the grasping contact points based on the stress risk map to obtain a contact point feature table; calculate the anti-slip force based on the contact point feature table to obtain an anti-slip force demand map; perform local-global constraint fusion based on the stress risk map and the anti-slip force demand map to obtain an initial force distribution scheme; perform local-global force balance iterative optimization on the initial force distribution scheme to obtain a multi-constraint drive instruction set;

[0027] In the embodiment of the present invention, based on the stress risk map and tactile feature tensor, the effective contact area is identified, the contact area and force state are calculated, the spatial position of key contact points such as the pressure center is determined, and a contact point feature table is formed. The characteristic parameters such as the material and weight of the object are obtained. Combining the contact point feature table with the object characteristics, a speed-dependent friction model is used. Calculate the speed-corrected friction coefficient. Considering the surface roughness Ra, use μ_effective= The effective friction coefficient is calculated. Based on the effective friction coefficient, local pressure, and tangential force, the contact area is divided into adhesion and potential slip zones, generating contact state partitioning data. Based on this contact state partitioning data, the minimum total normal force required to satisfy overall anti-slip conditions is calculated and optimally distributed to each zone based on contact area, local curvature, and contact state. This forms a normal force distribution scheme, which, together with the contact state partitioning data, constructs an anti-slip force demand map. Based on the anti-slip force demand map and the contact point characteristic table, the overall force and torque balance equations and friction cone constraints are established to construct global stability constraints. High-risk areas with a risk index > 0.8 are extracted from the stress risk map. Based on their adaptive thresholds, the maximum allowable stress σ_max_i is set. This is converted into constraints on the normal force P_i and tangential force F_friction_i in these areas (equivalent stress ≤ σ_max_i), thus constructing local safety constraints. The global stability and local safety constraints are integrated and solved using a quadratic programming method to obtain an initial force distribution scheme that satisfies all constraints. The gripper area is divided into a core gripping block, a safety-sensitive block, and a transition support block based on the stress risk map and contact morphology, forming a regional segmentation table. A composite objective function, J(f)=λG(f)+(1-λ)L(f), is constructed: G(f) minimizes the deviation from the anti-slip demand map, L(f) penalizes high-stress areas, and the weight w_i is determined based on regional sensitivity. The initial force distribution scheme is used as the starting point for iteration, with a maximum number of iterations set to 50 and a convergence threshold of 0.01. During the iteration, the forces of the other blocks are alternately fixed, and the force of the current block is optimized using gradient descent (first the core block, then the safety-sensitive block, and finally the transition support block), while λ is dynamically adjusted to balance G and L. After iterative convergence, the final force distribution scheme is converted into target air pressure / voltage values ​​for each actuation unit. A multi-constraint actuation instruction set is generated by combining the execution sequence, timing, monitoring thresholds, and contingency plans.

[0028] Step S4: progressively drive and execute the multi-constraint drive instruction set to obtain drive execution status data; evaluate the deformation effect data of the gripper in real time; monitor the contact slip risk based on the tactile feature tensor to obtain contact status monitoring results; generate fine-tuning compensation instructions based on the deformation effect data and contact status monitoring results; and drive a feedback loop in real time based on the drive execution status data and the fine-tuning compensation instructions to achieve intelligent control of the gripper.

[0029] In this embodiment of the present invention, the soft gripper is divided into eight independent control areas based on the physical structure and drive units. These areas are numbered and a topological relationship diagram (adjacency matrix) is constructed to form a regional division scheme. The multi-constrained drive instruction set is parsed, and the instructions of each drive unit are distributed and mapped to the corresponding area. A drive execution parameter table containing the target air pressure / voltage values ​​is generated. The number of drive segments in each area is dynamically determined based on the target value change amplitude ΔP_i in the drive execution parameter table and the stress risk map (the number of segments is increased in areas with a risk index > 0.7). , forming a table of regional segment numbers. Nonlinear parameter increments are designed for each region based on the number of segments (small increments in the early and late stages, larger increments in the middle stages), forming a segmented parameter table. Based on the regional topology and the segmented parameter table, the difference in parameter changes between adjacent regions is constrained to not exceed a threshold (for example, first-level adjacent difference ≤ C_1). A "lead-follow" model is introduced to obtain a coordinated segmented parameter table. Regional risks are ranked according to the stress risk map, and an execution order list is generated. Combining the coordinated segmented parameter table and the execution order list, a sequence of target parameter values ​​for each region is generated for each time step (for example, 50ms intervals), forming a phased execution plan. The underlying drive control module sends control signals to the drive units in each region according to the plan, while simultaneously monitoring the actual air pressure / voltage values ​​and the actual deformation (e.g., bending angle) calculated from strain sensor data in real time, recording these as drive execution status data. Using the tactile feature tensor (updated in real time), the current pressure distribution and friction force vector field are extracted to construct contact feature data. Dynamic contact characteristics such as pressure center, contact area, and key point pressure and friction are calculated based on the contact feature data. Based on the dynamic characteristics of the contact, the ratio of the local friction coefficient to the effective friction coefficient is calculated, or changes in the friction force direction are monitored to detect signs of microslip and generate a slip risk assessment map. Areas where pressure values ​​deviate from expectations are identified, generating a pressure anomaly map. The pressure anomaly map and the slip risk assessment map are combined to assess the local stability risk of each contact area and generate a local stability assessment map. Based on the slip risk assessment map and contact characteristic data, global stability indicators such as gripping torque margin and pressure center fluctuation are calculated. The local stability assessment map and global stability indicators are combined to generate contact state monitoring results (overall stability, unstable areas, and potential slip directions). The actual deformation deviation in the drive execution state data is compared with the expected deformation. Combined with the local unstable areas and slip risks in the contact state monitoring results, the required drive parameter fine-tuning is calculated (for example, increasing air pressure in a certain area to improve grip or adjusting torque to correct posture), and fine-tuning compensation instructions are generated. The drive execution status data, fine-tuning compensation instructions, deformation effect data and contact status monitoring results are integrated into a real-time data stream to build a closed-loop feedback mechanism. The execution status is monitored at a frequency of 20ms, deformation and contact are evaluated at 50ms, and fine-tuning instructions are generated and executed at 100ms, realizing dynamic adaptive intelligent control of the gripper's grasping process.

[0030] As an example of the present invention, refer to Figure 2 As shown, in this example, step S1 includes:

[0031] Step S11: collecting signals during the contact process with the object through a sensor array installed on the soft gripper to obtain raw sensor signals, wherein the raw sensor signals include pressure signals, friction signals, and strain signals;

[0032] In this embodiment of the present invention, a high-density flexible pressure sensor array is arranged in an 8×8 grid. Each sensor unit converts normal contact pressure into a resistance change based on the piezoresistive effect. This resistance change is then captured and digitized by an integrated circuit. A micro friction sensor is installed alongside each pressure sensor. Using multi-axis force sensor technology, it measures the tiny deformation of a cantilever beam or elastomer, decomposing the tangential contact force into two orthogonal components (e.g., along the x and y directions of the gripper surface). These components are then converted into voltage signals for digitization. A stretchable strain sensor network consists of 12 nodes distributed along key deformation zones of the soft gripper (e.g., joint bends and fingertip stretch zones). These sensors utilize stretchable conductive materials based on carbon nanotubes or liquid metal. Their resistance changes with deformation caused by tension or compression. This resistance change is detected by a Wheatstone bridge circuit and converted into a voltage signal for digitization. All sensor signals are synchronously acquired by a high-speed data acquisition module with a sampling frequency set to 200Hz, generating a dataset of raw pressure, friction, and strain signals with timestamps.

[0033] Step S12: performing signal filtering and noise elimination on the original sensor signal to obtain a filtered signal data set;

[0034] In an embodiment of the present invention, the collected raw pressure signal contains high-frequency noise caused by mechanical vibration or power supply fluctuations. A digital low-pass filter (for example, implemented using a Butterworth filter algorithm) is used to process the pressure signal with a cutoff frequency of 20 Hz to retain the valid contact force signal and remove high-frequency interference. The raw friction force signal contains instantaneous spikes or glitches. A median filter algorithm is used to process the friction force signal. This algorithm smoothes the signal by taking the median of the data within a sliding window, effectively eliminating abnormal spike interference. The raw strain signal is affected by temperature drift or long-term creep. A Kalman filter algorithm is used to process the strain signal. The Kalman filter optimizes the signal by combining sensor measurements with the system dynamic model, effectively reducing random noise and compensating for signal drift. These filtering algorithms are implemented on an embedded processing unit to perform real-time or quasi-real-time processing of the raw sensor signals.

[0035] Step S13: performing sensor data time synchronization on the filtered signal data set to obtain a time synchronized data set;

[0036] In an embodiment of the present invention, each set of sensor data collected in step S11 is accompanied by a precise timestamp. Although the sensor acquisition is synchronized, different sensor types have inherent response delays. The filtered pressure signal data is time-compensated by 5ms (the data points are shifted forward 5ms on the time axis), the friction signal data is time-compensated by 8ms, and the strain signal data is time-compensated by 12ms to calibrate their response delays. Although the compensated data stream has the same initial sampling frequency, the time points are not completely aligned due to compensation and slight acquisition differences. Therefore, all data are resampled into a unified time series, for example, one data point every 5ms, and the data values ​​at non-original sampling time points are calculated using a linear interpolation method to ensure that all sensor data are completely aligned in the time dimension to form a time-synchronized data set.

[0037] Step S14: constructing a contact pressure distribution map, a contact friction force vector field, and an internal strain field of the soft body according to the time-synchronized data set;

[0038] In this embodiment of the present invention, pressure sensor data from a time-synchronized dataset is used to map discrete pressure values ​​in an 8×8 grid onto a two-dimensional coordinate system representing the contact surface of the soft gripper. To obtain more detailed distribution information, bilinear interpolation is used between sensor points to generate a continuous 64×64 resolution pressure distribution heat map, which visually represents the normal pressure distribution at the contact interface. Friction sensor data from the time-synchronized dataset is used to construct the x- and y-direction tangential force components collected at each sensor location into the friction force vector at that location, forming a sparse friction force vector field. Simultaneously, at each sensor location, the friction coefficient μ_i = |F_friction_i| / P_i is calculated, where |F_friction_i| is the magnitude of the friction force vector at that location and P_i is the corresponding pressure value. Using data from 12 strain sensor nodes in the time-synchronized dataset, a radial basis function (RBF) interpolation method is used to reconstruct the three-dimensional continuous strain field within the soft gripper based on these 12 nodes. RBF interpolation approximates strain values ​​by constructing a linear combination of distance-based kernel functions. Its coefficients are determined by fitting known sensor data using the least squares method, thereby predicting the strain state of any point inside the gripper and further calculating the principal strain direction and strain energy density distribution.

[0039] Step S15: spatially organize the contact pressure distribution map, the contact friction force vector field, and the internal strain field of the soft body to obtain a tactile feature tensor.

[0040] In this embodiment of the present invention, the contact pressure distribution map (two-dimensional data), contact friction force vector field (two-dimensional vector data, which can be viewed as two two-dimensional component maps), and internal strain field (three-dimensional data, comprising multiple components of the strain tensor) generated in step S14 are spatially aligned and integrated. For each grid cell or key spatial point on a predefined three-dimensional grid representing the soft gripper's space, the corresponding pressure value, two-dimensional friction force vector information, and three-dimensional strain tensor component values ​​extracted from the reconstructed strain field are organized together. For example, the data for a spatial point includes the pressure value, friction force x component, friction force y component, and strain tensor xx component, yy component, zz component, xy component, yz component, and zx component. The data from these spatial points are combined to form a three-dimensional data structure representing the complete mechanical state of the soft gripper at a specific moment. By stacking these three-dimensional data structures in a time series, a four-dimensional data structure, namely the tactile feature tensor, is obtained. Its dimensions are [space_x × space_y × space_z × time], which contains the complete mechanical state information of the contact interface between the object and the gripper and the internal structure of the gripper over time.

[0041] Preferably, the calculation and mapping of stress gradients according to the tactile feature tensor in step S2 includes:

[0042] extracting a tactile feature map from the tactile feature tensor;

[0043] Calculating the stress gradient field according to the tactile feature mapping table;

[0044] Identify stress concentration areas in the stress gradient field and obtain a stress concentration area marking map;

[0045] Perform historical data correlation analysis on the stress concentration area marker map to obtain a historical correlation feature set;

[0046] The regional differentiation threshold is calculated based on the historical correlation feature set and the stress concentration area marking map to obtain the regional threshold mapping table.

[0047] In this embodiment of the present invention, a slice of the tactile feature tensor at the current moment is selected. This is a three-dimensional data structure [space_x × space_y × space_z], where each spatial point stores the pressure value, friction vector, and strain tensor. For subsequent analysis, key mechanical parameters are extracted from this three-dimensional data to construct a tactile feature mapping table. For the gripper surface area, the contact pressure value P and friction vector F_friction are mainly extracted, and the total force F_total = For the internal area of ​​the gripper, the components of the strain tensor ε are mainly extracted and the equivalent stress σ_e is calculated, for example using the VonMises equivalent stress criterion:

[0048] σ_e= ;

[0049] Where σ is the stress tensor, and its components σ_ij and the strain tensor ε_ij are calculated using the material's constitutive relations (e.g., Hooke's law for linear elastic materials, σ_ij = C_ijklε_kl, where C is the material's elastic constant tensor). These calculated equivalent stress values ​​or other key parameters (e.g., maximum principal stress) are stored in a three-dimensional mesh corresponding to the spatial structure of the soft gripper, forming a tactile feature map. Furthermore, regions in the tactile feature map where stress amplitudes exceed a preset baseline value (e.g., 0.3 MPa for soft silicone materials) are identified as initial areas of interest.

[0050] The stress gradient represents the rate of change of stress in space and is a key indicator for identifying areas of stress concentration. For each micro-region (grid point) in the tactile feature map (a three-dimensional stress value distribution) obtained in the previous step, the change in stress value relative to the adjacent micro-region is calculated. The three-dimensional central difference method is used to calculate the partial derivatives of stress in the three orthogonal directions of x, y, and z: , similar calculation and Δx, Δy, and Δz are the grid steps. The partial derivatives in these three directions are used as components to construct the stress gradient vector of each micro-region. These vectors form a stress gradient field, which represents the direction and magnitude of the steepest stress change. .

[0051] The stress concentration area is characterized by a stress gradient amplitude significantly higher than that of the surrounding area. A three-dimensional adaptive convolution algorithm is applied to detect sudden changes in the stress gradient field. A 3×3×3 three-dimensional convolution kernel is used to slide on the stress gradient field to calculate the local average gradient or gradient variance. The adaptability is reflected in the fact that the weight or size of the convolution kernel can be dynamically adjusted according to the local gradient characteristics. For example, a smaller kernel is used to focus on details in areas with drastic gradient changes. An initial gradient threshold (for example, 0.5MPa / mm) is set to The areas exceeding the threshold are marked as potential stress concentration areas. These marked areas form a three-dimensional binary map, namely the stress concentration area marking map, where the value of the marked area is 1 and the value of the non-marked area is 0.

[0052] A historical grasping database is established, storing tactile feature tensors, stress risk maps, actuation instruction sets, and final grasping results (success, failure, object damage type, and location) from previous grasping of various objects (especially fragile and irregular objects). For the currently identified stress concentration area marker, the historical database is searched for historical records with similar grasped object types, similar initial contact postures, or similar stress distribution patterns. Similarity scores are calculated between the current stress distribution and historical success and failure cases, for example, using methods based on image features (such as Hu moments and Zernike moments) or signal correlation (such as the Pearson correlation coefficient). The historical cases with the strongest correlation in spatial location and stress characteristics with the current stress concentration area are extracted. The stress evolution process in the corresponding areas, the established safety thresholds, and the final grasping results (whether damage occurred, the type of damage, and the location) in these historical cases are analyzed. The historical case information with strong correlation, similarity scores, and association rules between stress concentration areas extracted from historical cases and grasping results (for example, a stress gradient higher than X MPa / mm in this area caused object damage in Y% of historical failure cases) are integrated into a historical correlation feature set.

[0053] Based on the potential risk areas identified by the stress concentration region marker map and combined with information extracted from the historical correlation feature set, personalized stress thresholds are calculated for different regions of the soft gripper. For regions exhibiting a high risk of damage in historical data (e.g., stress concentration and object damage in multiple grasping failures), the safety stress threshold for these regions is set lower. For relatively safe regions in the historical data, a more relaxed threshold can be set. The threshold calculation also considers the soft gripper's material properties (e.g., yield strength and fracture strain) as well as the geometry of different regions (e.g., weak links and stress concentration design points). For example, for a region in the historical data, the historical safety stress upper limit is P_hist. Combined with the gripper material's safety factor SF (e.g., SF = 0.8), the baseline threshold for this region is set to P_hist × SF. These differentiated thresholds form a three-dimensional data structure corresponding to the spatial structure of the soft gripper, namely, a regional threshold mapping table, in which each microregion has a specific stress threshold.

[0054] Preferably, the adaptive adjustment of the stress gradient threshold according to the regional threshold mapping table and the stress gradient field in step S2 includes:

[0055] Classify the soft gripper's regional sensitivity and obtain a regional sensitivity classification table;

[0056] Obtain and calculate the similarity of the currently grasped object based on historical grasping data to obtain a set of similar cases;

[0057] Assign time weights to similar case sets to obtain weighted case sets;

[0058] Calculate the benchmark threshold value for the weighted case set and obtain the regional benchmark threshold table;

[0059] Monitor the stress change rate based on the stress gradient field and the tactile characteristic tensor to obtain a stress change rate map;

[0060] Identify the sensitive area of ​​stress change rate on the stress change rate map and obtain the rapid change area marking map;

[0061] Dynamically adjust the threshold of the fast-changing area marker map according to the regional threshold mapping table to obtain a preliminary adjustment threshold table;

[0062] Performing spatial smoothing on the preliminary adjustment threshold table to obtain a smoothed threshold table;

[0063] Generates an adaptive threshold distribution map based on a smoothed threshold table.

[0064] In an embodiment of the present invention, the surface and internal space of the soft gripper are divided into regions of different sensitivities based on the material properties, geometric structure, and design functions of the gripper. For example, regions with thinner material thickness, the highest density of embedded sensors, or those expected to withstand greater strain in design are classified as highly sensitive regions; regions with moderate material thickness, average sensor density, or those primarily serving as supports are classified as medium sensitive regions; and regions with thicker material thickness, lower sensor density, or those primarily serving as connectors are classified as low sensitive regions. This classification is determined based on prior knowledge and structural analysis of the gripper. Each spatial microregion is labeled with the sensitivity category to which it belongs (for example, integers 1, 2, and 3 are used to represent low, medium, and high sensitivity, respectively), forming a three-dimensional data structure corresponding to the spatial structure of the gripper, namely, a regional sensitivity classification table.

[0065] Get all stored grasping records from the historical grasping database. For the object currently being grasped, obtain its three-dimensional shape information through an external visual sensor (for example, a depth camera integrated with the gripper). Calculate the similarity between the shape features of the current object and the shape features of the grasped objects in the historical database, for example, using feature descriptors such as shape context and surface normal histogram for matching. At the same time, consider information such as the grasping task type (for example, grasping, placing, carrying), object material properties (for example, preliminary judgment of hardness and surface roughness by sensors), etc. for auxiliary matching. Select historical grasping records with a similarity score higher than a preset threshold (for example, shape similarity higher than 0.7, task type matching) to form a similar case set. The similar case set contains historical data that is somewhat comparable to the current grasping situation.

[0066] The historical records in the similar case set have different occurrence times. In order to make the recent cases that better reflect the current system status and environmental changes have a greater impact on the threshold calculation, a time weight is assigned to each historical case in the similar case set. The time weight calculation formula uses the exponential decay model: W_time= , where t is the difference between the case's occurrence time and the current time (time interval), and T is the time decay constant (for example, setting T to 100 hours means that the weight of cases 100 hours ago decays to approximately 37%). The shorter the time interval, the higher the weight. Each similar case is associated with its calculated time weight to form a weighted case set.

[0067] Based on the historical successful grasping records in the weighted case set, a baseline stress gradient threshold is calculated for each region of the soft gripper. For each region, the historical distribution of stress gradients that appeared in the weighted successful cases is statistically analyzed. The baseline threshold can be set as a statistic of the stress gradient in the region in the weighted historical data, such as the weighted average plus a safety margin, or the 95% quantile of the weighted historical stress gradient distribution. For example, the baseline threshold for region i is ,in is the stress gradient magnitude for region i in the jth weighted historical case, and SafetyOffset_i is the safety offset set for region i, which can be determined based on the regional sensitivity classification table (with larger offsets for highly sensitive areas). These calculated baseline thresholds constitute the regional baseline threshold table, which provides the basis for subsequent real-time adjustments.

[0068] During the grasping process, the current stress gradient field and tactile feature tensor (including time series data) are acquired in real time. The rate of change of each stress value (extracted from the tactile feature tensor) or stress gradient amplitude (extracted from the stress gradient field) over time is calculated. For example, the stress change rate is calculated using the finite difference method: ΔP / Δt ≈ (P(t) - P(t-Δt)) / Δt, where P(t) is the stress value at the current time t, P(t-Δt) is the stress value at the previous time t-Δt, and Δt is the sampling interval. The rate of change of the stress gradient amplitude is calculated similarly. These calculated stress or stress gradient change rate values ​​are stored in a three-dimensional mesh data corresponding to the spatial structure of the gripper, forming a stress change rate graph.

[0069] In the stress rate of change graph, identify areas where the stress or stress gradient rate of change exceeds a preset threshold (for example, a stress rate of change > 0.05 MPa / ms or a stress gradient rate of change > 0.1 MPa / mm / ms). These areas indicate rapidly changing stress states and require special attention. These rapidly changing areas are marked in 3D space to form a 3D binary map, the rapidly changing area marker map.

[0070] For regions marked in the rapidly changing region marker map, the corresponding stress gradient threshold in the regional threshold mapping table is dynamically adjusted based on the region's sensitivity classification and the current stress change rate. The stress change rate-based threshold adjustment formula is: New Threshold = Baseline Threshold × (1-α × Change Rate_Normalization). The baseline threshold is obtained from the regional baseline threshold table, Change Rate_Normalization normalizes the stress or stress gradient change rate in the region to a range of 0-1, and α is the sensitivity coefficient, whose value is determined according to the regional sensitivity classification table (highly sensitive regions have larger α values, such as 0.3; lowly sensitive regions have smaller α values, such as 0.1). The higher the stress change rate, the greater the decrease in the new threshold relative to the baseline threshold, making the system more sensitive to stress concentration in that region. For non-rapidly changing regions, the threshold remains unchanged or undergoes minor adjustments. These adjusted thresholds constitute the preliminary adjusted threshold table.

[0071] In the initial threshold adjustment table, significant differences in threshold values ​​between adjacent areas due to localized fluctuations or sensor noise can cause control instability. To ensure smooth spatial transitions and avoid sudden changes in threshold values, the table is spatially smoothed. Using a three-dimensional Gaussian filter or a moving average filter, the threshold value of each micro-area is weighted averaged with the threshold values ​​of its neighbors, ensuring that the difference in threshold values ​​between adjacent areas does not exceed a preset maximum allowable difference (for example, the difference between adjacent micro-areas does not exceed 30%). This smoothed threshold distribution results in a smoother threshold table.

[0072] The smoothed threshold table contains the stress gradient thresholds for each microregion of the soft gripper at the current grasping moment. These thresholds are dynamically adjusted and spatially smoothed based on historical data, regional sensitivity, and real-time stress change rates. The smoothed threshold table is stored as a standard three-dimensional data structure, an adaptive threshold distribution map. This map provides a real-time, regionally differentiated safety reference standard for subsequent stress risk assessments.

[0073] Preferably, performing stress risk prediction according to the adaptive threshold distribution map in step S2 includes:

[0074] The stress over-limit risk assessment is performed based on the adaptive threshold distribution diagram to obtain the stress risk index distribution;

[0075] The stress evolution trend is predicted based on the stress risk index distribution and the tactile characteristic tensor to obtain a stress evolution prediction map;

[0076] A stress risk map is constructed based on the stress risk index distribution, stress gradient field and stress evolution prediction map.

[0077] In an embodiment of the present invention, for each micro-area in the soft gripper space, its current actual stress value (for example, the equivalent stress value extracted from the tactile feature mapping table) and the corresponding adaptive stress threshold (obtained from the adaptive threshold distribution map) are obtained. The stress overlimit risk index R_i of each micro-area is calculated. The risk index is a value between 0 and 1, which indicates the degree of tension of the current stress value relative to its threshold. The calculation formula is: R_i=σ_i / T_adaptive_i, where σ_i is the actual stress value of micro-area i, and T_adaptive_i is the adaptive stress threshold of micro-area i. In order to limit the index to the range of 0-1 and highlight high risks, a sigmoid function can be used for mapping, for example, R_i= , where k is the gain coefficient, used to adjust the steepness of the curve. When σ_i is much smaller than T_adaptive_i, R_i approaches 0; when σ_i approaches or exceeds T_adaptive_i, R_i rapidly approaches 1. The risk index value of each micro-region is stored in a three-dimensional data structure corresponding to the spatial structure of the gripper, forming a stress risk index distribution.

[0078] The stress change trend of each micro-area in the future period is predicted using the time series data of the stress risk index distribution or the original stress value at the last five time points in the tactile feature tensor. For each micro-area, extract the sequence data of its stress risk index (or original stress value) changing over time. Use a linear prediction algorithm (for example, a linear regression model based on the least squares method) or a more complex time series prediction model (for example, an autoregressive moving average model ARMA or a long short-term memory network LSTM, where linear prediction is used as a simple example) to fit the sequence and predict the stress change rate (for example, MPa / ms) of the micro-area in the next 100ms and the time point at which the peak stress may occur. For example, for micro-area i, the stress value sequence at the most recent moment is σ_i(t-4Δt),...,σ_i(t). Fit this sequence through linear regression to obtain the linear model σ_i( )=a* +b, where represents the future time relative to t. The predicted stress change rate is a. The predicted stress change rate and predicted peak time of each micro-region are stored in a three-dimensional data structure corresponding to the spatial structure of the gripper to form a stress evolution prediction graph.

[0079] A comprehensive stress risk map is constructed by integrating the stress risk index distribution (indicating the current risk level), the stress gradient field (indicating the severity and direction of stress changes), and the stress evolution prediction map (indicating future risk trends) obtained from the stress overlimit risk assessment. The stress risk map is a three-dimensional data structure in which each spatial microregion contains the following information: 1) a stress overlimit risk index (a value between 0 and 1), indicating the risk level of the current stress relative to an adaptive threshold; 2) a stress gradient direction vector (a three-dimensional vector), indicating the direction of the fastest current stress change; and 3) a predicted stress change rate (a signed value), indicating whether the future stress will increase or decrease, and the rate of increase or decrease. To intuitively represent the risk level, the microregions can be divided into different levels based on the stress overlimit risk index. For example, a risk index less than 0.4 is considered a safe zone (marked in green), 0.4 to 0.8 is considered a caution zone (marked in yellow), and greater than 0.8 is considered a dangerous zone (marked in red). The stress risk map provides comprehensive, real-time stress safety information for subsequent mechanical stability constraint optimization and drive control.

[0080] Preferably, calculating the anti-slip force according to the contact point feature table in step S3 includes:

[0081] Obtain the characteristic parameters of the object, calculate the speed-dependent friction coefficient of the contact point based on the contact point characteristic table, and obtain the speed-corrected friction coefficient;

[0082] Surface roughness effect analysis is performed based on the speed-corrected friction coefficient and the contact point characteristic table to obtain the effective friction coefficient;

[0083] Conduct contact area state analysis based on the effective friction coefficient and object characteristic parameters to obtain contact state partition data;

[0084] Calculate the minimum normal force value based on the contact state partition data;

[0085] Optimize the normal force distribution according to the minimum normal force value to obtain the normal force distribution plan;

[0086] The anti-slip force demand map is constructed based on the normal force distribution scheme and contact state partition data.

[0087] In an embodiment of the present invention, before the grasping task begins or through preliminary sensing by sensors, characteristic parameters of the grasped object are obtained, such as the object material (such as glass, plastic, ceramic), surface treatment (such as smooth, rough, coated), estimated weight and center of gravity position. Combined with the contact point feature table (including the spatial position of the key contact points, the current grasping force, local pressure and friction information), the small relative sliding speed v0 of the gripper surface relative to the object surface at each key contact point is calculated. Based on advanced tribology theory, a speed-dependent friction model is used to calculate the friction coefficient μ(v) that takes into account the influence of sliding speed. The model formula is μ(v)= ,in is the static friction coefficient, μ∞ is the dynamic friction coefficient during high-speed sliding, and v0 is the characteristic velocity. ,μ∞, ) is determined by offline experiments or table lookup for a specific gripper material and a specific object material combination. For example, if the gripper material is silicone and the object is glass, the table result is =0.8,μ∞=0.5, =1mm / s. The μ(v) value calculated at each contact point is used as the velocity-corrected friction coefficient at that point.

[0088] Consider the effect of the microscopic topography of the object and gripper surfaces on the actual friction. Obtain the microscopic roughness parameters of the object and gripper surfaces, such as estimating the average roughness Ra through visual or tactile sensors. Based on the velocity-corrected friction coefficient μ(v), ​​the surface roughness effect model is used to calculate the effective friction coefficient μ_effective, which takes into account the influence of microscopic topography. The model formula is μ_effective=μ(v)×[1+k_r× ], where μ(v) is the velocity-corrected friction coefficient calculated in the previous step, Ra is the surface roughness parameter, Ra0 is the reference roughness, and k_r and α are fitting parameters determined experimentally. For example, k_r = 0.1 and α = 0.5. This model reflects that the rougher the surface, the higher the effective friction coefficient (within a certain range). The μ_effective value calculated for each contact point is used as the effective friction coefficient at that point.

[0089] At the contact interface between the gripper and the object, different regions experience different contact states, such as full adhesion (no relative sliding), microslip (localized minor sliding), or macroslip (significant sliding across the entire contact area). The contact state of each region is analyzed based on the effective friction coefficient μ_effective and the current normal pressure P and tangential force F_friction at each contact point. For example, when |F_friction| < μ_effective × P, the region is in adhesion; when |F_friction| approaches or reaches μ_effective × P, the region is in microslip. Combining object parameters (such as surface hardness and elastic modulus) with gripper material properties, a contact mechanics model (e.g., a simplified model based on finite element analysis or Hertzian contact theory) is used to perform a more detailed analysis of the contact interface, dividing the contact region into adhesion and potential slip regions. Each microregion is labeled with its contact state category (e.g., adhesion, microslip), generating contact state partitioning data.

[0090] To prevent macroscopic slip or postural instability during grasping, sufficient normal force must be applied to generate the friction required to resist slip. Based on the contact state partitioning data, the required normal force contribution is calculated for each micro-region in the sticking zone and the potential slip zone. In the sticking zone, the required normal force is primarily used to maintain the current contact state and resist external disturbances (such as gravity and inertia). In the potential slip zone, the minimum normal force required to maintain stable contact or prevent slip is calculated. This is typically related to the effective friction coefficient and the expected tangential force to be resisted in that region. For example, the required total tangential anti-slip force, F_slip_req, depends on the object's gravity, expected acceleration, and other factors. The total normal force, F_normal_total, must satisfy ∑(μ_effective_i×P_i) ≥ F_slip_req, where P_i is the normal pressure applied in micro-region i. The goal is to find a set of P_i that minimizes the total normal force while meeting the anti-slip requirements and taking into account the characteristics of the contact state partitioning (for example, a higher normal force is required in the slip zone to generate sufficient friction). Calculate the minimum total normal force value that meets the overall anti-slip requirement.

[0091] A minimum total normal force is a global requirement that must be rationally distributed across the various contact areas between the gripper and the object. Normal force distribution is an optimization problem. The goal is to find the optimal normal pressure distribution, P_i, while meeting the minimum total normal force requirement. This ensures balanced grasping torque, stable object posture, and minimizes stress concentration. Consideration is given to the geometric fit between the gripper and object, the size of the contact area, and contact state partitioning data. The total normal force is proportionally distributed across the contact areas. The proportionality factor can be positively correlated with the contact area and local curvature fit. For example, a higher normal force is allocated to areas with larger contact areas. Furthermore, based on the contact state partitioning data, a relatively higher normal force density is assigned to areas in the potential slip zone to increase the friction margin in those areas. An optimization algorithm (e.g., a gradient-based optimization method) can adjust the normal force distribution ratio across each area to optimize grasping torque balance while satisfying the total force constraint. The resulting target normal force value for each contact area or micro-region constitutes the normal force distribution solution.

[0092] The normal force distribution scheme obtained in the previous step (i.e., the target normal force value for each contact region or micro-region) is combined with the contact state partitioning data to construct an anti-slip force demand map. The anti-slip force demand map is a two-dimensional or three-dimensional data structure corresponding to the contact interface of the soft gripper, where each contact region or micro-region contains the following information: 1) the target normal force value, indicating the normal force that needs to be applied to the region to meet the overall anti-slip requirement; 2) the local effective friction coefficient, indicating the current friction characteristics of the region; and 3) the contact state marker, indicating whether the region is an adhesion zone or a potential slip zone. This map clearly shows the distribution of mechanical loads required in each contact region of the gripper to ensure a non-slip grip, providing key input for subsequent iterative optimization of the local-global force balance.

[0093] Preferably, in step S3, performing local-global constraint fusion according to the stress risk map and the anti-slip force requirement map includes:

[0094] Construct global stability constraints based on the anti-slip force demand graph and contact point characteristic table;

[0095] Extract high-risk areas from the stress risk map and construct local safety constraints;

[0096] Constraint fusion and initial solution generation are performed on local safety constraints and global stability constraints to obtain the initial force distribution scheme.

[0097] In this embodiment of the present invention, the global stability constraint is designed to ensure that the force applied by the gripper balances the object's gravity, inertia, and external interference forces, while generating sufficient friction to prevent the object from slipping and rotating. The object's force and torque balance equations are established using the target normal force and local effective friction coefficient for each contact region in the anti-slip force demand map, as well as the spatial locations of key contact points and the object's center of gravity in the contact point feature table. The force balance equation is ∑F_contact_i+F_gravity+F_inertial=0, where F_contact_i is the contact force applied by micro-region i (including normal and tangential components), F_gravity is the object's gravity, and F_inertial is the object's inertia. The torque balance equation is ∑(r_i×F_contact_i)+M_gravity+M_inertial=0, where r_i is the position vector of micro-region i relative to the object's center of gravity, and M_gravity and M_inertial are the torques generated by gravity and inertia, respectively. At the same time, the friction cone constraint, |F_friction_i|≤μ_effective_i×P_i, is considered to ensure that the tangential force in each contact micro-region does not exceed its maximum static friction. These equations and inequalities together form the global stability constraint, defining the global force and torque relationship and local friction bounds required to achieve stable grasping.

[0098] Local safety constraints are designed to prevent the gripper from injuring itself or the object during grasping, particularly in areas of stress concentration. Using a stress risk map, high-risk areas with a stress overlimit risk index exceeding a preset threshold (e.g., a risk index > 0.8) are identified. For each high-risk area, the maximum safe stress or load that the area can withstand is determined based on the current stress state, stress gradient, and predicted stress evolution trend displayed in the stress risk map. For example, a maximum allowable stress value σ_max_i is set for the area based on an adaptive threshold (obtained from the adaptive threshold distribution map) and a certain safety margin. This stress limit is converted into constraints on the normal pressure P_i and tangential force F_friction_i applied to the area. For example, the equivalent stress in the area must not exceed σ_max_i. This can be expressed as σ_e(P_i, F_friction_i, shape_i) ≤ σ_max_i, where σ_e is the equivalent stress calculation function and shape_i is the geometric shape parameter of the area. These maximum allowable forces or stress values ​​set for high-risk areas constitute the local safety constraints.

[0099] The global stability constraints constructed in the previous step (consisting of a series of equations and inequalities involving the normal and tangential forces in each region of the gripper) and the local safety constraints (maximum allowable force or stress limits imposed on high-risk areas) are integrated into a unified optimization framework. This optimization problem seeks to find a force distribution scheme at the gripper-object interface (i.e., the normal force P_i and tangential force F_friction_i to be applied in each microregion) that satisfies both global stability requirements (force balance, torque balance, and anti-slip) and local safety requirements (stress in high-risk areas does not exceed safety limits). This is a constrained optimization problem. Quadratic programming can be used to solve this problem and obtain an initial solution. Quadratic programming is an optimization technique applicable to problems with quadratic objective functions and linear constraints. Here, the objective can be set to minimize the total applied force or torque error, while the global stability and local safety constraints are included as linear or nonlinear constraints. The solver calculates an initial set of normal and tangential force distribution values ​​that satisfy all constraints. This set of values ​​constitutes the initial force distribution solution. Although this initial solution is not optimal, it is a feasible solution and provides a starting point for subsequent iterative optimization.

[0100] Preferably, performing local-global force balance iterative optimization on the initial force distribution scheme in step S3 includes:

[0101] The initial force distribution scheme is divided into grasping regions to obtain a regional block table, where the regional block table includes a core grasping block, a safety-sensitive block, and a transition support block;

[0102] According to the regional block table, a dual objective function is constructed to obtain an objective function parameter table;

[0103] Iteratively initialize the initial force distribution scheme according to the objective function parameter table to obtain the iterative initial state;

[0104] Fix the force values ​​of the safety-sensitive block and the transition support block, perform core block optimization on the initial state of the iteration, and obtain the core optimized force distribution;

[0105] The force values ​​of the core grab block and the transition support block are fixed, and the sensitive block optimization is performed on the core optimized force distribution to obtain the sensitive optimized force distribution;

[0106] Fix the force values ​​of the core grab block and the safety sensitive block, optimize the support block for the sensitive optimized force distribution, and obtain the complete optimized force distribution;

[0107] Perform target evaluation and convergence judgment on the complete optimized force distribution according to the objective function parameter table to obtain the convergence state;

[0108] Iterative control and instruction generation are performed according to the convergence state and the complete optimized force distribution to obtain a multi-constraint driving instruction set.

[0109] In an embodiment of the present invention, the contact area between the soft gripper and the object is divided into different functional blocks based on the structural characteristics of the soft gripper, the contact form with the object, and the stress risk map. The core gripping block is the area that mainly undertakes the gripping task and generates the main gripping force. It usually corresponds to the area with the largest contact area between the gripper and the object or the highest curvature matching. The safety-sensitive block is the area marked as a high-risk or attention area in the stress risk map. It is very sensitive to changes in mechanical loads and requires strict stress control. The transition support block is the area other than the core gripping block and the safety-sensitive block, which mainly plays an auxiliary support and stabilization role. This block division is dynamically determined based on the gripper structure, contact geometry, and real-time stress risk assessment. Each contact micro-area is marked with the functional block category to which it belongs (for example, integers 1, 2, and 3 are used to represent the core gripping block, safety-sensitive block, and transition support block, respectively) to form a regional block table.

[0110] Construct a composite optimization objective function J(f) that combines the global stability objective and the local safety objective. The global stability objective function G(f) aims to minimize the deviation between the actual applied force distribution and the ideal force distribution required to satisfy global stability, such as minimizing the total force or total moment error, or minimizing the difference between the normal force distribution scheme and the current scheme: G(f) = , where P_i and F_friction_i are the normal force and friction force of micro-region i in the current solution, P_target_i and F_friction_req_i are the target normal force and required anti-slip friction force given in the anti-slip force demand diagram. The local safety objective function L(f) is designed to penalize high stress areas and force the stress not to exceed the safety threshold: L(f) = , where σ_i is the actual stress in micro-region i (obtained from the tactile feature map), σ_safe_i is the safe stress threshold for micro-region i (obtained from the regional threshold map), and w_i is a weighting factor. w_i is set higher for micro-regions in the safety-sensitive block and lower for other regions. The composite objective function J(f)=λG(f)+(1-λ)L(f) is used, where λ is a balancing factor ranging from [0 to 1], which is used to balance global stability and local safety. The initial value of λ is set to 0.5. The specific form of these objective functions, the weights w_i, and the balancing factor λ constitute the objective function parameter table.

[0111] Use the initial force distribution solution generated by fusion of constraints and the initial solution in step S3 as the starting point for iterative optimization. Set the maximum number of iterations (e.g., 50) and the convergence threshold (e.g., objective function value change of less than 0.01%). Store the initial force distribution solution, the current λ value, the iteration counter (initialized to 0), and the objective function parameter table as the initial state for the iteration.

[0112] In the current iteration, the force distribution values ​​(i.e., normal force P_i and tangential force F_friction_i) for the micro-regions in the safety-sensitive and transition-support blocks in the regional block table are fixed. Only the force distribution for the micro-region in the core gripping block is used as the optimization variable. The optimization objective function is now simplified to J_core(f_core)=λG(f)+(1-λ)L(f), where only the force f_core of the core gripping block is variable. This sub-optimization problem is solved, for example, using gradient descent to adjust the force values ​​of the core gripping block along the negative gradient of the objective function to reduce the objective function value. The local friction cone constraints for all regions must still be satisfied during the optimization process. The optimization continues until the core block force values ​​converge or the predefined number of sub-iterations is reached. The optimized force distribution of the core gripping block, combined with the fixed force distributions of the safety-sensitive and transition-support blocks, constitutes the core optimized force distribution.

[0113] Based on the core block optimization in the previous step, the force distribution values ​​of the core grab block and the transition support block are fixed, and only the force distribution of the micro-region in the safety-sensitive block is used as the optimization variable. At this time, the optimization objective function is J_sensitive(f_sensitive)=λG(f)+(1-λ)L(f), where only the force f_sensitive of the safety-sensitive block is variable. To solve this sub-optimization problem, focus on the local safety target L(f), adjust the force value of the safety-sensitive block, especially reduce the force in the high-stress area, to minimize the local safety objective function. The gradient descent method is also used for optimization, and the local friction cone constraint is satisfied. The optimized force distribution of the safety-sensitive block, together with the fixed force distribution of the core grab block and the transition support block, constitutes the sensitive optimization force distribution.

[0114] Based on the optimization of the sensitive block in the previous step, the force distribution values ​​of the core grasping block and the safety-sensitive block are fixed, and only the force distribution of the micro-region in the transition support block is used as the optimization variable. At this time, the optimization objective function is J_support(f_support)=λG(f)+(1-λ)L(f), where only the force f_support of the transition support block is variable. When solving this sub-optimization problem, the focus is on the global stability goal G(f), and the force value of the transition support block is adjusted to better meet the overall force balance and torque balance requirements. The gradient descent method is used for optimization, and the local friction cone constraint is satisfied. The optimized force distribution of the transition support block, together with the fixed force distribution of the core grasping block and the safety-sensitive block, constitutes the complete optimized force distribution of the current iteration step.

[0115] Calculate the composite objective function value J(f) corresponding to the complete optimized force distribution obtained in the current iteration. Compare the current objective function value with the objective function value from the previous iteration. If the rate of change of the objective function value is less than the convergence threshold (0.01%) or the maximum number of iterations (50) is reached, the optimization process is considered converged. Simultaneously, evaluate the global stability metrics (e.g., total force / torque error) and local safety metrics (e.g., maximum stress exceedance rate) of the current solution. Based on these evaluation results, dynamically adjust the balance factor λ. For example, if the local safety metric is poor (high maximum stress exceedance rate), increase the value of λ in the next iteration to prioritize local safety. If the global stability metric is poor (large force / torque error), decrease the value of λ. Record the current λ value, objective function value, and evaluation metrics to form convergence status information.

[0116] If the optimization process has not converged and the maximum number of iterations has not been reached, the current fully optimized force distribution is used as the initial force distribution for the next iteration, the iteration counter is updated, and the cycle from core block optimization to support block optimization continues. If the optimization process converges or reaches the maximum number of iterations, the fully optimized force distribution solution obtained in the last iteration is used as the optimal force distribution solution. Based on this optimal force distribution solution and combined with the soft gripper's drive model (e.g., the air pressure-deformation-force model), the control parameters (such as air pressure and voltage) of each drive unit (such as the airbag and motor) required to achieve this force distribution are calculated. These drive parameters, execution sequence, timing requirements, as well as key area monitoring thresholds and emergency adjustment plans extracted from the stress risk map and anti-slip force demand map, are formatted into an executable instruction set for the control system, forming a multi-constraint drive instruction set. This instruction set contains all the control information required to drive the gripper for safe and stable grasping.

[0117] Preferably, the step S4 of progressively driving and executing the multi-constraint driving instruction set includes:

[0118] The soft gripper is divided into 8 independent control areas, and each control area is numbered and the relationship is established to obtain the area division scheme;

[0119] Distribute and map the multi-constraint drive instruction set according to the area division scheme to obtain a drive execution parameter table;

[0120] The number of segments of each control area is dynamically determined according to the drive execution parameter table and the stress risk map to obtain a regional segment number table;

[0121] Perform nonlinear segmentation design on the regional segmentation table to obtain the segmentation parameter table; construct a regional topology relationship diagram for the 8 independent control areas;

[0122] According to the regional topology relationship diagram and the segmentation parameter table, the inter-region change constraint coordination is carried out to obtain the coordinated segmentation parameter table;

[0123] Arrange regional risks in order according to the stress risk map to obtain an execution order list;

[0124] Generate a phased execution plan based on the coordinated segmentation parameter table and execution sequence list;

[0125] Drive execution and status monitoring are performed according to the phased execution plan to obtain drive execution status data.

[0126] In an embodiment of the present invention, the gripper is divided into 8 independent control areas according to the physical structure of the soft gripper and the layout of the drive unit. For example, in a three-finger soft gripper, each finger can be divided into three areas: the fingertip, the middle of the finger, and the base of the finger, for a total of 9 areas, or it can be divided into 8 or more independently controllable areas according to the position of the driving airbag / motor. A unique number (for example, 1 to 8) is assigned to each area. At the same time, a topological relationship between areas is constructed to identify which areas are adjacent and which areas have structural influences on each other (for example, the expansion of one area will stretch the adjacent area). This relationship can be represented by an adjacency matrix or a graph structure to form a region division scheme.

[0127] The multi-constraint drive instruction set contains the control parameters (such as air pressure and voltage) for each drive unit required to achieve optimal force distribution. Based on the region partitioning scheme, the global or drive unit-specific instructions in the multi-constraint drive instruction set are parsed and distributed to the corresponding control regions. For example, if the instruction set specifies the air pressure target value P_A for drive unit A, and drive unit A primarily affects regions 1 and 2, the air pressure target value P_A is mapped to the drive execution parameters for regions 1 and 2. At the same time, information such as the execution order, timing requirements, and safety margins contained in the instruction set is also associated with the corresponding regions. This process converts high-level drive instructions into specific execution parameters for each independent control region (for example, the target air pressure for region 1 is P_1, the target air pressure for region 2 is P_2, and so on), forming a drive execution parameter table.

[0128] Progressive drive execution requires breaking down the target drive parameter change process of each area into multiple small steps. The number of segments affects the smoothness and safety of execution. For the difference ΔP_i between the target parameter value and the current parameter value of each area in the drive execution parameter table, the initial number of segments is dynamically determined based on the change amplitude. , where P_threshold is a preset parameter change threshold, such as 0.1 atmospheres or 0.5 volts. Furthermore, based on the stress risk map, for regions with a high stress overlimit risk index (e.g., risk index > 0.7), the number of segments is increased, for example, N_i = N_i + 2, to make the driving changes in these sensitive areas slower and more refined. The calculated number of segments for each region is recorded to form a regional segment number table.

[0129] Based on the regional segmentation table, design a specific parameter change path for each region. Adopt a nonlinear segmentation strategy, for example, using smaller parameter increments at the beginning and end of the actuation process and larger increments in the middle. This helps to slowly establish contact in the early stages, fine-tune the position in the later stages, and quickly reach the target value in the middle stages. For example, for N-segment changes, the parameter increments for each segment can be distributed according to a sinusoidal or exponential function. At the same time, to coordinate the actuation between regions, clarify the topological relationship between regions. Construct a regional topological relationship diagram to show which regions are directly adjacent (first-level adjacency), which regions are indirectly affected by other regions (second-level adjacency), and the strength of the structural connections between them.

[0130] To prevent inconsistent drive parameter changes between adjacent regions, which could lead to gripper instability or shear stress on the object, constraints are imposed on drive changes between regions. Based on the regional topology graph and the segmented parameter table, drive parameter changes between adjacent regions are coordinated. For example, the difference in parameter changes between first-level adjacent regions within the same time step is constrained to not exceed a preset threshold: |ΔP_i(k)-ΔP_j(k)|≤C_1, where ΔP_i(k) is the parameter increment of region i at the kth time step, and C_1 is a constant. For second-level adjacent regions, the constraint can be relaxed appropriately: |ΔP_i(k)-ΔP_j(k)|≤C_2, where C_2>C_1. Furthermore, a "lead-follow" execution mode is introduced. For example, for regions with lower stress risk, their drive can slightly lead that of higher-risk regions to provide buffer time for adjustment. Based on these constraints and modes, the segmented parameters of each region are fine-tuned to obtain a coordinated segmented parameter table.

[0131] The eight control regions are ranked according to risk using information from the stress risk map, including the stress overrun risk index, stress gradient magnitude, and predicted stress change rate for each region. For example, regions with high current risk indexes, large stress gradients, and predicted rapid stress growth are ranked higher in the risk list. The execution order list ranks regions from low to high risk or according to specific coordination strategies (for example, starting with the gripping force core region and gradually expanding to the support region). This list guides the relative execution priority and timing of drive instructions for each region during phased execution.

[0132] The coordinated segmented parameter table (containing the segmented parameter increments and timing information for each region) is combined with the execution order list to generate a detailed phased execution plan. This plan specifies the drive parameter values ​​that each control region should achieve at each time step (for example, every 50ms) throughout the entire drive process. The plan takes into account the number of segments in each region, nonlinear segmentation design, coordination of variation constraints between regions, and the execution order of region risks. For example, the plan specifies: at time t_k, region i executes the drive of segment m_i with the target parameter value P_i(k); region j executes the drive of segment m_j with the target parameter value P_j(k); and the timing and variation constraints between regions are met. This plan is a detailed, time-discrete sequence of instructions used to guide the operation of the underlying drive unit.

[0133] The phased execution plan is sent to the underlying drive control module. According to the plan, the drive control module sends control signals to the drive units (such as air valves and voltage controllers) corresponding to each area. For example, the air valve is controlled to open / close to adjust the airbag pressure, or the voltage output is controlled to drive the electric actuator. While the drive is being executed, the actual working status of each drive unit is monitored in real time, such as measuring the actual air pressure value of the airbag, the actual voltage value or current value of the motor. At the same time, the actual deformation state of each area of ​​the soft gripper is monitored through the sensor data collected in step S1 (for example, the local bending angle or stretching rate is calculated through the strain sensor data). These real-time drive parameter values, actual deformation data, execution time, response delay, and any abnormal conditions (such as air pressure fluctuations, actuator failures) are recorded to form drive execution status data.

[0134] Preferably, performing contact slip risk monitoring according to the tactile feature tensor in step S4 includes:

[0135] Extract the current pressure distribution data and friction force vector field data from the tactile feature tensor to construct contact feature data;

[0136] Calculate the dynamic contact characteristics between the gripper and the object based on the contact feature data, where the dynamic contact characteristics include the pressure center position, contact area, key point pressure value and friction direction;

[0137] Perform micro-slip detection and analysis based on contact dynamic characteristics to obtain a slip risk assessment diagram;

[0138] Perform pressure anomaly pattern recognition based on contact dynamic characteristics and contact feature data to obtain a pressure anomaly area map;

[0139] Generate a local stability assessment map based on the pressure anomaly area map and the slip risk assessment map;

[0140] Calculate the global stability index based on the slip risk assessment map and contact characteristic data;

[0141] Contact status monitoring results are generated based on local stability evaluation diagrams and global stability indicators.

[0142] In an embodiment of the present invention, the latest tactile feature tensor is obtained in real time. From the current time slice of the four-dimensional tensor, a two-dimensional pressure distribution map and a two-dimensional friction vector field representing the contact interface between the soft gripper and the object are extracted. The pressure distribution map is a two-dimensional matrix, in which each element represents the normal pressure value of the corresponding spatial position. The friction vector field is a two-dimensional vector field, and each position contains a two-dimensional vector representing the magnitude and direction of the tangential force. These two two-dimensional data structures, as well as the relevant spatial position information and sensor raw data, are organized together to form the contact feature data at the current moment.

[0143] Analyze the contact feature data to extract macroscopic and local features describing the contact interface. Calculate the pressure center position of the contact interface, which is the weighted average position of all contact pressures: C = (∑_iP_i × r_i) / ∑_iP_i, where P_i is the pressure value of micro-region i and r_i is the position vector of micro-region i. Calculate the effective contact area, which is the total area of ​​micro-regions where the pressure exceeds a preset threshold (e.g., 0.01 MPa). Identify critical contact points, such as peak pressure points, peak friction points, or high-risk points in the stress risk map, and record the specific pressure values ​​and friction vectors at these points. Analyze the overall distribution of the friction vector field, for example, by calculating the total friction vector and average friction direction. These calculated macroscopic and local features constitute the contact dynamic characteristics.

[0144] Microslip is a localized, minute relative motion that precedes macroslip and is a key precursor to unstable grasping. Contact dynamics and contact feature data can be used to detect signs of microslip at the contact interface. One approach is to analyze whether the local friction coefficient reaches or approaches a critical value: for each contact microregion, the local friction coefficient μ_local_i = |F_friction_i| / P_i is calculated. If μ_local_i approaches or exceeds the effective friction coefficient μ_effective_i for that region (derived from the anti-slip force demand map), this indicates that the region is on the verge of microslip. Another approach is to monitor changes in the direction of the local friction force vector: During stable grasping, the direction of the local friction force is typically correlated with the object's gravity or the intended direction of motion. Abnormal deflections or rapid fluctuations in the friction force direction in a particular region indicate the occurrence of microslip. Rapid, unexpected changes in the local pressure distribution can also be analyzed, which can cause a redistribution of contact points associated with microslip. Based on these analyses, a micro-slip risk index is calculated for each micro-region of the contact interface (e.g., based on the value of μ_local_i / μ_effective_i or the severity of the change in friction force direction), forming a two-dimensional or three-dimensional slip risk assessment map.

[0145] Monitor the contact interface for abnormal pressure distribution patterns, such as localized excessive pressure (causing damage to the object) or localized excessive pressure (causing loss of contact). Use the pressure distribution map in the contact feature data to compare and analyze expected or historically successful pressure distribution patterns, such as using image comparison algorithms or statistical methods. Identify areas where pressure values ​​deviate significantly from expected values ​​(for example, exceeding a certain multiple of the average pressure or exceeding the historical safety range). Pay particular attention to the location and amplitude of the pressure peak, as well as the uniformity of the pressure distribution. Combined with the stress risk map, identify whether the pressure anomaly occurs in a safety-sensitive block or high-risk area. Mark the identified pressure anomaly areas to form a pressure anomaly area map.

[0146] The pressure anomaly area map (reflecting the abnormality of local load) and the slip risk assessment map (reflecting the uncertainty of local contact state) are combined to perform a local stability assessment on the contact interface between the gripper and the object. Abnormal areas in the pressure anomaly area map and high slip risk areas in the slip risk assessment map are marked as local unstable areas. Different local stability risk levels are assigned according to their specific anomaly types (too high / too low pressure, degree of micro-slip). For example, an area with both high pressure and high micro-slip risk is assessed as the highest local instability level. These local stability risk levels are mapped to the spatial areas corresponding to the gripper contact interface to form a local stability assessment map.

[0147] Assess the overall stability of the entire grasping system. One global stability metric is the grasping torque margin, which is the margin between the stabilizing torque generated by the force applied by the gripper and the interfering torques (such as gravity torque and external disturbance torque) that cause the object to become unstable. The pressure distribution and friction force vector field in the contact feature data are used to calculate the total force and total torque applied by the gripper on the object. This is then balanced with the object's gravity, inertia, and other forces. The distance of the grasping torque relative to the friction cone boundary is calculated; this distance can be used as a measure of anti-slip stability. For example, the grasping force and torque are calculated to determine whether they are within a safety margin within the friction cone. Another global stability metric is the rate of change of the contact area or the degree of fluctuation in the pressure center position. If the contact area decreases rapidly or the pressure center position drifts dramatically, it indicates that the overall grasp is unstable. These calculated numerical metrics constitute the global stability metric.

[0148] The local stability assessment map (which provides detailed risk information for each area of ​​the contact interface) and the global stability index (which provides an overview of the overall gripping status) are combined to generate the final contact state monitoring result. The contact state monitoring result is a comprehensive report containing the following information: 1) The overall gripping stability status (e.g., stable, caution, unstable), determined based on the global stability index; 2) A list of local unstable areas, indicating which areas of the gripper are at risk of pressure anomalies or micro-slip, and their risk levels (obtained from the local stability assessment map); 3) Potential slip directions or pressure overload directions (obtained from the slip risk assessment map and the pressure anomaly area map); and 4) A prediction of future contact state evolution trends (e.g., how long it will take for macro-slip to occur if the current state persists). This result provides real-time, multi-dimensional contact state feedback for subsequent fine-tuning compensation calculations and emergency response.

[0149] 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.

[0150] 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 intelligent control method based on the Internet of Things, characterized in that: The following steps are involved: Step S1: Perform multimodal data acquisition and preprocessing on the contact process between the soft gripper and the object to obtain a time-synchronized dataset; construct a tactile feature tensor containing pressure, friction, and strain based on the time-synchronized dataset; Step S2: Calculating and mapping the stress gradient based on the tactile feature tensor to obtain a stress gradient field and a regional threshold mapping table; adaptively adjusting the stress gradient threshold based on the regional threshold mapping table and the stress gradient field to obtain an adaptive threshold distribution map; Stress risk prediction is performed based on the adaptive threshold distribution map to obtain a stress risk map; Step S3: performing grasping contact point analysis based on the stress risk map to obtain a contact point feature table; Anti-slip force is calculated based on the contact point characteristic table to obtain an anti-slip force demand map. Local-global constraint fusion is performed based on the stress risk map and the anti-slip force demand map to obtain an initial force distribution scheme. Perform local-global force balance iterative optimization on the initial force distribution scheme to obtain a multi-constraint driving instruction set; Step S4: progressively drive and execute the multi-constraint drive instruction set to obtain drive execution status data; evaluate the deformation effect data of the gripper in real time; monitor the contact slip risk based on the tactile feature tensor to obtain the contact status monitoring result; Fine-tuning compensation instructions are generated based on deformation effect data and contact status monitoring results; the feedback loop is driven in real time based on the drive execution status data and fine-tuning compensation instructions to achieve intelligent control of the gripper.

2. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting signals during the contact process with the object through a sensor array installed on the soft gripper to obtain raw sensor signals, wherein the raw sensor signals include pressure signals, friction signals, and strain signals; Step S12: performing signal filtering and noise elimination on the original sensor signal to obtain a filtered signal data set; Step S13: performing sensor data time synchronization on the filtered signal data set to obtain a time synchronized data set; Step S14: constructing a contact pressure distribution map, a contact friction force vector field, and an internal strain field of the soft body according to the time-synchronized data set; Step S15: spatially organize the contact pressure distribution map, the contact friction force vector field, and the internal strain field of the soft body to obtain a tactile feature tensor.

3. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: The calculation and mapping of stress gradients based on the tactile feature tensor in step S2 includes: extracting a tactile feature map from the tactile feature tensor; Calculating the stress gradient field according to the tactile feature mapping table; Identify stress concentration areas in the stress gradient field and obtain a stress concentration area marking map; Perform historical data correlation analysis on the stress concentration area marker map to obtain a historical correlation feature set; The regional differentiation threshold is calculated based on the historical correlation feature set and the stress concentration area marking map to obtain the regional threshold mapping table.

4. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: In step S2, adaptively adjusting the stress gradient threshold according to the regional threshold mapping table and the stress gradient field includes: Classify the soft gripper's regional sensitivity and obtain a regional sensitivity classification table; Obtain and calculate the similarity of the currently grasped object based on historical grasping data to obtain a set of similar cases; Assign time weights to similar case sets to obtain weighted case sets; Calculate the benchmark threshold value for the weighted case set and obtain the regional benchmark threshold table; Monitor the stress change rate based on the stress gradient field and the tactile characteristic tensor to obtain a stress change rate map; Identify the sensitive area of ​​stress change rate on the stress change rate map and obtain the rapid change area marking map; Dynamically adjust the threshold of the fast-changing area marker map according to the regional threshold mapping table to obtain a preliminary adjustment threshold table; Performing spatial smoothing on the preliminary adjustment threshold table to obtain a smoothed threshold table; Generates an adaptive threshold distribution map based on a smoothed threshold table.

5. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: The stress risk prediction according to the adaptive threshold distribution map in step S2 includes: The stress over-limit risk assessment is performed based on the adaptive threshold distribution diagram to obtain the stress risk index distribution; The stress evolution trend is predicted based on the stress risk index distribution and the tactile characteristic tensor to obtain a stress evolution prediction map; A stress risk map is constructed based on the stress risk index distribution, stress gradient field and stress evolution prediction map.

6. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: Calculating the anti-slip force according to the contact point feature table in step S3 includes: Obtain the characteristic parameters of the object, calculate the speed-dependent friction coefficient of the contact point based on the contact point characteristic table, and obtain the speed-corrected friction coefficient; Surface roughness effect analysis is performed based on the speed-corrected friction coefficient and the contact point characteristic table to obtain the effective friction coefficient; Conduct contact area state analysis based on the effective friction coefficient and object characteristic parameters to obtain contact state partition data; Calculate the minimum normal force value based on the contact state partition data; Optimize the normal force distribution according to the minimum normal force value to obtain the normal force distribution plan; The anti-slip force demand map is constructed based on the normal force distribution scheme and contact state partition data.

7. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: In step S3, local-global constraint fusion is performed based on the stress risk map and the anti-slip force demand map, including: Construct global stability constraints based on the anti-slip force demand graph and contact point characteristic table; Extract high-risk areas from the stress risk map and construct local safety constraints; Constraint fusion and initial solution generation are performed on local safety constraints and global stability constraints to obtain the initial force distribution scheme.

8. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: The local-global force balance iterative optimization of the initial force distribution scheme in step S3 includes: The initial force distribution scheme is divided into grasping regions to obtain a regional block table, where the regional block table includes a core grasping block, a safety-sensitive block, and a transition support block; According to the regional block table, a dual objective function is constructed to obtain an objective function parameter table; Iteratively initialize the initial force distribution scheme according to the objective function parameter table to obtain the iterative initial state; Fix the force values ​​of the safety-sensitive block and the transition support block, perform core block optimization on the initial state of the iteration, and obtain the core optimized force distribution; The force values ​​of the core grab block and the transition support block are fixed, and the sensitive block optimization is performed on the core optimized force distribution to obtain the sensitive optimized force distribution; Fix the force values ​​of the core grab block and the safety sensitive block, optimize the support block for the sensitive optimized force distribution, and obtain the complete optimized force distribution; Perform target evaluation and convergence judgment on the complete optimized force distribution according to the objective function parameter table to obtain the convergence state; Iterative control and instruction generation are performed according to the convergence state and the complete optimized force distribution to obtain a multi-constraint driving instruction set.

9. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: The step S4 of progressively driving and executing the multi-constraint driving instruction set includes: The soft gripper is divided into 8 independent control areas, and each control area is numbered and the relationship is established to obtain the area division scheme; Distribute and map the multi-constraint drive instruction set according to the area division scheme to obtain a drive execution parameter table; The number of segments of each control area is dynamically determined according to the drive execution parameter table and the stress risk map to obtain a regional segment number table; Perform nonlinear segmentation design on the regional segmentation table to obtain the segmentation parameter table; construct a regional topology relationship diagram for the 8 independent control areas; According to the regional topology relationship diagram and the segmentation parameter table, the inter-region change constraint coordination is carried out to obtain the coordinated segmentation parameter table; Arrange regional risks in order according to the stress risk map to obtain an execution order list; Generate a phased execution plan based on the coordinated segmentation parameter table and execution sequence list; Drive execution and status monitoring are performed according to the phased execution plan to obtain drive execution status data.

10. The intelligent control method based on the Internet of Things according to claim 1, characterized in that: The contact slip risk monitoring according to the tactile feature tensor in step S4 includes: Extract the current pressure distribution data and friction force vector field data from the tactile feature tensor to construct contact feature data; Calculate the dynamic contact characteristics between the gripper and the object based on the contact feature data, where the dynamic contact characteristics include the pressure center position, contact area, key point pressure value and friction direction; Perform micro-slip detection and analysis based on contact dynamic characteristics to obtain a slip risk assessment diagram; Perform pressure anomaly pattern recognition based on contact dynamic characteristics and contact feature data to obtain a pressure anomaly area map; Generate a local stability assessment map based on the pressure anomaly area map and the slip risk assessment map; Calculate the global stability index based on the slip risk assessment map and contact characteristic data; Contact status monitoring results are generated based on local stability evaluation diagrams and global stability indicators.

Images